What should an AI prompt for a real estate letter always include?

Five things — who you are, who the reader is, why you're writing now, what you want them to do, and hard constraints (a word cap around 250 for a single page, plain Australian English, no marketing clichés, and an instruction to use only the facts you supply). Prompts missing the constraints produce the generic, over-polished drafts agents complain about.

How do I stop ChatGPT inventing suburb statistics in my letters?

Tell it explicitly to use only the facts you provide and to leave a placeholder like [recent sale] where a local detail belongs. Language models will otherwise fabricate plausible-sounding medians and "strong demand" claims. Then check every remaining claim yourself — the homeowner reading it knows the street better than the model does.

Should I just use the same AI prompt for every campaign?

Keep the same scaffold — identity, reader, reason, ask, constraints — but change the situation and goal each time. A just-sold letter to neighbours, an expired-listing approach and a cold farming intro have different jobs, and one generic prompt flattens them into the same beige letter.

AI Prompts for Real Estate Letters That Don't Sound Like AI

The difference between an AI draft you can use and one you bin isn't the model — it's the brief. Here's the prompt scaffold that produces usable agent letters, plus ready-to-paste prompts for the five letters you send most.

Most agents' first AI prompts for real estate letters are one line: "Write a prospecting letter for a real estate agent." And the draft that comes back is exactly as generic as the request — polite, polished, interchangeable with every other agent's letter in the suburb.

The model isn't the problem. The brief is. A language model gives you the average of everything it's read unless you pin it down — so the craft has moved from writing the letter to writing the instructions for the letter. This post gives you the scaffold, then ready-to-adapt prompts for the letters agents actually send. (For whether you should be drafting with AI at all — short answer: yes, for drafts — start with the pillar: can you use ChatGPT for prospecting letters?)

The five-part prompt scaffold

Every good letter prompt answers five questions. Miss one and the model fills the gap with its blandest guess.

Who you are. "I'm a local real estate agent who has sold in [suburb] for eight years" — identity shapes voice.
Who the reader is. Owners 10+ years in their home. Landlords. The neighbours of a house you just sold. The tighter the reader, the sharper the draft.
Why you're writing now. A nearby sale, a market shift, a new listing, the season. Letters without a "why now" read as spam because they are.
What you want them to do. One action — book an appraisal, scan the link, reply for the report. Give the model your actual offer or it will invent "don't hesitate to reach out".
Hard constraints. This is the part almost everyone skips, and it's where the AI flavour gets stripped out:

Under 250 words. Plain, warm Australian English, written the way a person speaks at a front door. No marketing clichés — never "I hope this letter finds you well", "dynamic market", "don't hesitate". Use ONLY the facts I've given you; where a local detail belongs, leave a placeholder like [recent sale]. Sign off as [name], [agency].

The 250-word cap matters more than it looks: an A4 page holds roughly 300 handwritten words, and a letter that has to be shrunk or spilled onto page two loses the personal feel that makes it work. Shorter also reads more human.

Five prompts to adapt

Paste, then swap the brackets. Each already carries the constraint block above — keep it.

1. Cold farming introduction

I'm a real estate agent introducing myself to homeowners in [suburb], where I've just [reason — opened, sold, listed]. Audience: owners who've been in their homes 5+ years and get plenty of agent mail. Goal: familiarity, not a hard sell — I want them to keep the letter and remember the name. Offer: a free, no-obligation [suburb] price guide for anyone who [response action]. Warm, modest, zero pressure. [Constraints block.]

2. Just-sold letter to the neighbours

I've just sold [address/street descriptor] and I'm writing to nearby homeowners. Audience: neighbours who watched the campaign happen. Goal: turn curiosity about the result into appraisal conversations. Facts you may use: [result/detail you're allowed to share]. Offer: a no-obligation chat about what the sale means for their own home's value. Confident but not boastful. [Constraints block.]

3. Expired / withdrawn listing approach

I'm writing to a homeowner whose property was listed with another agency and didn't sell. Audience: someone frustrated, possibly embarrassed, definitely over agent promises. Goal: a low-pressure conversation about what I'd do differently. Tone: empathetic first, credible second — acknowledge how draining a failed campaign is before offering anything. No criticism of the other agent. [Constraints block.]

4. Downsizer letter

I'm writing to long-term owners of larger family homes in [suburb]. Audience: empty-nesters who may be years from selling. Goal: plant the idea gently and be the obvious call when they're ready. Offer: [equity update / downsizing guide]. Unhurried, respectful of the emotional weight of leaving a family home. [Constraints block.]

5. Past-client reactivation

I'm writing to a past client I [sold for / helped buy] in [year]. Goal: re-open the relationship, not pitch. Offer: an updated market appraisal given how [suburb] has moved since. Warm and personal — we know each other. [Constraints block.]

These pair naturally with a proven skeleton — if you'd rather start from a known-good structure and have AI redraft it in your situation, the template library and farming letter examples are built for exactly that.

The prompt is half the job

Even a perfect brief produces a draft, not a letter. The model still can't know the sale that just happened on the street, the thing you noticed at Saturday's open home, or how you actually talk. That ten-minute humanising pass — check the facts, add the sentence only you could write, kill the remaining tells, rebuild the close — is its own craft, covered in how to make AI writing sound human.

And remember what no prompt can fix: a brilliant letter, laser-printed into a windowed envelope, still gets binned in the two-second mail sort. Draft with AI, make it yours — then send it in real ink. (If you're drafting inside Scribbly, the writing assistance in the order form already works this way — it drafts to your exact paper size, keeps your response link in, and signs off as you.)

Next: the editing pass that strips the last of the AI flavour — make AI writing sound human.