Uncategorized | coachemilyterrell.com

AI Texting Rules for Real Estate Agents: TCPA in 2026

Written by Coach Emily on August 28, 2026. Posted in Uncategorized.

By Emily Terrell — Top Coach and Speaker at Tom Ferry International. 4 years coaching at Tom Ferry, 9 years prior as a client.

AI texting rules for real estate agents come from the TCPA, not from any AI-specific law: outbound automated texts and AI voice calls require prior express consent, and marketing versions require prior express written consent. The FCC confirmed in 2024 that AI-generated voices count as artificial. This guide covers the consent tiers, the Texas rules, and the workflow.

Key Takeaways

The question isn’t whether AI is ethical. It’s who the sender is — and whether that person consented to hear from a machine.
AI drafting a message you read and send yourself triggers nothing. AI sending on your behalf triggers the TCPA.
The FCC confirmed in February 2024 that AI-generated voices fall inside the TCPA’s “artificial or prerecorded voice” restrictions, which means prior express consent before you dial.
The TCPA carries a private right of action at $500 per violation, and a court can treble it. Your CRM’s AI feature does not carry that liability. You do.
Texas has no AI disclosure duty for licensees, but TREC Rule 535.155 still governs every message designed to attract the public to your brokerage services.

What is TCPA compliance for AI in real estate?

TCPA compliance for AI in real estate is the practice of confirming you have the right level of consent before an automated system sends a text or places a call on your behalf. It’s not a new rule written for AI. It’s a 1991 statute that AI walked into.

That framing matters because agents keep looking for the AI rule and missing the one that already applies. The Telephone Consumer Protection Act doesn’t care what wrote the message. It cares whether a machine sent it and whether the person on the other end agreed to that.

Why this matters for real estate agents

Most agents adopted the tools and skipped the consent question entirely. According to NAR’s 2025 Technology Survey (September 2025), AI adoption reached 68% of agents while only 17% reported a significant positive impact and 46% noticed no difference at all. Two-thirds of the industry is running AI. A fraction is getting anything from it. Almost none of them have audited what their tools are sending.

The economics make the exposure sharper. According to NAR’s 2026 Member Profile (June 2026), the typical agent closed nine transaction sides in 2025 with a median gross income of $59,200, while median business expenses rose to $9,530 from $8,010 the year before. Now put a TCPA claim on top of that. Under 47 U.S.C. § 227(b)(3), a private plaintiff can recover $500 per violation or actual loss, whichever is greater, and a court may increase the award up to three times that if it finds the violation willful or knowing.

Here’s the thing nobody wants to tell you: per violation means per message. A single automated campaign to a purchased list isn’t one problem. It’s as many problems as you have contacts.

“Your AI vendor is not on the hook for your consent record. You are. If you can’t produce the moment a lead agrees to automated contact, you don’t have consent — you have a contact.” — Emily Terrell, Tom Ferry Coach

The three sender tiers — and which ones need consent

Sort every AI communication task into one of these before you turn anything on.

Does AI drafting a message trigger the TCPA?

No. AI writes your follow-up email, you read it, you hit send. The sender is you. There’s no automated system placing the call or the text, so the TCPA’s artificial-voice and automated-messaging rules aren’t in play.

This is the same tier covered in what real estate agents have to disclose when using AI — no disclosure required, because nothing about the tool changes what the client believes. Your only obligation is accuracy, and it’s the same obligation you’d have if you’d typed it yourself.

Do AI-sent texts to leads need consent?

Yes, and this is where most agents are exposed without knowing it. The moment your CRM’s AI responds to a new lead automatically, at 11:47 PM, without you touching it, you’ve moved from drafting to sending. Consent rules apply to automated messaging, and marketing content raises the bar to prior express written consent rather than general consent.

A pre-checked box buried in a landing page footer is not the same as a documented opt-in you can produce two years later in a demand letter. Neither is an existing relationship with a past client — a relationship can exempt you from do-not-call obligations for manual outreach without giving you consent for automated contact.

Do AI voice calls need consent?

Yes, unambiguously. On February 8, 2024, the FCC issued a Declaratory Ruling confirming that the TCPA’s restrictions on the use of “artificial or prerecorded voice” encompass current AI technologies that generate human voices — and that calls using those technologies require the prior express consent of the called party (FCC).

The ruling didn’t create a carve-out for voice agents that sound convincingly human. The opposite: how lifelike the voice is has nothing to do with it. If a machine generated the voice, it’s artificial.

Inbound is a different picture. When a consumer calls you and an AI answers, the prior-consent analysis that governs outbound calls isn’t the same one in play. If you’re going to run AI on the phone at all, start there.

This is general information, not legal advice. TCPA rules are complex, consent standards have shifted in recent years, and state telemarketing laws add requirements on top of the federal floor. Confirm your setup with your broker and an attorney licensed in your state before you deploy anything outbound.

What Texas agents need to know

Texas licensees do not have an AI disclosure duty. TRAIGA took effect January 1, 2026, and the enacted version put the consumer AI-notice requirement on governmental agencies, with a separate provision for health care providers — not on private businesses generally (Texas Attorney General). This is covered in more depth in AI real estate compliance: what actually risks your license.

What does apply is the rule that’s always applied. TREC Rule 535.155 requires every advertisement to include the license holder’s or team’s name plus the broker’s name in at least half the size of the largest contact information for any sales agent, associated broker, or team name in the ad (TREC). TREC’s definition of an advertisement reaches text messages and social media, and Chapter 1101.652(b)(23) states that a license holder’s advertising can’t be misleading.

Now think about what your AI lead-response tool sends. It has no idea your broker’s name belongs in that message.

Common mistakes

Treating an existing client relationship as consent for automated contact. It isn’t. Those are two different permissions, and the artificial-voice rule doesn’t care how well you know the person.
Assuming the vendor’s compliance page covers you. The consent obligation sits with the calling or texting business. In a dispute, you’re the one producing the record.
Turning on the CRM’s AI auto-responder without auditing the list it runs against. Speed-to-lead is worth building. Speed-to-lead against contacts who never opted in is a per-message liability.
Publishing AI-drafted texts with no broker identification. The tool strips it every time because it doesn’t know the rule exists.
Deploying an AI voice agent because a competitor did. Outbound voice is the highest-exposure tier on this list. Start with inbound.
Keeping no consent record at all. If you can’t show when and how someone opted in, the question isn’t whether you’ll win. It’s what it costs to find out.

Frequently Asked Questions

Can real estate agents use AI to text leads?

Yes, with consent. If AI drafts a message and you send it manually, no special rules apply. If an automated system sends on your behalf, TCPA consent requirements attach, and marketing content requires prior express written consent. Document where each contact’s consent came from and what it covered, because the burden of proving it sits with you rather than your vendor.

Does the TCPA apply to AI-generated text messages?

The TCPA governs automated telephone messaging regardless of what generated the content. What matters is whether an automated system sent it and whether the recipient consented. AI writing the words doesn’t change the analysis. AI sending them does. Statutory damages under 47 U.S.C. § 227(b)(3) run $500 per violation, with courts able to increase the award up to three times for willful or knowing violations.

Do I need consent to use an AI voice agent to call leads?

Yes. The FCC’s February 2024 Declaratory Ruling confirmed that AI technologies generating human voices fall within the TCPA’s restrictions on artificial or prerecorded voice, requiring prior express consent of the called party. How natural the voice sounds is irrelevant to the analysis. Marketing calls carry a higher standard than informational ones. Get legal review before deploying outbound voice.

Is it legal to use AI to answer inbound calls?

Inbound calls a consumer initiates don’t raise the same prior-consent question that outbound calls do, which is why inbound is the safer place to start with voice AI. State call-recording and identification rules can still apply. If you’re testing AI on the phone at all, put it on calls people make to you before you put it on calls you make to them.

Do I have to tell someone they’re talking to an AI?

In Texas, there’s no such duty on real estate licensees — TRAIGA’s consumer notice requirement landed on governmental agencies and health care providers, not private businesses. Separately, artificial or prerecorded voice messages carry federal identification and opt-out requirements. Rules vary by state and are moving toward more disclosure, so build identification in rather than retrofitting it.

Am I liable if my CRM vendor’s AI sends a noncompliant message?

Generally yes, as the calling or texting business. Vendor compliance features help; they don’t transfer the obligation. Ask any vendor directly how consent is captured, where the record lives, how you export it, and what happens when someone opts out mid-sequence. If they can’t answer in specifics, that’s your answer.

Does TREC require my broker’s name in an AI-generated text?

TREC Rule 535.155 requires advertisements to carry the license holder’s or team’s name plus the broker’s name at half the size of the largest agent contact information, and TREC’s definition of advertisement reaches text messages and social media. AI drafts strip it every time. Build broker identification into the prompt and the template rather than catching it in review.

Is it unethical to use AI for client communication?

The ethics question is usually a proxy for a marketing worry — will clients think less of me. Nobody discloses that a transaction coordinator drafted the letter. The real obligations are accuracy, your own judgment on every message before it goes out, and consent before a machine contacts anyone. An agent who reads and edits an AI draft is on firmer ground than one sending unread boilerplate they wrote in 2019.

Bring this to your team or event

Emily Terrell speaks at brokerage events, real estate conferences, and team trainings on AI, systems, and social media — the exact playbook in this post, delivered live to your audience. As a Top Coach and Speaker at Tom Ferry International and an active agent closing 70+ transactions a year, Emily speaks from the stage about what’s working right now, not theory. Recent stages include NAHREP and eXp Con.

Book Emily to speak at your next event:
Email: eterrell@yourcoach.com
Phone: (210) 400-9191
Web: coachemilyterrell.com

For real estate agents who want to implement this: Get the weekly real estate prompt library at weeklyrealestateprompts.com or follow @coachemilyterrell on Instagram for daily systems and AI breakdowns.

Claude for Real Estate Agents: The Setup That Actually Works

Written by Coach Emily on August 28, 2026. Posted in Uncategorized.

By Emily Terrell — Top Coach and Speaker at Tom Ferry International. 4 years coaching at Tom Ferry, 9 years prior as a client.

Claude for real estate agents works best when you stop treating it as a chatbot and set it up as a workspace. Two features do the heavy lifting: Projects, which hold your market data and brand voice permanently, and Skills, which store a repeatable workflow you never re-explain. This post walks the exact three-layer setup, including where compliance guardrails live.

Key Takeaways

Claude isn’t in the top three AI tools agents use, which means the agents using it well have less competition for the output quality.
A Claude Project is a persistent workspace with its own instructions and uploaded files, so every chat starts already briefed on your business.
A Skill is a reusable playbook Claude loads on its own, which is where your fair housing and TREC guardrails belong permanently.
The setup is a one-time build of about 90 minutes; after that, prompts get shorter and output gets more consistent.
Compliance is a structural problem, not an editing problem — build the constraint into the system, not into your final read-through.

What is Claude, and how is it different for real estate agents?

Claude is a general-purpose AI assistant built by Anthropic, used through claude.ai or its desktop and mobile apps. For agents, the meaningful difference isn’t the model — it’s the two organizing features around it.Projects let you create self-contained workspaces with their own chat histories and knowledge bases, where you upload documents, provide context, and have focused chats. Skills go further: they package your workflows and institutional knowledge so Claude applies them consistently, and anyone can create one by writing instructions in Markdown — no coding required. Claude Claude Platform Docs

That combination matters because the real problem with AI in a real estate business isn’t output quality. It’s that you re-explain your market, your voice, and your compliance rules every single time you open a blank chat.

Why this matters for real estate agents

Adoption is settled. The argument is over. RPR’s February 2026 survey of 225 NAR members found 82% currently use AI in their business, 68% use it daily or several times per week, and 34% save four or more hours per week. (RPR, February 2026) Coachemilyterrell

Here’s the thing nobody wants to tell you: the agents saving four hours a week and the agents saving zero are using the same tools. The difference is setup.

