AI writes listing remarks and client emails that sound polished and certain, even when the square footage, the school, or the HOA dues are simply made up. Here is why it happens, why it matters, and the habit that catches it before a buyer relies on it.
Educational guide · written for residential agents, team leads, and brokers · tool details current as of September 2026 · not legal advice
Why AI invents property facts
A general-purpose chatbot does not look up the county assessor's record or your MLS listing unless it has been connected to that data and told to use it. On its own, it predicts plausible text. A listing is a highly patterned kind of text: a bedroom count, a bath count, a square footage, a lot size, a school name, an HOA figure. When the model does not have the real number, it produces one that fits the pattern, and it states that number with the same confidence as a real one.
Even when you paste in accurate details, the model can garble them: swapping a lot size in acres for square feet, rounding 1,940 square feet up to "nearly 2,100," turning a half bath into a full bath, or carrying a fact from an earlier listing in the same chat into a new one.
The mental model
Treat AI as a fast copywriter who has never seen the house and never checks a record. Useful for tone and structure. Never a source for a single number, name, or boundary.
What wrong facts look like
The examples below are illustrative, written for this guide, and not drawn from real listings. The dangerous errors are the believable ones:
Listing remarks. "Four bedrooms" when the fourth is an unpermitted basement room; "0.5-acre lot" when the plat shows 0.25; "updated roof" when the seller said only the flashing was repaired.
School assignments. "Zoned for Lincoln Elementary" when the district drew new boundaries last year. Assignments change, and a neighboring school's name is a very plausible guess.
Taxes and HOA. Annual taxes pulled from a prior year or a neighbor's parcel; HOA dues that omit a special assessment.
Zoning and use. "Zoned for an ADU" or "short-term rentals allowed" stated as fact with no check of local ordinances or HOA rules.
CMAs and comps. An AI-written market summary citing a "recent sale at $612,000" that never happened, or real comps with the wrong dates, square footage, or sale prices.
Client emails. A friendly recap telling a buyer the home "has a transferable warranty" or "sits outside the flood zone" when nobody verified either one.
Why it matters
Once a fact is in your marketing or your email, it is your representation, not the software's. Depending on your state and the facts, a material misstatement can support a complaint to your real estate regulator, a misrepresentation or negligence claim, or a dispute that surfaces at closing. "The AI wrote it" is not a defense you want to test.
The REALTOR® Code of Ethics. For NAR members, Article 12 requires REALTORS® to be honest and truthful and to "present a true picture in their advertising, marketing, and other representations." Standard of Practice 12-8 asks members to keep website information current and to correct inaccurate content promptly.
MLS rules. MLSs set their own accuracy, data-entry, and photo requirements, and they vary. Read your MLS rules; do not assume one market's rules apply in another.
State law and license law. Disclosure and misrepresentation standards differ by state. Your managing broker or a real estate attorney can tell you what applies to you.
AI-altered and virtually staged photos
AI photo tools can add furniture, swap a gray sky for blue, erase a power line, or "fix" a cracked driveway. Some of that is routine editing; some of it changes what a buyer believes they will get. Disclosure rules are tightening, and they vary by state and MLS:
California. AB 723 (Chapter 497, Statutes of 2025), effective January 1, 2026, requires licensees who use a digitally altered image in advertising to include a conspicuous statement on or next to the image that it was altered, plus a link, URL, or QR code to the original. Common adjustments such as lighting, color correction, cropping, and exposure are excluded from the definition.
ARMLS (Arizona). Starting May 28, 2026, ARMLS requires a Disclosure of Digitally Altered Images when subscribers upload virtually staged or otherwise altered photos. Basic brightness, contrast, and cropping edits do not trigger it.
If your state or MLS has no specific rule, the safer practice is the same: label altered images clearly, keep the unedited originals, and never use editing to hide a defect. Confirm the current rule with your MLS and managing broker.
The verification workflow
Every fact AI puts in front of you is unverified until you have matched it to a source you trust.
Assume every number is a guess. Beds, baths, square footage, lot size, year built, taxes, dues, and distances are claims to check, not facts.
Match to your own listing data. Compare against the MLS input sheet and the seller's documents, not against the AI's summary of them.
Check public records. Confirm parcel size, tax amounts, and ownership with the county assessor or recorder, and note where records and the listing disagree.
Go to the source for HOA, zoning, and schools. Use the HOA's current documents, the municipality's zoning information, and the school district's own boundary tool. Tell clients to verify school assignments directly.
Rebuild comps from the MLS. Every sale in a CMA should come from an actual MLS or public record you pulled, with the date and price checked.
State sources and hedge honestly. Say where a number came from ("per county records") and avoid presenting estimates as measurements.
If you cannot verify it, cut it. An unverified fact does not go in a listing, a CMA, or a client email.
Keep client confidences out of consumer AI
Agents owe clients duties that often include confidentiality, and the specifics vary by state and agency relationship. Whatever your state's rules, some information should never go into a consumer chatbot:
Client financials, pre-approval amounts, and how high a buyer will really go.
Negotiating position or motivation: a seller's divorce, relocation deadline, or lowest acceptable price.
Lockbox codes, gate or alarm codes, and when a home will be vacant.
Identity documents, account numbers, and anything else you would not put on a postcard.
As of September 2026, the consumer tiers of major chatbots can use your chats to improve their models depending on your settings: ChatGPT has an "Improve the model for everyone" control, Claude's consumer plans have a model-training setting in Privacy Settings, and consumer Gemini has a "Keep Activity" control, with some chats reviewed by humans. Business, enterprise, and API tiers (for example ChatGPT Business and Enterprise, Claude for Work, and the vendors' APIs) are not used for training by default. Settings and terms change, so re-check them. Use your brokerage's approved tools (see tool setup) and its data sensitivity rules, and strip names and addresses when a task does not need them.
The bottom line
Let AI help with tone, structure, and first drafts. Take every fact from the MLS, public records, and source documents. Keep those roles separate and AI becomes a time-saver instead of a liability.
Before you publish or send
Beds, baths, square footage, lot size, and year built match the MLS input, seller documents, and public records.
Tax, HOA, and assessment figures come from current records or HOA documents.
School, zoning, flood, and permitted-use statements were verified at the source, or removed.
Every comp in a CMA was pulled from the MLS or public records and checked.
Altered or virtually staged photos are labeled as your state and MLS require, and originals are saved.
No client financials, negotiating position, or access codes were pasted into a consumer AI tool.
A licensed agent, not the AI, has read the final version and stands behind every fact in it.
Sources & further reading
National Association of REALTORS®, 2026 Code of Ethics & Standards of Practice (Article 12) — nar.realtor
California Legislature, AB-723 Real estate: digitally altered images: disclosure — leginfo.legislature.ca.gov