AI prompts for listing descriptions

Written for real estate agents. A listing that states the facts on the sheet, runs your fair-housing check before it goes live, and does not read like every other listing on the block.

• 01 Ten sections, one per writing job a listing takes — from the facts block to the claims audit before you publish
• 02 30 fill-in prompts, brackets left open for your market, your voice, and your brokerage’s writing rules
• 03 30 follow-up steps you send in the same chat, starting with the read-back that quotes every sentence with a person in it
• 04 30 named failure modes, each paired with the prompt that repairs it
• 05 A 105-page manual, a printable quick-reference page, and a plain-text file every prompt copies cleanly from

The kit starts a step before the description. Section one turns your data sheet and walkthrough notes into a facts block — one fact per line, each labeled with where it came from: the sheet, your tape measure, your own eyes, or nowhere yet. Whatever the model could not find comes back as a list of gaps instead of a plausible number, because a gap you can see is a phone call, and a gap the model closed for you is a figure a buyer will quote back at you.

The description is written from that block and nothing else. The rule the whole kit runs on — describe the property, never the people — sits inside the prompt itself, so it is working before a draft exists rather than after you have read one that sounds fine. Then the first chain step turns the model on its own output: read the draft back one sentence at a time and quote anything with a person in it. In our test runs the in-prompt rule got a full draft down to one stray people-sentence; the read-back caught that one. The other sections cover the rest of the job: three lengths cut from one long version so the short one never learns new facts, the property you have been putting off writing, the pass that strips the phrases that sound like every other listing on your MLS, the rewrite for a listing that has not moved, and the claims audit before anything goes live.

The thirty failure modes are the kit’s spine, because a wrong listing draft rarely looks wrong. The facts block comes back flattened — thirteen lines all wearing the same certainty, your half-remembered furnace year dressed like a measurement. The draft describes the block instead of the house: a welcoming street, neighbors who look out for each other, none of it on your sheet and all of it about people. Square footage drifts upward and a serviced roof turns into a newer one — every slip in the expensive direction. The anti-sameness pass comes back as a thesaurus run — it swaps charming for captivating and calls it fresh, or worse, replaces the adjective with a specific cabinet material your kitchen does not have. Asked to revive a stale listing, the model reaches for urgency — statements about price, timing and your seller’s position that you never handed it.

The one the manual calls most dangerous: the model grades its own work and hands you a verdict — reviewed, nothing to worry about, ready to post. It is not in a position to make that call, and that confident sentence is the one that stops you reading. The kit’s checks are built to flag and quote, never to clear, and the repair for an unasked-for verdict ships alongside the repair for the check that returned a cleaner draft instead of showing you which sentences were the problem.

The free starter card carries ten cross-cutting prompts and the four rules behind them, at principle level. It does not carry this kit’s fair-housing read-back, the follow-up chains, the thirty repairs, or the quick-reference page. That boundary is deliberate and it holds.

Nothing identifying enters a chat window: no street address, no client or seller name, no contact detail, no money, and no word about why they are selling. Every paste bracket in the kit names what to strip before you paste. The fair-housing language pass flags and quotes sentences that describe or imply a person — it never tells you a listing is compliant, and it is not legal advice. Your broker and your MLS rules govern what you publish.

The prompts are model-agnostic by design and work in ChatGPT, Claude and Gemini. The test runs behind this kit were against Claude only, and a dated model-notes page inside the manual says so plainly.

Real Estate Agent Kit 01 · $7 · 105-page manual · on sale now