The concern data makes the point sharper. In that same RPR survey, 63% named accuracy of outputs as their top concern, 49% named compliance or legal issues, and 28% named Fair Housing specifically. Those aren’t tool problems. They’re architecture problems — and they get solved once, at setup, or they get re-litigated on every listing. Coachemilyterrell

And on the competitive side: NAR’s 2025 Technology Survey found the most common AI tools among Realtors were ChatGPT at 58%, Gemini at 20%, and Copilot at 15%. (NAR, September 2025) Claude doesn’t crack that list. In a market where the typical agent closed nine transaction sides in 2025 (NAR 2026 Member Profile), being one of the few agents in your market running a properly configured system is a real edge. Andreessen Horowitz Mile High Title Guy

“Most agents are still typing into a blank box. The agents winning with AI built a workspace once and never started from zero again. That’s a ninety-minute investment, not a personality trait.”
— Emily Terrell, Tom Ferry Coach

The three-layer Claude setup for a real estate business

Layer 1: What goes in a Claude Project?

A Project is the room. Everything you’d otherwise re-explain goes in here once. Projects are available to all users including free accounts, with free users capped at five projects, and paid plans automatically scaling large knowledge bases through retrieval so you can upload more without breaking anything. (Anthropic, “What are projects?”) Claude

Build one Project called Listings. Upload:

Your last five listing descriptions you were proud of, so it learns your voice from evidence rather than adjectives
Your local market stats sheet, refreshed quarterly
Your brokerage’s advertising disclosure requirements
A one-page brand voice doc

Then write Project instructions: who you are, what market you serve, what you never say, and what format you want back. Every chat inside that Project loads all of it before you type a word.

Layer 2: What should your first Skill be?

A Skill is the playbook. Skills are available on Free, Pro, Max, Team, and Enterprise plans, and require code execution to be enabled in settings before you can turn them on in Customize > Skills. (Anthropic, “Use skills in Claude”) Claude

Don’t build five. Build one: New Listing Marketing Set. Its instructions should specify the exact deliverables — MLS description at your target word count, three social captions, one email to sphere, one thirty-second video script — plus the fair housing constraints and your required disclosures. Anthropic’s own guidance is to keep skills focused, write clear descriptions so Claude knows when to invoke them, and start with plain Markdown before adding anything complex. Tom Ferry

You can also describe your process out loud and have Claude build the Skill file for you, which is how most agents should do it.

Layer 3: Where do the compliance guardrails live?

In the Skill. Not in your head, and not in the final proofread.

This is the part most AI training skips, and it’s why the 49% compliance-concern number stays stuck. If your fair housing constraint lives in the Skill instructions, it applies to every listing automatically, including the one you generate at 9pm on a Friday. If it lives in your review process, it applies on the days you’re sharp.

Texas license holders have a second layer: TREC Rule 535.155 governs advertising, and AI-generated marketing is still advertising. Agents outside Texas need the equivalent rule from their own commission. I cover this in more depth in what AI training for real estate agents must actually cover.

This is general information, not legal advice. Consult your broker or attorney on specific listings.

How I use this in my own business

I close 70+ transactions a year on roughly five hours of active management per week, and the Listings Project is a load-bearing part of that math.

When I take a listing in Stone Oak, I don’t open a blank chat. I open the Project, paste the property specs, and the marketing set comes back in my voice with the disclosures already in place. I edit for accuracy — always, every time, because the specs are mine to verify and no tool changes that — and it goes into the MLS once. From there the syndication system handles distribution and Follow Up Boss fires the sequence to my sphere.

The setup took me an afternoon. It has run for months without maintenance beyond a quarterly stats refresh. That’s the whole pitch: build it once, and it’s scalable and repeatable.

Common mistakes

Using chats instead of Projects. If you’re pasting your brand voice into the prompt every time, you’re doing manual labor a Project does for free.

Building nine Skills before finishing one. Focused skills compose well; sprawling ones misfire. Ship one, run it for thirty days, then build the second.

Treating the compliance review as the guardrail. Review catches errors. Structure prevents them. You need both, but the structure is what scales.

Pasting client financial or contact data into a general-purpose chatbot. Decide what goes in and what never does before you’re in a hurry.

Uploading your best listing descriptions and skipping the market data. Voice without local facts produces a fluent copy that says nothing specific. The structured listing data is the input; the marketing is the output.

Frequently Asked Questions

Is Claude better than ChatGPT for real estate agents?

Neither wins outright, and the framing is the problem. ChatGPT leads on adoption at 58% of Realtors per NAR’s 2025 Technology Survey, which means more peers to ask for help. Claude’s advantage is the Projects-plus-Skills architecture for repeatable, voice-consistent workflows. Pick one, configure it properly, and stop tool-shopping — configuration beats tool choice every time.

Is Claude free for real estate agents?

Yes, with limits. Free accounts can create up to five Projects and access Skills once code execution is enabled in settings. Paid plans add automatic retrieval that scales large knowledge bases without hitting context limits, which matters once you’re uploading a full market stats sheet plus years of listing copy. Start free, upgrade when your Project outgrows it.

What should I put in a Claude Project for my real estate business?

Four things: five listing descriptions you’re proud of, your current local market stats, your brokerage’s advertising and disclosure requirements, and a one-page brand voice document. Then write Project instructions covering your market, your audience, your format preferences, and the phrases you never use. That combination is what makes every subsequent chat start informed instead of blank.

Can Claude write MLS listing descriptions that comply with fair housing?

It can produce a compliant copy, but the responsibility stays with you. Build the fair housing constraint into your Skill instructions rather than catching violations during edit, then review every description before input. RPR’s February 2026 survey found 28% of agents cite Fair Housing as a top AI concern, which is exactly why the guardrail belongs in the system.

Can I upload client data to Claude?

Decide this deliberately, not in a hurry. Property specs, public market data, and your own marketing copy are straightforward. Client financial details, contact records, and anything you’d be uncomfortable seeing outside your CRM are a different category with real consequences. Write your rule down before you need it, and give the same rule to anyone on your team.

How long does it take to set up Claude for a real estate business?

About ninety minutes for the first Project and one Skill, done properly in a single sitting. Budget a quarterly fifteen-minute refresh of your market stats after that. The payoff shows up in prompt length: once the workspace is built, requests drop from three paragraphs of context to one line of property specs.

Bring this to your team or event

Book Emily to speak at your next event:
Email: eterrell@yourcoach.com
Phone: (210) 400-9191
Web: coachemilyterrell.com

Can I Use an AI Avatar for Real Estate Videos?

Written by Coach Emily on August 28, 2026. Posted in Uncategorized.

By Emily Terrell — Top Coach and Speaker at Tom Ferry International. 4 years coaching at Tom Ferry, 9 years prior as a client.

Yes, with a narrow boundary: use an AI avatar for repeatable, script-driven content like market updates and FAQ answers and keep your real face on anything where a client is deciding whether to trust you. No state currently requires disclosing an agent’s own avatar, but your advertising rules still apply. Here’s the three-question test.

Key Takeaways

Every state AI disclosure law written so far regulates images of the property, not a synthetic version of the agent — which means the avatar question is a judgment call, not a compliance checkbox.
The avatar is safe where it delivers information and dangerous where it substitutes for relationship.
TREC Rule 535.155 governs an avatar video exactly like any other advertisement, and your AI script will strip your broker’s name out every single time.
An avatar delivering a client testimonial is not a gray area — that’s an FTC problem.
Build the avatar from your own likeness with documented consent, never from a stock presenter, or you’ve quietly told buyers the person representing you doesn’t exist.

What is an AI avatar in a real estate video?

An AI avatar is a synthetic on-camera presenter generated from recorded footage of a real person, driven by a text script instead of a live take. Tools like HeyGen build a model of your face and voice, then produce a new video of “you” saying anything you type, in any language, with matched lip-sync.

The output looks like footage. It isn’t. Nothing was filmed. That single fact is the entire subject of this post.

Why this matters for real estate agents

Most agents will reach for an avatar for the wrong reason: they hate filming. That instinct sends the avatar straight to the content that matters most, because the content agents avoid filming is usually the content where they have to be personally persuasive.

The production pressure behind that instinct is real. According to NAR’s 2026 Member Profile (June 25, 2026), the typical agent closed nine transaction sides in 2025 and now has 13 years of experience — a more seasoned field competing for a smaller pool of transactions. Content volume is one of the few levers left.

But volume without judgment is exactly where AI has been failing agents. According to NAR’s 2025 Technology Survey (September 18, 2025), 68% of agents now use AI, yet only 17% reported a significantly positive impact on their business and 46% reported no noticeable impact at all.

Read that as a warning about avatars specifically. An avatar multiplies your output. If your output wasn’t converting, you’ve just built a faster machine for producing content nobody responds to — and added a trust liability on top.

“An avatar can deliver your information. It can’t deliver your credibility. The moment a seller senses the person on screen was never actually in the room, you’re not saving time — you’re spending trust you’ll need at the listing table.”
— Emily Terrell, Tom Ferry Coach

The three-question avatar test

Run every planned video through these in order. Stop at the first “yes.”

Is the avatar making a claim about the property?

If your avatar says the kitchen was renovated, the roof is four years old, or the lot backs to greenbelt, stop. You’ve left avatar territory and entered listing-advertisement territory, where a different and much stricter set of rules governs what you can show and what you have to disclose. That’s covered in detail in what real estate agents have to disclose when using AI.

The short version: property claims need verification against the MLS record and, in some states, disclosure of any generated media. Don’t hand that job to a script you wrote from memory.

Is the avatar standing in for you in a moment of trust?

Listing presentations. Buyer consults. Follow-up after a lost offer. Anything a client will remember as a conversation with you. If the avatar goes there, the answer is no — not because a regulator will catch it, but because a client eventually will.

Here’s the thing nobody wants to tell you: the cost of getting caught isn’t a fine. It’s that every past interaction gets re-evaluated. A client who learns the “personal video” was synthetic doesn’t just distrust that video. They wonder what else wasn’t real.

Is the content repeatable, scripted, and about information?

If you got here, you have a green light. Monthly market updates. Answers to the twelve questions every buyer asks. Listing-goes-live announcements. Translated versions of content you already filmed with your actual face.

These share a property: the value is in the information, not in the delivery. Nobody watches a market update to feel connected to you. They watch it to learn what happened to inventory. That’s the band where the avatar earns its keep.

What the law actually says about agent avatars

Does any state require you to disclose your own AI avatar?

Not as written. California’s AB 723, effective January 1, 2026, added Section 10140.8 to the Business and Professions Code and requires disclosure when a digitally altered image of the real property appears in an advertisement, plus access to the original. It defines a digitally altered image as one changed to add, remove, or modify elements of the property. Your face isn’t an element of the property.

Wisconsin’s 2025 Act 69 takes the same shape and lands in 2027. Texas is quieter still — TRAIGA took effect January 1, 2026, but its consumer AI-notice duty applies to governmental entities, not private licensees, a point covered in what actually risks your real estate license.

So the honest answer: no statute currently reaches an agent’s own avatar. That’s a gap in drafting, not a permission slip. Legislatures wrote for the problem in front of them — fake listing photos — and video of a synthetic agent arrived after the pen went down.

This is general information, not legal advice. Confirm your obligations with your broker, your MLS, and an attorney licensed in your state before changing your workflow.

What TREC still requires on every avatar video

All of it. TREC’s advertising rules define an advertisement as any communication by or on behalf of a license holder designed to attract the public to use real estate brokerage services, explicitly including electronic media, social media, and the internet. Rule 535.155 requires the name of the license holder or team plus the broker’s name at a minimum of half the size of the largest contact information in the ad. On social media, that information can live on a directly linked profile page.

