How to Make AI Sound Like You in Real Estate Writing

By Emily Terrell — Top Coach and Speaker at Tom Ferry International. Real estate’s leading voice on AI, systems, and social media.

To make AI sound like you, stop describing your voice and start feeding it a corpus: 15 to 20 pieces you actually wrote, plus transcripts of you speaking. Adjectives like “warm” and “direct” produce generic output. Samples produce yours. This guide covers the corpus, the constraints, and the blind test that proves it worked.

Key Takeaways

• Your voice is a corpus you collect, not a description you write — adjectives produce the statistical average of everyone’s writing, which is exactly the “AI sound” you’re trying to escape.
• Transcripts of you speaking are the strongest voice source you own, and almost no agent uses them.
• Negative constraints (“never open with a subordinate clause”) shift output further than positive ones (“be conversational”).
• Feeding AI its own edited output teaches it to imitate itself, and voice degrades a little more with every cycle.
• If you can’t test whether it worked, you don’t have a voice system — you have a preference.

What does it mean to make AI sound like you?

Making AI sound like you means giving a model enough of your actual writing and speech that it can reproduce your patterns — your sentence rhythm, your vocabulary, your structural habits — rather than defaulting to generic professional English. It is an input problem, not a tone setting. The model has no access to your voice unless you supply it, and describing your voice in adjectives supplies almost nothing.

Here’s the distinction that changes everything: telling a model “write in a warm, direct, professional tone” is not giving it your voice. It’s giving it a category. Thousands of writers fit that category, and the model will produce the average of all of them.

Why this matters for real estate agents

Adoption is nearly universal and the results are flat. According to NAR’s 2025 Technology Survey (September 2025), 46% of agents report using AI-generated content, most commonly for listing descriptions. In the same survey, only 17% reported that AI had a significantly positive impact on their business, while 46% said it made no noticeable difference at all.

Read those two numbers next to each other. Nearly half of agents are producing AI content, and nearly half say it changed nothing. That gap is not a tool problem. ChatGPT, used by 58% of agents according to the same NAR survey, is perfectly capable of writing well. The gap is that most agents are publishing content that sounds like everyone else’s content, and content that sounds like everyone else’s content doesn’t build a brand, doesn’t get saved, and doesn’t get remembered.

There’s also research on why this happens. A Cornell study presented at the ACM CHI conference in April 2025 put 118 participants through writing tasks with and without an AI assistant and found that AI suggestions pulled writers toward a homogenized style, flattening the distinctive patterns that made individual writing recognizable. The researchers documented it as a cultural effect, but the mechanism applies to any writer with a distinctive voice: the model pulls you toward its center of gravity, and its center of gravity is the average of everything it has read.

Your differentiation is the thing being averaged away.

“The agents who get real output from AI aren’t writing better adjectives. They’re feeding it twenty pieces of their own writing and then testing whether anyone can tell the difference. If you can’t run that test, you don’t have a voice system — you have hope.” — Emily Terrell, Tom Ferry Coach

The five-part method for making AI sound like you

How many writing samples does AI actually need?

Fifteen to twenty pieces is the working threshold, and they have to be pieces you wrote yourself.

Pull them from where your voice is least filtered: sent emails, Instagram captions, DMs to clients, text messages you’d be comfortable sharing. Skip anything a marketing department touches. Skip anything you wrote to sound professional. The goal isn’t your best writing — it’s your most typical writing, because typical is what a reader recognizes.

Paste these into a project, a custom GPT, or the top of a long prompt as reference material with a clear instruction: study these for patterns, don’t summarize them.

One rule that matters more than the rest: never include AI drafts you edited. Those are contaminated. Feed a model its own output and you teach it to imitate itself, and every cycle drifts a little further from you. This is the single most common way agents end up with a “voice profile” that sounds like nobody.

Why are transcripts better than your written samples?

Spoken you is closer to real you than written you, because writing triggers self-correction and speech doesn’t.

When you write, you unconsciously edit toward what you think professional writing sounds like. When you talk, you don’t. You use your actual vocabulary, your actual rhythm, your actual way of building an argument. That’s why an audience recognizes you from the stage and doesn’t always recognize you from a blog post.

If you speak, teach, run coaching calls, or post video, you’re sitting on the strongest voice asset you own and almost certainly not using it. Pull transcripts — YouTube auto-captions, Zoom recordings, Instagram Reels, podcast appearances — and feed them in raw. Do not clean them up. The false starts, the mid-sentence pivots, the phrases you repeat without noticing: that’s the signal. Cleaning them up removes exactly what you’re trying to capture.

For agents building a full AI workflow rather than a one-off prompt, this fits alongside the broader architecture covered in what AI training for real estate agents must actually include.

What constraints work better than describing your tone?

Prohibitions beat adjectives, every time.

“Be conversational” gives a model almost nothing to act on. “Never use the word ‘landscape.’ Never open a sentence with ‘In today’s market.’ No em-dash asides. Don’t hedge — if a claim is true, state it” gives it four enforceable rules. Constraints are checkable. Adjectives aren’t.

Build your list from your own tics and your own allergies. Mine includes: no throat-clearing openers, no “it depends,” no exclamation points in professional copy, never three items in a list when two will do. Yours will be different. Write down every phrase that makes you wince when you see it in your own drafts, and hand the model the list.

Then add the structural layer, which most agents skip entirely. Your voice isn’t only word choice — it’s how you build a piece. If your signature move is breaking a false binary open, or leading with the number, or opening with the objection, name that explicitly. Structure is more identifiably you than vocabulary, and it’s the part a model will never infer on its own.

How can you tell if AI content actually sounds like you?