Your avatar tool doesn’t know any of this. Feed it a script and it produces a compliant-looking video with no broker identification anywhere. Build the broker block into your lower-third template once, and it rides along on every export.

Where the FTC rule bites

One use case is not a gray area at all. Under the FTC’s final rule on consumer reviews and testimonials, 16 CFR Part 465, businesses are prohibited from creating or disseminating testimonials that misrepresent that they come from someone who does not exist — the rule names AI-generated fake reviews specifically — or from someone who never actually had the experience described.

An avatar delivering a client testimonial is that violation with a face on it. So is a synthetic “past client” in a marketing video. Don’t build it, don’t buy it from a vendor who offers it as a feature.

How I use this in my own business

I run 70+ transactions a year in San Antonio on roughly five hours a week of active management, and video is part of that system. My rule is one sentence: the avatar gets the information, my face gets the people.

Here’s the actual how. My monthly Stone Oak market update runs through an avatar built from my own footage, because the value in that video is the absorption rate and the median days on market, not my energy. I write the script from MLS data, run it, export it, and it’s live in twenty minutes. Feet on the desk, coffee in hand.

Everything on the seller side gets filmed. The listing presentation follow-up, the “here’s what happened at the open house” video, the call after a low appraisal — I film those on a phone, badly lit, in whatever I was wearing that day. Those convert better precisely because they’re rough. The imperfection is the proof I was there.

The dividing line came out of my batch filming system, which is where I first drew it: use AI for the repeatable stuff, keep your face where trust is being built. The avatar didn’t change that rule. It just made it easier to break.

Common mistakes

Using a stock avatar instead of your own likeness. A generic presenter with your brokerage logo tells a buyer your marketing features a person who doesn’t exist. That’s a different problem from a synthetic version of you, and a worse one.

Sending avatar video into one-to-one client communication. The moment a video is addressed to a specific person, it reads as personal attention. Automating that is the one thing that can’t be walked back.

Letting the script write property claims. Your avatar will state a square footage with total confidence. Verify every property fact against the MLS record before it reaches a script, not after.

Skipping the broker block. AI-generated video has no idea Rule 535.155 exists. Build it into the template so you’re never relying on memory at export.

Building an avatar without documented consent and a revocation path. Get your tool’s consent, retention, and deletion policy in writing before you upload training footage of your own face. You’re handing over a biometric asset you’ll want to control later.

Assuming no law means no risk. State AI statutes haven’t reached agent avatars yet. Misleading-advertising rules and consumer expectations already have.

Frequently Asked Questions

Do I have to disclose that a real estate video uses an AI avatar?

No state currently requires it for an avatar of the agent. California’s AB 723 and Wisconsin’s Act 69 both regulate altered images of the property, not the agent’s likeness. That said, misleading-advertising rules apply regardless, and a simple on-screen note costs nothing. Check your MLS rules and brokerage policy separately, since both can impose requirements beyond state law.

What kind of real estate videos work best with an AI avatar?

Repeatable, script-driven content where the information carries the value: monthly market updates, answers to common buyer and seller questions, listing-live announcements, and translated versions of videos you have already filmed. These are videos viewers watch for the content, not for connection with you. Anything a client experiences as a personal conversation should be filmed with your actual face.

Can I use an AI avatar for a listing tour video?

Not for the property claims. If the avatar describes features, condition, or square footage, you’re making listing representations that require verification and, in some states, disclosure of generated media. An avatar introducing a tour is different from an avatar narrating one. Keep the property claims tied to verified MLS data and filmed or photographed reality.

Is it legal to use an AI avatar in real estate marketing in Texas?

Yes. TRAIGA took effect January 1, 2026, but its consumer AI-notice duty applies to governmental entities, not real estate licensees. TREC Rule 535.155 still governs the video as an advertisement, meaning broker identification is required and the ad cannot mislead. Legal to use is not the same as advisable everywhere — the three-question test still applies.

Will clients trust me less if they find out I used an AI avatar?

It depends entirely on where you used it. A synthetic market update reads as efficient. A synthetic video that a client believed was filmed for them reads as deception, and it retroactively taints every other interaction. The trust cost isn’t in using the tool. It’s in using it somewhere a person expects you.

Can an AI avatar deliver a client testimonial?

No. The FTC’s rule at 16 CFR Part 465 prohibits creating or disseminating testimonials that misrepresent that they come from a person who does not exist or who never had the experience described, and it names AI-generated content specifically. A synthetic client is a fabricated endorsement. Film real clients with written permission, or use written testimonials with attribution.

Do I still need my broker’s name on an AI avatar video?

Yes, on every one. TREC defines an advertisement to include electronic media, social media, and the internet, and Rule 535.155 requires the license holder or team name plus the broker’s name at half the size of the largest contact information. Avatar tools never add this. Build it into your lower-third template so it exports automatically.

Bring this to your team or event

Book Emily to speak at your next event:
Email: eterrell@yourcoach.com
Phone: (210) 400-9191
Web: coachemilyterrell.com

Brokerage AI Adoption: Why Most Training Doesn’t Stick

Written by Coach Emily on August 28, 2026. Posted in Uncategorized.

By Emily Terrell — Top Coach and Speaker at Tom Ferry International. 4 years coaching at Tom Ferry, 9 years prior as a client.

Brokerage AI adoption fails after training because the session is designed as an event and adoption is a system. Agents are independent contractors, so a brokerage can’t mandate a tool — it can only make the new workflow easier than the old one. This guide covers the five failure points and the 90-day structure that fixes them.

Key Takeaways

Your agents already use AI — the gap isn’t adoption, it’s frequency and depth.
A brokerage has no compliance authority over an independent contractor’s tech stack, so mandates don’t work and incentives do.
Training without a named internal owner and a 30/60/90 cadence decays inside three weeks.
No pre-session baseline means no post-session proof, which means no budget next quarter.
Compliance silence reads as risk to agents, and cautious agents quietly stop using the tool.

What is brokerage AI adoption?

Brokerage AI adoption is the share of your agents who use an AI workflow frequently enough and deeply enough to change a business result — not the share who attended the training. Attendance measures the event. Adoption measures behavior thirty days later. Most brokerages track the first number and assume it stands in for the second.

Why this matters for real estate agents and the leaders who train them

The industry data makes the gap unusually clear. According to NAR’s 2025 Technology Survey, 17% of agents reported a significantly positive impact from AI, 33% saw a moderately positive impact, and 46% said AI had no noticeable impact at all. Nearly half the profession is using a technology and getting nothing measurable from it. Inman

The frequency breakdown explains why. The same survey found 20% of agents use AI tools daily, 22% weekly, 27% a few times a month, and 32% haven’t used AI in their business at all. That “few times a month” cohort is the signature of training that didn’t stick. Those agents learned something, tried it, and drifted back. Daily use is where behavior change actually lives. Inman

Here’s the part that should reframe how you budget. Two out of three agents either agree (38%) or strongly agree (29%) that their brokerage provides all the tech tools they need. Your agents are telling you the tools aren’t the problem. Buying another platform won’t move the number. And 24% of agents already spend over $500 per month on technology out of their own pockets — they’re not waiting on you for access. Inman

Adoption has been near-universal for a while. Realtors Property Resource’s February 2026 survey of 225 NAR members put AI adoption at 82%, and industry reporting has shifted accordingly: the divide that matters now is between agents who adopted a tool and agents who rebuilt a workflow around one. Cameron Walker of Clever Offers, quoted in that Inman piece, made the point plainly — buying the tool isn’t what produces results; rebuilding the strategy around it is. Coachemilyterrell + 2

Your training either produces a rebuilt workflow or it produces a pleasant morning.

The five reasons AI training doesn’t stick

Why can’t a brokerage just mandate AI use?

Because your agents are independent contractors, and every other failure below is downstream of this one. You control the room for ninety minutes and you control nothing after. Corporate training models assume compliance authority that a brokerage structurally does not have, so the rollout gets designed like an employee mandate and lands like a suggestion.

The workaround isn’t authority. It’s friction. Make the new workflow measurably faster than the old one, prove it in the room with the agent’s own listing, and let self-interest carry it. Agents adopt what saves them time on the task they already hate.

Who owns adoption after the trainer leaves?

Usually nobody, which is why it dies. There’s no internal owner, no slot in the weekly sales meeting, no check-in date. The session is a line item on a calendar rather than phase one of anything.

Name a person before you book the trainer. Not the broker — a producing agent your people already respect, ideally someone who’ll be visibly better at this in sixty days. Their job is a five-minute standing agenda item: one workflow, one agent demoing it, every week for twelve weeks.

Why does nothing get measured?

Because measurement requires a baseline, and baselines have to be captured before the session starts. If you don’t know what your agents spent on listing copy, follow-up drafting, and captions last month, you have no way to prove the training worked and no way to defend the spend at renewal.

The fix takes ten minutes at the top of the session. Every agent writes down their rough hours on those three tasks from the previous month. That number is the hook — agents badly underestimate it until they add it up — and it’s the only honest scorecard you have at day 90.

Why do agents leave with notes instead of a working asset?

Because the session was a tour. Watching a demo produces admiration. Building produces adoption. If an agent walks out without a saved prompt in their own logged-in account, tied to their own live listing, retention is functionally zero by Thursday.

This is the single highest-leverage change you can make to an agenda. One live build on real material beats five tool demos, every time. The full session agenda, timing, and pre-work list is worth reading before you brief any trainer.

Why does compliance silence kill adoption quietly?

Because an agent who isn’t sure whether AI-assisted listing copy is allowed will simply stop using it, and they won’t tell you why. Fair Housing exposure in generated descriptions, state advertising rules on how agents describe themselves, and client data going into consumer chatbots are all real questions. When the brokerage says nothing, the cautious agents — often your best ones — read the silence as risk.

Treat compliance as a working segment with a real example, not a closing disclaimer. Build the guardrails into the prompt itself so the compliant version is the default output, the way it works when you automate listing syndication with a reusable prompt. This is general information, not legal advice — consult your broker or attorney on specific listings and campaigns.

The 90-day structure that makes it stick

“Training is a phase, not an event. If your only line item is the speaker fee, you bought a morning — not adoption.”
— Emily Terrell, Tom Ferry Coach

Days minus-14 to 0: pre-work

Agents arrive with a real active listing or real cold lead open on their device, a logged-in and tested account on the tool being used, and their hours baseline written down. No account setup in the room. Setup time is build time you’re paying for and not getting.

Days 1 to 30: one workflow, one owner

Pick a single workflow and refuse to add a second. Follow-up drafting or listing copy — not both. The internal owner runs the five-minute weekly slot. One agent demos their version. The rest of the meeting proceeds as normal.

Days 31 to 60: wire it into the system

Whatever agents built has to live somewhere it persists — a CRM template, a saved prompt library, an email sequence. Faster content production with no system underneath it is just faster chaos. The same logic applies to any process you eventually want to hand off: build the workflow yourself first, then document it.

Days 61 to 90: measure against the baseline

Same three tasks, same question, same agents. Hours now versus hours in month zero. Report it to the room, not just to leadership — agents who see the aggregate number move are the ones who tell the holdouts.

How I run this in my own business

When I rolled an AI follow-up workflow out to my own San Antonio team, the first attempt failed exactly the way I’ve described. I ran a great session, everyone was enthusiastic, and three weeks later two people were using it. The second attempt, I did one thing differently: I made my transaction coordinator the owner and gave her five minutes at the top of every Monday meeting. Same content, same tool. The difference was that somebody’s name was on it every week.

That’s also how I close 70+ transactions a year on roughly five hours a week of active management. Not because the tools are better than yours — because there’s a cadence underneath them.

Common mistakes

Booking a tool tour instead of a build session. Tool tours produce interest. Builds produce adoption. If the agenda has more than two tools in it, it’s a tour.
Handing out a prompt list. A prompt list has a shelf life of roughly one model release. Teach prompt architecture so the skill survives the next update.
Ending with energy instead of a date. No named workflow, no owner, no check-in — nothing survives the drive home.
Training the whole roster at once. A mixed room gets pitched at the middle and lands for nobody. Split by production tier or by comfort level, not by office.
Treating the trainer’s contract as the whole engagement. If reinforcement isn’t in the scope, adoption isn’t in the outcome.

Frequently Asked Questions

Why doesn’t AI training stick at most brokerages?

Because the training is designed as an event and adoption is a system. Agents are independent contractors, so mandates don’t apply, and without a named internal owner, a reinforcement cadence, and a measured baseline, the workflow decays within about three weeks. NAR’s 2025 Technology Survey shows 46% of agents report no noticeable impact from AI despite widespread use.

How do I measure whether AI training actually worked?

Capture a baseline before the session: ask every agent to write down their hours from the previous month on listing copy, follow-up drafting, and social captions. Re-ask the same three questions at day 90 with the same agents. The delta is your ROI number. Satisfaction surveys measure the morning, not the behavior change.

Should a brokerage mandate specific AI tools for its agents?

No, and structurally you can’t. Independent contractor status means you have no authority over an agent’s tech stack. What works instead is making the new workflow demonstrably faster than the old one and proving it in the room on the agent’s own live listing. Self-interest carries adoption where a mandate stalls.

Who should own AI adoption inside a brokerage?

A producing agent your people already respect — not the broker, and not an admin. Their entire job is a five-minute standing item in the weekly sales meeting for twelve weeks: one workflow, one agent demonstrating it. Naming this person before you book the trainer is the highest-return decision in the whole process.

How long should a brokerage AI training session be?

Sixty minutes is the working minimum for a session that includes a live build. Anything shorter forces a demo instead of a build, which produces interest but not adoption. Half-day formats allow two builds plus in-room CRM configuration. A 45-minute keynote slot works if you cut the CRM block and keep the build and compliance segments intact.

Why do agents stop using AI tools after training?

Two reasons dominate. First, they left with notes instead of a working asset saved in their own account. Second, they’re unsure what’s compliant — around Fair Housing language in generated descriptions or client data in consumer chatbots — and brokerage silence reads as risk. Cautious agents quietly stop rather than ask.

Bring this to your team or event

Book Emily to speak at your next event:
Email: eterrell@yourcoach.com
Phone: (210) 400-9191
Web: coachemilyterrell.com

How to Train Your Real Estate Team on AI in 90 Days

Written by Coach Emily on August 28, 2026. Posted in Uncategorized.

By Emily Terrell — Top Coach and Speaker at Tom Ferry International. 4 years coaching at Tom Ferry, 9 years prior as a client.

To train your real estate team on AI, run one 60-minute build session, then protect 90 days of reinforcement behind it. Week one installs a single workflow. Weeks two through twelve enforce it with a named owner, a weekly checkpoint, and a day-30 measurement. This guide gives the full 90-day plan, the ownership map, and the adoption metric.

Key Takeaways

Training is an event; adoption is a schedule — the session is week one of thirteen, not the whole plan.
Two-thirds of agents say their brokerage already gives them every tech tool they need, and most still report AI hasn’t changed their business.
Roll out one workflow per month, not a toolkit — coverage isn’t the constraint, retention is.
Every workflow needs one named owner on the team, because a group commitment is nobody’s commitment.
Measure how many agents are still running the workflow on day 30. Room energy isn’t a metric.

What is an AI rollout for a real estate team?

An AI rollout is a scheduled sequence that installs a small number of AI workflows into a team’s existing operations over a defined period, with named owners and a measurement date. It’s not a training session. The session is one hour inside it.

The distinction matters because most team leaders buy the hour and skip the twelve weeks. Then they conclude AI doesn’t work for their team. What didn’t work was a one-hour intervention against a fifteen-year habit.

Why this matters for real estate teams

Teams have more to gain and more to lose here than solo agents, because a workflow that fails on a team fails five times.

According to NAR’s 2026 Member Profile (June 2026), 21% of Realtors worked as part of a team in 2025, with a median of four members. The production gap is the whole argument: the typical individual agent closed nine transaction sides, while the median team closed 32 sides on $17.5 million in volume against $2.7 million individually, per HousingWire’s coverage of the report. Leverage already exists on a team. AI either multiplies it or adds noise to it.

Here’s the number that should stop you. According to NAR’s 2025 Technology Survey (September 2025), 38% of agents agree and another 29% strongly agree that their brokerage provides all the technology tools they need. Two out of three. And in that same survey, only 17% reported AI had a significantly positive impact on their business, while 46% saw no noticeable difference at all — with adoption already at 68%, per HousingWire’s analysis.

Your team almost certainly has the tools. What it doesn’t have is a rollout.

“Nobody on your team is failing at AI because they lack a tool. They’re failing because no one told them which task to stop doing manually, by when, and who’s checking.”
— Emily Terrell, Tom Ferry Coach

The 90-day AI rollout plan

Three phases. One workflow per phase. If that sounds slow, look at your last rollout of anything and count what survived.

Days 1–30: install one workflow

Pick the task your team does most often and enjoys least. For most teams that’s listing copy or follow-up drafting. One task, not a category.

Run the build session in week one. The full agenda, timing, and pre-work requirements are in what AI training for real estate agents must cover — the short version is that agents build on their own live listing or their own cold lead, or nothing sticks.

Then three things in weeks two through four. The workflow gets written down in one page, stored where the team already looks. It gets a five-minute slot at the top of your existing team meeting, where one agent shows their actual output. And you set the day-30 count date on the calendar in week one, before anyone has an excuse ready.

Days 31–60: add the guardrail and the second workflow

Month two is where teams either mature or start generating liability.

Write your team’s AI use rule as a one-page document every agent signs. Four items: every AI-generated property fact gets verified against the MLS record before publishing, generated listing copy gets reviewed against fair housing language standards (see NAR’s fair housing resources), AI-generated marketing follows your state’s advertising rules the same as any other advertising, and client financial or contact data never goes into a general-purpose chatbot. This is general information, not legal advice — have your broker or attorney review the document before your agents sign it.

Then add workflow two. Something adjacent, not something new — if month one was listing copy, month two is the marketing assets that follow it. Automating MLS listing syndication covers where that handoff actually happens.

Days 61–90: transfer ownership and audit

You stop running it. Somebody else does.

Name a workflow owner for each of the two workflows. That agent maintains the one-page doc, updates the prompts when a model changes and the output degrades, and runs the five-minute meeting slot. If a workflow doesn’t have a person’s name on it by day 90, it’s already dead and you haven’t noticed.

Then audit. Count how many agents ran each workflow in the last two weeks. Compare against the baseline hours you captured in week one. Report the number to your team out loud, including if it’s bad.

Who owns what

Three roles, and one person can hold more than one on a four-person team.

The team leader owns the calendar and the count. Not the teaching, not the prompts — the schedule and the measurement. This is the role that gets abandoned first because it produces no visible output.

The workflow owner owns one workflow’s documentation, prompt maintenance, and demo slot. Give this to the agent who was most frustrated during the build session, not the most enthusiastic. The enthusiastic one already has it working and won’t notice when it breaks for everyone else.

The compliance checker owns spot-checking published output against the one-page rule. Rotate this monthly so every agent has to read the document at least once with real attention.

If you’re weighing whether these roles go to agents or to a VA, the tracking and accountability side of that decision is the subject of my Social Handoff session.

How I use this in my own business

I close 70+ transactions a year in San Antonio in roughly five hours a week of active management, and the reason that works is that nothing lives in my head. Every AI workflow I run has a document and a destination.

When I rolled AI listing copy into my own operation, month one was only that — listing copy. Nothing else. My transaction coordinator owned the prompt doc, not me, because I’m the least reliable person to maintain something I already know how to do from memory. Month two added the social assets that come off the same listing data; the batch approach I use for that is in the video editing batch system.

The part I got wrong the first time: I measured whether people liked it. Month two I switched to counting how many listings went out using the workflow. The number was lower than the enthusiasm, which is the only reason I found the gap.

Common mistakes

Rolling out a toolkit instead of a workflow. Five tools in one session produces zero adopted workflows. One task, one month.
Treating the training session as the rollout. The hour is 1/13th of the work and the easiest part to buy.
Leaving workflows unowned. “The team will maintain it” means it breaks silently at the next model update.
Writing the AI use rule after something goes wrong. It takes an hour in month two and costs a license in month nine.
Measuring sentiment instead of usage. Ask how many agents ran it in the last two weeks. That’s the only number.
Skipping the baseline. Without week-one hours captured, you can’t prove the rollout worked to your team or to yourself.

Frequently Asked Questions

How long does it take to train a real estate team on AI?

The session takes 60 minutes. The rollout takes 90 days. Plan on one build session in week one, a five-minute reinforcement slot in your existing weekly meeting, and a formal count at day 30 and day 90. Teams that budget only the session consistently report no measurable change, which matches what the broader adoption data shows.

How many AI tools should my team use?

One tool for the first 90 days, and one workflow per month inside it. Most agents already have access to a general AI assistant, so building where the team already has accounts removes a setup barrier. Adding a second tool before the first workflow is running daily is the most common way teams stall.

Who should own AI training on a real estate team?

Split it. The team leader owns the schedule and the measurement. A named workflow owner — ideally the agent who struggled most in the build session — owns documentation and prompt maintenance for each workflow. A rotating compliance checker spot-checks published output. Unowned workflows break silently and nobody reports it.

Do I need an AI policy for my real estate team?

Yes, and one page is enough. Cover verification of AI-generated property facts against the MLS record, fair housing review of generated listing copy, state advertising rules applied to AI-generated marketing, and a hard rule on client data in consumer AI tools. This is general information, not legal advice — have your broker or attorney review it.

How do I get agents who resist AI to actually use it?

Stop selling the technology and assign the task. Resistance usually isn’t philosophical, it’s that no one specified which task to stop doing manually or by when. Give a resistant agent the workflow owner role rather than exempting them. Ownership converts faster than persuasion, and their skepticism makes the documentation better.

How do I measure whether AI training worked for my team?

Count how many agents ran the named workflow in the last two weeks, at day 30 and again at day 90. Capture baseline hours on the target task in week one so you have a before number. Adoption rate and hours recovered are the metrics. Survey scores and post-session energy predict nothing.

Bring this to your team or event

Book Emily to speak at your next event:
Email: eterrell@yourcoach.com
Phone: (210) 400-9191
Web: coachemilyterrell.com

Should Real Estate Agents Hire a VA or Use AI in 2026?

Written by Coach Emily on August 28, 2026. Posted in Uncategorized.

By Emily Terrell — Top Coach and Speaker at Tom Ferry International. 4 years coaching at Tom Ferry, 9 years prior as a client.

Hire a VA for work that fails loudly — transaction deadlines, client follow-through, database hygiene. Use AI for work that’s high-volume, low-judgment, and reversible — drafts, summaries, first-pass research. The two aren’t substitutes. This guide gives the delegation test I run before hiring anyone, plus what TREC actually lets an unlicensed VA do.

Key Takeaways

The dividing line isn’t cost — it’s failure mode. AI fails silently. A VA fails loudly, and loud failures get caught.
If you can’t write the SOP, neither option helps. You’ll just pay to have your chaos executed faster.
AI adoption is already near-universal, and most agents report it changed nothing about their business.
Texas law limits what an unlicensed VA can legally touch — showing property and prospecting calls are off the table.
The highest-leverage setup isn’t VA or AI. It’s a VA running AI inside a documented workflow.