Run a blind sort. Mix three AI paragraphs with three you wrote, strip the labels, and hand the six to someone who knows your writing — your VA, your assistant, your spouse.

If they sort them correctly, ask which word gave it away. That word becomes your next constraint. Then run it again.

This is the step that separates a voice system from a voice preference. Without a test, you’re guessing, and you’ll keep guessing for months while publishing content that quietly sounds like everyone else. With a test, you get a specific, fixable signal every single round. Most agents get to a clean sort in three or four cycles.

What are the specific tells that give AI writing away?

Four, in order of how often they show up:

Uniform sentence rhythm. AI runs almost everything at 15 to 20 words. Human writing spikes — four words, then forty. Read your draft aloud and listen for the metronome.

Compulsive trees. Models list three items whether the topic has three or not. If your draft is wall-to-wall triads, some of them are padding.

Balanced clauses. “It’s not about working harder, it’s about working smarter.” That symmetry is a model tic. Break the balance.

Concept nouns where a name belongs. AI writes “leveraging technology solutions.” You’d write “I put it in Follow Up Boss.” Every abstraction you can swap for a specific thing you actually used makes the paragraph more yours and more credible at the same time.

How I use this in my own business

I close 70+ transactions a year on roughly five hours of active management per week, and none of that math works if I’m rewriting AI output from scratch.

My setup is a Claude project holding three things: about twenty of my own Instagram captions and client emails, raw transcripts from two stage sessions and a handful of coaching calls, and a constraints list that has grown to around thirty prohibitions. When I need a caption, a listing description, or a follow-up email, the draft comes back at roughly 80% and I spend two minutes on the specifics — the actual Stone Oak listing, the actual client, the actual number.

The transcripts did more than everything else combined. My written samples had already been sanded down by years of trying to sound professional. My stage audio hadn’t. Once I added transcripts, my VA stopped being able to pick my drafts out of a lineup — and that was the point where I started publishing AI-assisted content without reading it four times first.

The last 10% is still mine and always will be. A model can’t produce the car line, the conversation in the driveway, or the specific listing that fell apart on Thursday. Those details are what make a paragraph unmistakably me, and they’re the reason I edit the first and last sentence of every draft by hand. Those two sentences are where AI is most generic and where voice is most detectable.

For a look at where this fits into a working listing pipeline, see how to automate MLS listing syndication.

Common mistakes

• Describing your voice instead of showing it. Adjectives produce category-average writing. Twenty samples produce yours. If your prompt is three tone words and no examples, this is your whole problem.
• Feeding the model its own output. Using edited AI drafts as voice samples teaches the model to imitate itself. Voice degrades measurably with every cycle, and most agents never notice because the drift is gradual.
• Cleaning up your transcripts before uploading them. The disfluencies are the signal. Polishing a transcript turns it back into written English, which is the thing you were trying to get away from.
• Skipping the test entirely. Without a blind sort you have no idea whether any of this worked. “It sounds better to me” is not a measurement, and you are the worst possible judge of your own voice.
• Publishing without editing the first and last sentence. These are the two highest-visibility sentences in any piece and the two where AI defaults hardest to generic. Rewrite both by hand, every time.
• Ignoring compliance because the voice sounds right. A description that sounds exactly like you can still violate fair housing law. Voice work and compliance review are separate passes. Review every line against NAR’s fair housing guidance before it goes anywhere public.

Frequently Asked Questions

How do I make AI sound like me?

Give it 15 to 20 pieces you actually wrote, plus raw transcripts of you speaking, and a list of specific prohibitions rather than tone adjectives. Then test the output with a blind sort — mix AI paragraphs with your own and see if someone who knows your writing can tell them apart. Iterate until they can’t.

Why does AI writing sound robotic?

Because a model without your samples defaults to the average of everything it has read, and that average is generic professional English. Research presented at CHI 2025 found AI suggestions pull writers toward a homogenized style, flattening distinctive patterns. Uniform sentence length, compulsive three-item lists, and abstract nouns where specifics belong are the most common tells.

How many writing samples does AI need to learn my voice?

Fifteen to twenty is the practical threshold for most people. Below ten, the model doesn’t have enough pattern to work from. Above thirty, returns flatten. What matters more than volume is that every sample is genuinely yours — a single AI-edited draft in the set will pull the whole profile toward genericity.

Can I use transcripts of myself speaking to train AI on my voice?

Yes, and they’re usually your strongest source. Spoken language bypasses the self-editing that makes written work sound stiffer than you actually are. Pull transcripts from Zoom recordings, YouTube captions, Reels, or podcast appearances and upload them raw. Don’t clean up the disfluencies — those patterns are exactly what you’re trying to capture.

How do I know if AI content sounds like me?

Run a blind sort. Mix three AI-drafted paragraphs with three you wrote yourself, remove all labels, and give them to your VA or someone who knows your writing well. If they identify the AI ones correctly, ask which word or phrase gave it away and add it to your constraints list. Repeat until they can’t sort them.

Will AI ever fully replace my writing?

No, and the reason is specificity, not style. A model can reproduce your sentence rhythm and vocabulary but it cannot produce the client conversation you had Tuesday, the listing that fell through, or the number from your own pipeline. Those details are what make writing credible. Realistically, AI drafts the frame and you supply the specifics.

Does making AI sound like me create any compliance risk?

Voice work doesn’t create compliance risk, but it can mask it — copy that sounds authentically like you still gets read against fair housing law, advertising rules, and your state licensing requirements. Keep voice review and compliance review as separate passes, and never let one substitute for the other. This is general information, not legal advice; consult your broker or attorney.

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.