What is the VA vs. AI decision actually about?

The VA vs. AI decision is a delegation question disguised as a budget question. A virtual assistant is a person who takes accountability for outcomes over time. AI is a tool that produces output on demand with no accountability at all. You’re not choosing between two versions of “help” — you’re choosing which kind of failure you can afford in a given task.

That distinction gets lost because both options get pitched the same way: save time, cut costs, do more. Neither claim tells you which work to hand over.

Why this matters for real estate agents

The economics have narrowed, and the wrong hire is expensive. According to NAR’s 2026 Member Profile (June 2026), the typical agent closed nine transaction sides in 2025 with a median gross income of $59,200, while median business expenses climbed to $9,530 from $8,010 the year before. Expenses grew faster than income. A VA hire that doesn’t reduce a real bottleneck comes straight out of that gap.

Meanwhile, most agents have already run the AI experiment and gotten nothing back. NAR’s 2025 Technology Survey found that 17% of agents reported a significantly positive impact from AI, 33% a moderately positive one, and 46% no noticeable impact at all (NAR). Nearly half of the people who tried it saw their business look exactly the same afterward.

Here’s the thing nobody wants to tell you: those two numbers are the same problem. Agents hiring a VA to fix an undefined process and agents buying an AI tool to fix an undefined process are making the identical mistake, and both get the identical result.

“Delegation doesn’t create a system. It exposes whether you had one. If you hand off chaos, you don’t get relief — you get chaos with a second person’s name on it.”
— Emily Terrell, Tom Ferry Coach

The test that decides it: does the failure make noise?

Why does failure mode matter more than cost?

Because a mistake you catch is cheap and a mistake you don’t catch closes a file. AI produces confident, fluent, wrong output and gives you zero signal that anything went sideways. A generated listing description with an invented square footage reads exactly as well as an accurate one. It goes live, it sits there, and nobody notices until a buyer’s agent does.

A person fails differently. A VA who misses a deadline tells you, or your client does, usually within a day. That’s not a smaller failure — it’s a visible one, and visible failures get fixed before they compound.

So the first question isn’t “what’s cheaper.” It’s: if this task goes wrong, how long until I find out?

What should you hand to AI?

Text in, text out. High volume. Low judgment. Reversible.

That means MLS listing copy drafts, social captions, email first drafts, meeting and call notes, market summary drafts, script variations, and prompt-driven content batches. Nobody dies if the third draft is bad. You read it, you fix it, you move on.

AI is also the right answer when the work is structured and repetitive rather than relational. Pulling comparables into a recurring format, reformatting data, generating variations on a template — that’s what MLS automation for agents covers in detail, and it’s real leverage.

What AI can’t do: persist across time, chase a lender who isn’t responding, notice that a client sounded off on the phone, or take responsibility for anything.

What should you hand to a VA?

Anything requiring persistence, judgment about people, or accountability to a deadline.

Transaction coordination. Chasing title and lender. Scheduling photographers and inspections. Database hygiene — the unglamorous work of keeping your CRM from rotting. Following up when an automation didn’t fire, which it will. And the one agents forget when they’re comparing price tags: reviewing AI output before it publishes.

That last function is why the “VA or AI” framing collapses. Someone has to be the review step.

When is the answer neither?

When you can’t write the process down. If you can’t describe a task in ten steps or fewer, you don’t have a process — you have a habit of performing inconsistently. Handing that to a VA produces a confused person asking you questions all day. Handing it to AI produces output shaped like the wrong thing.

Write the SOP first. That week of documentation is the actual work, and it’s the part everyone skips.

What can an unlicensed VA legally do in Texas?

This is where the cost comparison stops being the interesting question. Under Texas law, an unlicensed person may not perform activities that require a real estate license, and TREC is explicit that the broker or sales agent who employs them can be exposed alongside them (TREC).

Off limits for an unlicensed VA in Texas:

Showing property. TREC amended the rule to clarify that “show” includes opening doors, allowing access to a property, or hosting an open house. An unlicensed assistant cannot do any of it.
Prospecting calls. Calling to determine whether someone wants to buy or sell, then setting an appointment for a licensed agent, requires a license. TREC calls this out by name as telemarketing.
Reviewing contracts or making deals work. An unlicensed person may not direct or advise agents in their work as license holders.
Qualifying callers. They can confirm previously advertised details about a specific property, after identifying themselves as unlicensed — but they cannot qualify the caller in any respect.

Permitted, at the direction of a license holder:

Scheduling an appointment for the agent to show a home
Inputting data and typing contracts as specifically directed
Placing signage and advertisements as directed
Bookkeeping and office administration

Two practical consequences. First, the offshore VA marketplaces that advertise “real estate lead qualification” are describing licensed activity. Second, your job description needs to be written down before day one, not reverse-engineered after a complaint. TREC’s own guidance recommends brokers establish written guidelines and training for agents and unlicensed personnel.

This is general information, not legal advice. Consult your broker and an attorney licensed in your state before defining a VA’s scope of work.

The advertising rule that catches both

Whether the caption came from a VA or from a model, it’s still your advertisement. TREC Rule 535.155 requires every advertisement to include the name of the license holder or team placing it, plus the broker’s name in at least half the size of the largest contact information for any sales agent, associated broker, or team name in the ad (TREC). Social media gets flexibility — the ad complies if it links to a profile page or separate page carrying the required information.

Neither a VA nor an AI tool knows this by default. Batch thirty captions from a model and you may have produced thirty non-compliant advertisements. I go deeper on that failure mode in what actually risks your license when you use AI.

What the data says about relying on AI alone

Adoption is not the constraint anymore. An RPR survey of 225 NAR members published in February 2026 found that 82% of agents currently use AI, 68% use it daily or several times a week, and 34% save four or more hours a week (RPR). Time savings are real and they’re being captured.

The same survey found what agents are worried about: 63% cited accuracy of outputs as their top concern, and 49% cited compliance or legal issues. Read that against the NAR finding that 46% of agents saw no noticeable business impact, and the picture is consistent. Agents are producing faster and trusting the output less, which is exactly what happens when volume goes up and the review step doesn’t exist.

A VA is a review step with a name attached. That’s not a soft benefit. That’s the thing that converts AI speed into usable work.

The setup that actually wins: a VA running AI

Stop treating them as alternatives. The compounding version is one person operating your AI stack inside a documented workflow you own.

What that looks like in practice:

You write the SOP. Ten steps or fewer per task. What triggers it, what the input is, what “done” looks like, who gets notified.
AI drafts. Listing copy, follow-up sequences, captions, summaries — generated against saved prompts that already contain your brand voice, your broker identification, and your fair housing constraints.
The VA reviews and routes. Facts verified against the MLS record. Broker name present. Then it goes to the destination, not to a chat window.
The system holds it. Output lands in your CRM, your transaction platform, or your content queue. Speed without a destination is just faster chaos — the same principle that makes listing syndication automation work or fail.
You check the checkpoints. One weekly review of exceptions, not a daily review of everything.

That’s the arrangement that produces leverage. A VA without AI is an expensive pair of hands. AI without a VA is unaccountable volume.

Common mistakes

Hiring to escape a process you never built. The VA becomes a full-time question-asker and you conclude that delegation doesn’t work. It does. Documentation was the missing step.
Treating AI output as finished work. Fluent is not accurate. Every property fact gets verified against source data before it is published.
Assigning licensed activity to an unlicensed VA. Showing property, hosting open houses, and prospecting calls are not administrative tasks in Texas, regardless of what the job board listing said.
Buying tools instead of building workflows. Five tools with no destination produce less than one tool wired into your CRM. This is the same failure I see in rooms full of agents who’ve sat through AI training that was really a tool tour.
Comparing hourly cost instead of failure cost. A $600/month VA who prevents one blown deadline a quarter has already paid for the year.
Skipping the written scope. No SOP, no job description, no compliance boundaries — that’s the setup that turns into a TREC complaint or a resignation.

Frequently Asked Questions

Should I hire a VA or use AI first?

Use AI first if your bottleneck is content production — listing copy, captions, emails, drafts. Hire a VA first if your bottleneck is transaction volume, follow-up consistency, or anything with a deadline attached. Track one week of tasks before deciding. Most agents assume they have a content problem and discover they have a coordination problem.

Can a virtual assistant do real estate transaction coordination in Texas?

Yes, within limits. An unlicensed assistant can input data, type contracts as specifically directed by a license holder, handle bookkeeping, and schedule appointments. They cannot review contracts, advise on deal terms, show property, or make prospecting calls. TREC’s guidance is explicit that unlicensed personnel may not direct or advise agents in licensed work. Get the scope in writing before day one.

What tasks should real estate agents never give to AI?

Anything where a silent error is expensive. Final pricing decisions, contract interpretation, fair housing-sensitive language released without review, and any client-facing communication about legal or financial terms. AI is also the wrong tool for persistence — it won’t chase a lender, notice a client going quiet, or take ownership of a deadline.

Is a VA cheaper than an AI stack for real estate?

Per hour, usually not. Per outcome, often yes. Run the comparison against a specific bottleneck rather than against the general idea of help. NAR’s 2026 Member Profile puts median agent business expenses at $9,530 for 2025, so either investment is a meaningful share of your operating budget. Price both against the cost of the problem you’re actually solving.

How do I know if I’m ready to delegate at all?

Write the task down in ten steps or fewer. If you can, you’re ready to delegate it. If you can’t — if the steps change depending on the deal, or you’ve never done it the same way twice — you’re not delegating yet, you’re outsourcing confusion. Document first. That week of writing is the highest-return work you’ll do all quarter.

Can I use AI to replace a transaction coordinator?

Not currently, and the reason is structural rather than technical. Transaction coordination is mostly persistence across time — following up, escalating, noticing when a party has stopped responding, and being accountable when something slips. AI has no continuity between sessions and no accountability. The productive version is a coordinator using AI to draft communications and summaries faster.

What should I look for when hiring a real estate VA?

Prior real estate experience beats general admin experience, because the vocabulary and deadline pressure are specific. Ask how they’d handle a lender who’s stopped responding three days before closing — you’re testing for escalation instinct. Confirm they understand what unlicensed assistants can and cannot do in your state. Then give them a written SOP on day one.

Bring this to your team or event

Book Emily to speak at your next event:
Email: eterrell@yourcoach.com
Phone: (210) 400-9191
Web: coachemilyterrell.com

Can AI Violate Fair Housing in a Listing Description?

Written by Coach Emily on August 28, 2026. Posted in Uncategorized.

By Emily Terrell — Top Coach and Speaker at Tom Ferry International. Active San Antonio agent closing 70+ transactions a year.

AI can’t violate fair housing law — you can, by publishing what it wrote. Section 3604(c) of the Fair Housing Act bans any statement about a dwelling that indicates a preference based on a protected class, and intent isn’t required. This guide covers what triggers liability, what HUD’s 2026 guidance withdrawal changed, and the prompt guardrail that prevents it.

Key Takeaways

The Fair Housing Act attaches liability to the statement, not the author — “the AI wrote it” has never worked as a defense and never will.
Section 3604(c) requires no discriminatory intent, which means a model with no intent at all can still produce a violation you own.
HUD withdrew its digital advertising guidance in 2026, but the regulation governing listing copy survived untouched — less guidance, identical liability.
AI drifts in a predictable direction: it describes the buyer instead of the property, because that’s what persuasive copy has always done.
The multiplier isn’t one bad line. It’s one bad prompt template running across every listing you take.

What is a fair housing violation in a listing description?

A fair housing violation in a listing description is any wording that signals a preference, limitation, or exclusion based on a protected class. Under 24 CFR 100.75, it’s unlawful to make, print, or publish any notice, statement, or advertisement about the sale or rental of a dwelling that indicates a preference, limitation, or discrimination because of race, color, religion, sex, handicap, familial status, or national origin — and the prohibition covers all written or oral notices or statements by a person engaged in the sale or rental of a dwelling, including flyers, brochures, signs, and any document used in the transaction. Legal Information Institute

Two things about that rule matter more than agents realize.

First, it says nothing about intent. Federal courts read §3604(c) against an “ordinary reader” standard — what the copy conveys to the person reading it, not what the writer meant (Ragin v. New York Times Co., 2d Cir. 1991). A generative model has no intent by definition. That doesn’t help you. It removes the only defense most agents assume they have.

Second, it covers “statements,” not just “advertisements.” Agent remarks. Showing instructions. A text to a cooperating broker. The exposure doesn’t start when the copy hits Zillow.

Why this matters for real estate agents

You’re producing marketing at a volume your review process was never built for. According to NAR’s 2025 Technology Survey, released September 18, 2025, AI adoption reached 68% of agents. Two-thirds of the industry is now generating listing copy faster than any human wrote it, and most of that output goes live without a second read. Coachemilyterrell

The economics make the shortcut tempting. According to NAR’s 2026 Member Profile (June 2026), the typical agent closed nine transaction sides in 2025 with a median gross income of $59,200, while median business expenses climbed to $9,530 from $8,010 the year before. When your margin is compressing, skipping ninety seconds of review per listing feels like efficiency. coachemilyterrell

Here’s what nobody wants to tell you: it’s the cheapest ninety seconds in your business, and it’s the one everyone cuts.

“A discriminatory listing description doesn’t cost you a fine. It costs you a two-year window in which any tester, any buyer, any fair housing organization can walk that copy into federal court — and by then you’ve published it 40 more times because it lives in your template.”
— Emily Terrell, Tom Ferry Coach

Did HUD’s 2026 guidance withdrawal make this safer?

No. It made it quieter, which is worse.

What HUD actually withdrew

On April 6, 2026, HUD’s Office of Fair Housing and Equal Opportunity published a Federal Register notice withdrawing eight guidance documents effective September 17, 2025 — including the April 29, 2024 “Guidance on Application of the Fair Housing Act to the Advertising of Housing, Credit, and Other Real Estate-Related Transactions through Digital Platforms.” That was the document specifically addressing algorithmic and AI-driven housing advertising. It’s gone. HUD stated the withdrawn documents have been removed from active use and should not be relied upon as authoritative. Fordham Law ReviewFordham Law Review

Separately, HUD issued a proposed rule on January 14, 2026 that would repeal its Fair Housing Act disparate impact regulations and leave the development of disparate impact standards entirely to the courts. Compliance Alliance

Read those two headlines together and it sounds like fair housing enforcement is receding. For listing copy, it isn’t.

What survived the withdrawal

The distinction that matters is guidance versus regulation. HUD withdrew guidance — non-binding documents explaining the agency’s interpretation. It did not touch 24 CFR 100.75, the regulation that actually governs what you can say in a listing description. That section is still current, still in force, and still reads exactly as it did.

Disparate impact and §3604(c) are also different animals. Disparate impact is about a neutral policy producing an uneven outcome. Section 3604(c) is about what the words say. Rescinding the first does nothing to the second.

HUD said so directly in the withdrawal notice: any actions that do not comply with the text of the Fair Housing Act continue to be subject to enforcement by the Department. Fordham Law Review

Where the enforcement now comes from

The same notice contains the sentence agents should actually pay attention to. HUD noted that regardless of its own enforcement determinations, the Fair Housing Act allows complainants to file a civil action in federal district court or state court within two years of the alleged discriminatory housing practice, and that nothing in the notice affects parties’ rights to seek redress in court. Fordham Law Review

That’s the agency telling you the private right of action is untouched. Less agency guidance doesn’t mean less risk. It means the next person explaining the rule to you is a plaintiff’s attorney.

Where AI-generated listing copy actually drifts

These models learned property copy from decades of MLS archives written before anyone audited them. So the drift is fluent, confident, and completely predictable. Four patterns produce nearly all of it.

Does it describe the buyer instead of the property?

This is the root cause of everything else. “Ideal for a growing family.” “Great starter home for a young couple.” “Perfect for empty nesters.” Each one names a person, not a house — and family composition and age are protected or protected-adjacent territory. AI defaults here because describing the buyer is what makes copy convert. The model is optimizing for persuasion and hitting a statute on the way.

Does it characterize the neighborhood’s people?

“Safe neighborhood.” “Quiet, established community.” “Up-and-coming area.” Courts and regulators have long treated safety and desirability language as potential proxies for the racial or ethnic composition of an area. You can describe a house. You can state a verifiable fact. You cannot characterize who lives nearby, in either direction.

Does it use proximity as a proxy?

Naming a specific house of worship as a selling point — “steps from St. Anne’s,” “walking distance to the synagogue” — signals religious preference even when you meant it as a landmark. Distance in miles to a named amenity is a fact. A named religious institution offered as a lifestyle benefit is a signal.

Does it describe accessibility as a limitation?

“Not suitable for wheelchairs.” “Stairs make this a poor fit for anyone with mobility issues.” Agents write these thinking they’re being helpful. They’re stating a limitation based on handicap. Describe the feature — “two-story with no first-floor bedroom” — and let the buyer draw the conclusion.

The prompt guardrail that prevents most of it

Compliance belongs in the prompt, not in your memory at 9 p.m. Add this block to the end of every listing-copy prompt you save:

Describe only the property. Use physical features, materials, measurements, systems, and distance in miles to named non-religious landmarks. Do not characterize who the property suits, the composition or character of the neighborhood’s residents, the safety or desirability of the area, or the suitability of the property for any person or household type. Do not reference schools by quality rating, houses of worship, or family composition. Output facts only.

Two notes on using it. It reduces drift; it does not eliminate it, and it is not a substitute for reading the output. And it doesn’t cover state-level advertising requirements — TREC Rule 535.155 requires every advertisement to include the license holder or team name plus the broker’s name at a minimum size, and AI drafts never include it. That’s covered in AI real estate compliance: what actually risks your license.

Common mistakes

Reading the deregulation headlines as permission. HUD withdrew guidance, not the statute. Agents who conclude the rules loosened are about to publish into a two-year private litigation window with less warning than before.

Treating fair housing as an edit-stage catch. If the constraint isn’t in the prompt, you’re auditing every output by hand forever. Put it upstream and the drafts arrive closer to clean.

Assuming a disclaimer fixes it. There’s no footer that cures a listing description signaling a preference. The statement is a violation.

Auditing one listing instead of the template. One flagged description is a mistake. A saved prompt producing the same phrasing across thirty listings is a pattern, and patterns are what cases are built on.

Letting the model name schools by rating or houses of worship by name. Both read as lifestyle detail to an agent and as proxy signals to a regulator.

Skipping the segment in team training. This is the block brokerage leaders actually buy, because it’s their liability in the room — covered in what AI training for real estate agents must include.

This is general information, not legal advice. Fair housing law is enforced federally, by state, and by local ordinance, and requirements vary. Consult your broker, your state real estate commission, and an attorney licensed in your state before setting policy.

Frequently Asked Questions

Can AI violate fair housing laws in a listing description?

An AI tool can’t violate the law — it holds no license and bears no legal duty. You can, by publishing what it produced. The Fair Housing Act prohibits making, printing, or publishing any statement about a dwelling that indicates a preference based on a protected class. Liability attaches to whoever publishes. That’s the licensee and the responsible broker. Legal Information Institute

Does using AI change my fair housing liability?

No. The standard is identical whether you typed the description or a model did. What changes is volume and speed — you’re producing more copy with less review time, which raises the odds a violation gets published and repeated. Using AI doesn’t create a new legal duty; it stresses the review process you already needed.

Did HUD’s 2026 guidance withdrawal make AI listing copy safer?

No. HUD withdrew eight guidance documents effective September 17, 2025, including its 2024 guidance on digital and algorithmic housing advertising. Guidance is non-binding interpretation. The regulation governing listing statements, 24 CFR 100.75, was not withdrawn and remains in force. HUD confirmed that non-compliant actions remain subject to enforcement. Fordham Law ReviewFordham Law Review

Does the disparate impact rollback affect listing descriptions?

Not meaningfully. HUD proposed repealing its disparate impact regulations on January 14, 2026. Disparate impact concerns neutral policies producing uneven outcomes. Listing copy falls under §3604(c), which prohibits discriminatory statements outright. The two operate independently, so a change to one doesn’t relax the other. Compliance Alliance

Do I need discriminatory intent to violate fair housing in a listing?

No, and this is the point agents miss most often. Section 3604(c) prohibits statements that indicate a preference — courts apply an ordinary-reader standard focused on what the words convey, not what the writer meant (Ragin v. New York Times Co., 2d Cir. 1991). A model has no intent whatsoever and can still produce copy that violates the statute.

What specific phrases should I remove from the AI listing copy?

Anything describing the buyer rather than the property: “perfect for families,” “ideal for empty nesters,” “great starter home.” Anything characterizing residents or area safety: “safe neighborhood,” “quiet community.” Named houses of worship offered as amenities. Statements about who a property is or isn’t suitable for. Describe the house; let buyers decide if it fits.

Is my broker liable if AI wrote the listing description?

Yes, alongside you. Broker supervision duties extend to the tools used to conduct licensed activity, and the Fair Housing Act reaches anyone involved in making, printing, or publishing the statement. That’s why an AI policy — approved tools, required review step, documented compliance — belongs at the brokerage level rather than being left to individual agent preference.

Bring this to your team or event

Book Emily to speak at your next event:
Email: eterrell@yourcoach.com
Phone: (210) 400-9191
Web: coachemilyterrell.com

AI Avatar Videos for Real Estate Agents: The Trust Line

Written by Coach Emily on August 24, 2026. Posted in Uncategorized.

By Emily Terrell — Top Coach and Speaker at Tom Ferry International. Licensed since 2016. Closing 70+ deals/year while coaching agents nationwide.

AI avatar videos work for real estate agents on repeatable content — market updates, FAQ answers, and translated versions of videos you already filmed. They fail on anything relational: consultations, testimonials, and negotiation. This guide gives you the trust line, the TREC and FTC rules that apply, and the disclosure language to copy.

Key Takeaways

Avatar the repeatable. Film the relational. That single rule resolves 90 percent of the decisions you’ll face.
An AI avatar delivering a client testimonial isn’t a gray area — it’s a federal violation under the FTC’s consumer reviews rule.
Texas advertising rules don’t care how a video was produced. If it’s designed to attract clients, TREC Rule 535.155 applies and your broker’s name still has to be on it.
The Texas Responsible AI Governance Act does not require agents to disclose AI use. Disclose anyway, because the reputational cost of getting caught is bigger than the legal one.
One disclosure line, placed once in your description, removes almost all of the risk.

What is an AI avatar video?

An AI avatar video is a video where a synthetic version of you — your face, your voice, your cadence — delivers a script you wrote, without you ever turning on a camera. Tools like HeyGen build a digital twin from a short recording and then generate unlimited videos from typed text, including translated versions with matched lip movement. The output looks filmed. It wasn’t.

Why does this matter for real estate agents?

Because the economics are brutal on your side and invisible to the consumer. Filming a monthly market update takes setup, lighting, four takes, and an hour you don’t have. Generating it takes four minutes. According to NAR’s 2026 Member Profile, the typical individual agent closed nine transaction sides in 2025 with a median gross income of $59,200 — that’s a business with no room for an hour of avoidable production time. The same profile puts the typical Realtor at 13 years of experience, which means most of your competition has been doing video the slow way for a decade and is exhausted by it.

So the temptation is obvious: avatar everything, post daily, win.

Here’s the thing nobody wants to tell you. The agent who avatars everything doesn’t get caught by a regulator. They get caught by a past client who watches thirty seconds of a “personal” market update and notices the blink rate is wrong. That client doesn’t file a complaint. They just quietly stop referring to you. You never find out why.

That’s the actual risk. Not a fine. A silent discount applied to every video you post afterward.

Where is the line between an avatar and your own face?

Three questions. Run any video through them before you decide.

Is a specific person’s decision on the other side of this video?

If the answer is yes, film it. A seller deciding whether to reduce price, a buyer deciding whether to write, a past client deciding whether to refer you — those people are buying your judgment about their situation. An avatar delivering judgment about a specific person’s money reads as a shortcut, because it is one.

Would the viewer feel misled if they found out?

Run the disclosure test in your head. If you’d be uncomfortable adding “this video used an AI avatar” to the description, don’t make the video. That discomfort is accurate information, not squeamishness.

Is this content identical for every viewer, every time?

If yes, the avatar is doing exactly what it’s good at. Market stats, definitions, process explanations, and translations don’t change based on who’s watching. Nobody expects a bespoke performance of what an option period is.

Green light — avatar it, disclose once: monthly market updates read from published data, FAQ library answers, process explainers, Spanish or Vietnamese versions of content you already filmed in English, internal team training.

Yellow light — avatar with disclosure plus full advertising compliance: listing promotion videos and paid ad creative. These are advertisements under Texas rules, so the avatar question is the second question, not the first.

Red light — never: client testimonials, consultation follow-up, negotiation updates, condolence or hardship communication, and any video depicting another person without their documented consent.

What do TREC and the FTC actually require?

This is general information, not legal advice. Confirm anything here with your broker and your attorney before you publish.

Testimonials are the bright line. The FTC’s Trade Regulation Rule on the Use of Consumer Reviews and Testimonials took effect on October 21, 2024, and it bans creating or disseminating testimonials that misrepresent themselves as coming from someone who doesn’t exist or who had no actual experience with the business — AI-generated testimonials are named explicitly. The rule’s definition of a consumer testimonial covers depictions of a person’s likeness, not just their words, so an avatar performing a real client’s review is squarely inside it. If a past client loved working with you, film the client. Don’t generate them.

Texas advertising rules apply regardless of production method. TREC Rule 535.155 requires every advertisement to include the name of the license holder placing it and the broker’s name in a readily noticeable location, and the rule’s definition of “advertisement” expressly covers social media, electronic media, and the internet. A synthetic video promoting a listing is still an advertisement. The rule also prohibits materially misleading advertising, which is where an undisclosed avatar starts to look like a problem.

Texas AI law is not your compliance backstop. The Texas Responsible Artificial Intelligence Governance Act took effect January 1, 2026, and its consumer disclosure requirement lands on government agencies and healthcare providers, not real estate licensees. Most agents assume the opposite. TRAIGA won’t make you disclose. Your sphere will.

Consent isn’t optional on the platform side either. HeyGen requires a recorded consent video from the person depicted before a digital twin can be generated, and its terms require documented consent for any likeness a user uploads. That covers you when you avatar yourself. It also means you cannot avatar your broker, your TC, or a client because you have their photo.

The disclosure line to copy

Put this in the video description, once, on any avatar-generated content:

This video was produced using an AI avatar. The script was written by [Agent Name]. [Broker Name], [License #].

One line. Placed in the description, not buried in a comment. That’s the entire lift.

Common mistakes

Generating a testimonial because the client is camera-shy. The client’s discomfort doesn’t transfer the permission. Use their written review as on-screen text with attribution instead.

Disclosing in the comments. A comment isn’t a disclosure. It’s not visible on the platform surface where the video plays, and it disappears under engagement.

Avatar-ing the listing video and forgetting the broker name. Agents get so focused on the AI question that they skip the rule that was already there. TREC Rule 535.155 doesn’t have an AI exception.

Using an avatar for anything time-sensitive about a live transaction. Multiple offers, inspection findings, appraisal gaps — pick up the phone. Video isn’t the right medium and a synthetic one is worse.

Cloning a voice from a podcast appearance. Your own recordings of yourself are fine. Recordings a third party owns are a rights problem before they’re an AI problem.

Treating the avatar as a content strategy. Volume without discoverability is a content graveyard. The system that makes video work is search structure, and I’ve written the full framework for treating YouTube as a search authority system rather than a social channel.

Frequently Asked Questions

Are AI avatar videos legal for real estate agents?

Yes, with conditions. No federal or Texas law bans real estate agents from using AI avatars in marketing. What is prohibited is deception: AI-generated testimonials violate the FTC’s consumer reviews rule, and advertising that omits required broker identification or creates a misleading impression violates TREC Rule 535.155. Production method doesn’t change your existing advertising obligations.

Do I have to disclose that a video used an AI avatar?

Texas law doesn’t require it for real estate licensees — TRAIGA’s disclosure duty applies to government agencies and healthcare providers. Disclose anyway. If a viewer would feel misled discovering it later, non-disclosure creates a misleading-advertising exposure under TREC rules and a much larger trust problem with your sphere. One line in the description handles it.

Can I use an AI avatar for client testimonials?

No. The FTC’s consumer reviews rule, effective October 21, 2024, prohibits creating or disseminating testimonials misrepresenting that they come from a real person with actual experience, and it explicitly covers AI-generated content. The rule’s definition of a testimonial includes depictions of someone’s likeness. Film the actual client or use their written review as attributed on-screen text.

What content should real estate agents actually avatar?

Anything identical for every viewer: monthly market updates read from published data, FAQ answers about process and terminology, listing announcements with proper broker identification, and translated versions of videos you already filmed. Anything where a specific person is deciding something about their own money should be filmed. That’s the line, and it holds up almost every time.

Can I create an AI avatar of my broker or a past client?

Only with their documented consent. HeyGen requires a recorded consent video from the depicted person before generating a digital twin, and its terms require legal rights and explicit consent for any likeness uploaded. Beyond platform policy, generating someone’s likeness without permission raises right-of-publicity and deepfake exposure. Get written consent before, not after.

Does using AI avatars hurt how AI search engines see my content?

No. Tools like ChatGPT and Perplexity read transcripts and structured metadata, and a synthetic video produces a clean transcript the same as a filmed one. What matters for citation is accuracy, structure, and topical depth. The risk with avatars is human trust, not machine visibility. Optimize for both, but don’t confuse the two.

Bring this to your team or event

If you’re evaluating speakers for an upcoming event, here’s how to tell the difference between a stage presence and an operator.

Book Emily to speak at your next event:
Email: eterrell@yourcoach.com
Phone: (210) 400-9191
Web: coachemilyterrell.com

The AI Workflow for Real Estate Agents Doing 20 Deals a Year

Written by Coach Emily on August 24, 2026. Posted in Uncategorized.

By Emily Terrell — Top Coach and Speaker at Tom Ferry International. Licensed since 2016. Closing 70+ deals/year while coaching agents nationwide.

The realistic AI workflow for a real estate agent closing 20 deals a year runs about 90 minutes a week across three tools: Claude for drafting, Follow Up Boss for follow-up, and one video tool. It targets database reactivation and transaction communication, not lead generation. This guide gives the weekly schedule, the per-listing build, and the compliance step.

Key Takeaways

A 20-deal agent doesn’t have a lead problem — they have a transaction-load problem, and most AI advice is written for someone else.
The whole workflow fits in 90 minutes a week across three tools, not seven.
The highest-return block is 20 minutes on Monday reactivating a cold database segment.
Transaction milestone messages get written once with AI, then live as CRM templates — that’s what makes it a system instead of a habit.
Every AI-generated marketing asset needs a Fair Housing and TREC advertising review before it publishes.

What is an AI workflow for real estate agents?

An AI workflow is a fixed, repeatable sequence where AI drafts a specific business asset on a specific trigger, and the output lands somewhere it persists. It’s not a tool subscription and it’s not a prompt list. The trigger, the prompt, and the destination are all defined before you open the app.

Most agents don’t have a workflow. They have a habit of opening ChatGPT when they’re stuck. Those produce very different results.

Why this matters for real estate agents at 20 deals

Twenty sides a year is roughly double the national median. According to NAR’s 2026 Member Profile (June 2026), the typical individual agent reported nine transaction sides in 2025, with a median sales volume of $2.7 million for brokerage specialists. So the 20-deal agent isn’t lead-starved. They’re four to six files deep at any moment, running their own transaction coordination, and losing the database because every available hour goes to whoever is under contract right now.

Point AI at content creation in that situation and you make a busy person busier.

There’s also evidence that tool access alone changes nothing. According to NAR’s 2025 Technology Survey (September 2025), 68% of agents already use AI, but only 17% reported a significantly positive impact on their business while 46% reported no noticeable impact at all. That gap isn’t a tool gap. It’s a workflow gap.

And the spend is real. NAR’s 2026 Member Profile put median total business expenses at $9,530 in 2025, up from $8,010 the year before. Adding a fourth AI subscription to a stack you’re not using is a line item, not a strategy.

“At 20 deals, your AI budget should be zero new tools and 90 minutes of calendar. The agents who get results aren’t the ones with the best stack — they’re the ones whose prompts have a home in the CRM.”
— Emily Terrell, Tom Ferry Coach

The 90-minute weekly AI workflow

Three tools. Four blocks. Everything else is a tab you’ll stop opening by week three.

Monday, 20 minutes: database reactivation

Pull the Follow Up Boss Smart List of every contact with no touch in 60 days or more. Export the names with last-contact notes. Paste that into Claude with a standing prompt that drafts one specific, non-generic message per person — referencing the last real conversation, not “just checking in.”

You edit and send. Twenty minutes.

This is the highest-return block in the workflow and nobody teaches it, because it isn’t impressive on stage. NAR’s 2026 data showed the industry leaning harder on referrals and repeat clients than it has in years. At 20 deals, your database is the asset. AI makes working it survivable.

Per listing, 30 minutes once: the full marketing build

One prompt, one sitting: MLS description, three Instagram captions, the database email, the open house script, the neighborhood one-pager. Not five separate sessions spread across the listing period.

The mechanics of generating listing marketing from structured property data — and where the MLS rules actually sit — are covered in detail in Real Estate MLS Automation: What Agents Can Automate and How to Automate MLS Listing Syndication in 2026. Build it once per listing and stop.

Monthly, 60–90 minutes: batch the content

One session. Twelve pieces. Film in one block, caption with AI, export every ratio, schedule in one pass. The sequencing is the whole game here, and it’s broken down in Real Estate Video Editing: The Batch System That Works.

At 20 deals a year, monthly batching is realistic. Weekly is not, and pretending otherwise is how the whole system collapses in week six.

Per appointment, 10 minutes: consult prep

Comps context, the three objections this specific seller is most likely to raise, and a pricing conversation script tuned to their situation. Ten minutes before you walk in the door. You still bring the judgment — AI just stops you from prepping in the car.

The part that turns a habit into a system

Here’s the thing nobody wants to tell you: generating the same “we’re clear to close” email 20 times a year isn’t using AI. It’s typing faster.

Transaction milestone communication gets written once with AI — the offer-accepted note, the option-period reminder, the appraisal update, the clear-to-close message, the post-closing check-in — and then those drafts live inside your CRM action plans or transaction management tool as triggered templates. Twenty deals a year times roughly eight milestone touches is 160 messages you now never write again.

That’s scalable and repeatable. Everything upstream of it is just drafting.

Don’t automate: the actual client conversation, price reduction discussions, or anything a seller will forward to their attorney.

Common mistakes

Adding tools instead of blocks. Every new subscription is a decision you have to make again every week. Three tools, four calendar blocks.

Generating content nobody asked for. More Instagram posts don’t fix a transaction bottleneck. Fix the bottleneck.

Leaving output in the chat window. If the prompt result doesn’t land in the CRM, a saved prompt library, or a template, it evaporates by Thursday.

Skipping the compliance read. AI-generated property copy defaults straight into Fair Housing problems — “great family neighborhood,” “walkable to churches,” “quiet street perfect for retirees.” Every one of those is a steering risk under the Fair Housing Act. And under TREC Rule 535.155, Texas advertising must include the broker’s name in a readily noticeable location, at least half the size of the largest agent or team contact information — AI output doesn’t get an exemption. A 20-deal agent generating five assets per listing is producing 100 pieces of advertising a year. Build the review into the prompt and into the workflow.

Pasting client financial data into a general-purpose chatbot. Decide what goes in and what never does before you’re in a hurry.

This is general information, not legal advice. Consult your broker or attorney on specific listings and campaigns. Agents outside Texas should confirm the equivalent advertising rule with their own commission.

Frequently Asked Questions

How many AI tools does a real estate agent actually need?

Three. A general-purpose model like Claude for drafting, a CRM with AI-assisted follow-up like Follow Up Boss, and one video tool. Beyond that, each addition costs more in decision fatigue than it returns in time. NAR’s 2025 Technology Survey found 68% of agents already use AI while 46% report no noticeable business impact — the constraint isn’t tool count.

How much time does an AI workflow actually save at 20 deals a year?

Realistically, four to six hours a week once the templates exist, concentrated in listing marketing and transaction communication. The first three weeks save nothing because you’re building prompts and templates. Agents who quit before week four never see the return. Measure hours saved against a baseline you captured before you started, not against how the workflow feels.

Should a 20-deal agent use AI for lead generation?

No. At 20 sides a year you’re producing roughly double the national median, which means the constraint is capacity, not lead volume. Point AI at your existing database and your transaction load first. Reactivating past clients and referral sources returns more per hour than any AI lead-gen tool at this production level.

Is AI-generated listing copy compliant with Fair Housing rules?

Not by default. AI models routinely produce language that describes the likely occupants rather than the property — references to families, schools, churches, or age groups — which creates steering exposure under the Fair Housing Act. Write the constraint into your prompt, then read every draft before it publishes. The license holder is responsible for the advertisement, not the tool.

What’s the single highest-return AI task for a busy agent?

Database reactivation. Twenty minutes on Monday: pull everyone with no contact in 60 days, feed the list and last-contact notes to AI, get one specific message per person, edit and send. It’s unglamorous, it doesn’t demo well on stage, and it produces more closings than any content workflow at this production tier.

How do I keep an AI workflow from falling apart after a month?

Give every prompt a permanent home. A saved prompt library, a CRM template, an email action plan — somewhere the output persists without you remembering it. Workflows that live in a chat window die in three weeks. Workflows wired into the system you already open every morning survive. Pick one workflow, install it, then add the next.

Bring this to your team or event

Book Emily to speak at your next event:
Email: eterrell@yourcoach.com
Phone: (210) 400-9191
Web: coachemilyterrell.com

Can AI Accurately Price a Home? What Agents Must Know

Written by Coach Emily on August 24, 2026. Posted in Uncategorized.

By Emily Terrell — Top Coach and Speaker at Tom Ferry International. Licensed since 2016. Closing 70+ deals/year while coaching agents nationwide.

AI can price a home within roughly 2% when that home is already listed — and roughly 7% when it isn’t. Zillow publishes both numbers, and the gap between them is the whole story. This guide breaks down why the off-market estimate is the one your seller is looking at, and how to run that conversation.

Key Takeaways

Zillow reports a median error rate of 1.9% for on-market homes and 7.0% for off-market homes — and every pre-listing seller is looking at the off-market number.
Accuracy is measured against the most recent estimate before sale, not the estimate your seller saw six months ago.
In Texas, TREC rules bar license holders from using the word “value” for their own analysis and require a verbatim disclaimer on any estimated sale price.
The winning move at the listing table isn’t arguing with the algorithm. It explains it better than the seller can.

What is an automated valuation model?

An automated valuation model, or AVM, is a computer program that estimates a property’s likely sale price from public records, tax data, prior sales, and market trends — without anyone physically inspecting the home. The Zestimate, the Redfin Estimate, and the institutional models lenders use are all AVMs. A Zestimate is not an official appraisal, and Zillow says so directly on its own accuracy page.

What every AVM has in common is pattern recognition. The model looks at what similar properties sold for, then extrapolates. It does not value your listing. It runs a match against history.

Why this matters for real estate agents

Here’s the thing nobody wants to tell you: the accuracy number your seller quotes and the accuracy number that applies to their house are two different numbers.

Zillow reports a median error rate of 1.9% for U.S. homes currently on the market, and 7.0% for homes that are not on the market. (Zillow, “How Accurate Is My Zestimate?”) That second figure is the one that matters, because a seller checking their estimate before they call you is — by definition — off-market.

Run the math on a $600,000 Stone Oak home. A 7% median error means the estimate could reasonably land anywhere from $558,000 to $642,000. That’s an $84,000 spread. If your seller anchors to the top of it, you’re negotiating against a number the algorithm itself isn’t confident in.

And “median” is doing heavy lifting. A 7% median error doesn’t mean estimates are within 7%. It means half are closer and half are further off.

“The seller isn’t wrong to look at the Zestimate. They’re just looking at the least accurate version of it. My job in the first ten minutes is to show them which number they’re holding — and then show them the one that actually applies.” — Emily Terrell, Tom Ferry Coach

How is AVM accuracy actually measured?

This is the part that reframes the entire conversation, and almost no agent knows it.

Accuracy is scored against the most recent estimate, not the original

Appraiser Ryan Lundquist has documented that Zillow computes accuracy by comparing the final sale price to the Zestimate on or before the sale date — not the estimate from before the home was listed. (Sacramento Appraisal Blog)

Why that matters: a home could list at $380,000, drop through a series of reductions, and sell at $350,000 — 8% below the original estimate — while the most recent estimate had already slid to $353,000, letting the reported error come in at 1%. (Sacramento Appraisal Blog)

The estimate moves toward the market once the market gives it something to move toward. That’s not a scandal. It’s a measurement choice — and it’s why the on-market number looks so much better than the off-market number.

What the model can see, and what it can’t

What the AVM can see	What the AVM cannot see
Recorded sale price and date of comps	Seller concessions not captured in the recorded price
Public record square footage and room count	Unpermitted additions, conversions, or renovations
Assessed property tax information	Interior condition, finish quality, deferred maintenance
Geographic proximity to recorded comps	Micro-location factors — views, street noise, privacy
Historical price trend for the area	Market shifts inside the last 30 to 90 days
School district and basic neighborhood data	HOA restrictions or pending special assessments
Listed amenities from public fields	Staging, curb appeal, and buyer emotional response

Every row on the right is a conversation you can have. Every one is a reason your judgment is the layer the algorithm was never built to supply.

Where AVMs break down worst

Thin comp pools are the common thread. Luxury tiers with three or four sales a year, custom homes with no true match, rural and low-turnover markets where the model reaches back eighteen months for a comparable, and markets that moved sharply in the last quarter. In all four cases the model is doing its best math on data that doesn’t support the question.

What are the Texas rules on quoting a home’s value?

If you practice in Texas, this section is not optional reading.

Under TREC rules, a real estate license holder may not perform an appraisal of, or provide an opinion of value for, real property unless licensed or certified as an appraiser. (22 TAC §535.17, via Cornell Law School) You can give an estimated sale price. You cannot call it a value.

When a license holder provides a broker price opinion, comparative market analysis, or estimated worth or sale price, they must also provide a written statement reading: “This represents an estimated sale price for this property. It is not the same as the opinion of value in an appraisal developed by a licensed appraiser under the Uniform Standards of Professional Appraisal Practice.” That statement must be part of the written analysis and reproduced verbatim in at least 12-point font. (22 TAC §535.17, via Cornell Law School)

It reaches your marketing too. TREC’s guidance on Rule 535.155 lists as a potentially misleading advertisement one including the value of a property, unless it is based on a disclosed appraisal readily available on request, or given in compliance with §535.17. (TREC, “TREC’s Advertising Rules — What You Need To Know”) So the “what’s your home worth” graphic you were about to post has a compliance dimension.

This is general information, not legal advice. Confirm your own advertising and disclosure practices with your broker or an attorney.

The five-step AVM conversation

Knowing this is worth nothing if you can’t deploy it under pressure. Here’s the system.

Bring it up first. Don’t wait for the phone to come out. Pull the estimate yourself, early. That reads as confidence.
Name which number they’re holding. On-market versus off-market. This is the single highest-leverage sentence in the appointment.
Pull the comp set. Most platforms show at least partial comps. Walk through which ones don’t fit and why.
Point at two or three specific gaps. The unpermitted casita. The greenbelt lot. The roof was replaced last spring.
Position your analysis as the completing layer. Not a replacement. The part the model was never built to do.

Five steps, under three minutes, same every time. That’s not talent. That’s a system — the same way the rest of your business should run. (More on building repeatable systems around your listing data)

Where AI actually fits

AI tools like Claude and ChatGPT are not valuation tools. They’re research and communication tools. Use them to draft your client-facing AVM explainer, build a side-by-side comparison sheet, or script the five-step conversation in your own words. Do not use them to produce a price. A properly built comparative market analysis anchored in real local data is still the work.

Common mistakes

Getting defensive. The second you sound like you’re attacking Zillow, you’ve lost the room.
Quoting the on-market accuracy figure to a pre-listing seller. It’s the wrong number for their situation and a sharp seller will catch it.
Skipping the comp set. Abstract explanations of methodology don’t move anyone. Their actual comps do.
Using the word “value” in Texas. It’s a rules problem, not a style preference.
Treating this as knowledge instead of a script. Knowing it and delivering it under pressure are different skills.

Frequently Asked Questions

How accurate is the Zestimate?

Zillow reports a median error rate of 1.9% for homes currently on the market and 7.0% for off-market homes. (Zillow) Median means half of estimates fall inside that range and half fall outside. Accuracy also depends heavily on how much public data exists in a given area, so thin or unusual markets perform worse than the national figure suggests.

Why is the off-market number so much worse than the on-market number?

Once a home lists, the model gains a major new input — the listing price and current market activity — and its estimate adjusts toward it. Accuracy is then scored against that adjusted figure rather than the original. Zillow computes accuracy by comparing the final sale price to the estimate on or before the sale date. (Sacramento Appraisal Blog)

Can AI replace a comparative market analysis?

No. AI can draft your explanation, format your comparison, and script your presentation. It cannot inspect a property, assess condition, or read a buyer pool. Current large language models are not valuation engines, and the AVMs that do produce estimates work from public data that misses condition, unpermitted work, and micro-location entirely.

Can a Texas agent tell a client what their home is worth?

Not in those words. A Texas license holder may not provide an opinion of value unless licensed as an appraiser, though they may give an estimated sale price accompanied by a verbatim written disclaimer in at least 12-point font. (22 TAC §535.17) This is general information, not legal advice — confirm your practice with your broker.

What should I say when a seller quotes their Zestimate?

Ask which tool they used, then show them whether they’re looking at the on-market or off-market estimate. Explain that the off-market figure carries a materially higher published error rate, walk them through the specific comps the model chose, and identify two or three property-specific factors it cannot see. Position your analysis as the completing layer, not a correction.

Do AVM estimates differ between platforms?

Yes, often by tens of thousands of dollars on the same property. Different providers use different data sources, comp selection logic, and calibration. Pulling two or three estimates side by side in a listing appointment is one of the most effective demonstrations available, because it shows the seller directly that these are modeled opinions rather than facts.

Bring this to your team or event

Book Emily to speak at your next event:
Email: eterrell@yourcoach.com
Phone: (210) 400-9191
Web: coachemilyterrell.com