How do real estate agents use AI?

A lot of a real estate agent's day is writing: listing descriptions, emails to buyers and sellers, social media posts, neighborhood overviews and follow-ups. Much of it follows a pattern, which is exactly what chatbots such as ChatGPT, Claude or Gemini handle well. Less than two months after ChatGPT's launch, agents told CNN they were drafting listings in seconds from a few keywords, rewriting tense client emails, and running quick mortgage-payment calculations from the car.

The same agents described the output as a starting point. One realtor said he edited the generated listing before it went online; another sent AI-drafted contract addendums to lawyers for approval. That is the right model for the whole profession: AI writes the draft, a licensed professional decides what is true and what gets published.

Three tasks where AI helps most

1. Listing descriptions. Turning a fact sheet into readable copy, in several lengths for different sites.
2. Market summaries for clients. Explaining numbers you already have, such as recent sales and days on market, in plain language.
3. Routine messages. Showing confirmations, document checklists, follow-up notes after an open house.

A Fordham real estate professor quoted by CNN added a useful caution: drafting listings saves time, but fully automating replies to clients may cost more than it saves, because deals depend on personal relationships.

A worked listing prompt

The key is to supply the facts yourself and forbid the model from adding any. A chatbot fills gaps with plausible-sounding details, and in a listing a plausible detail can be a false one.

You are helping a licensed real estate broker write a listing description.
Use ONLY the facts below. Do not add features, measurements, distances,
school names or neighborhood claims that are not listed. If something
would help but is missing, list it as a question at the end instead.

Facts:
- 3-bedroom bungalow, 1 bathroom, finished basement
- Built 1978; roof replaced 2021 (seller's invoice on file)
- Lot about 6,000 sq ft (to confirm with certificate of location)
- Heat pump installed 2022
- 10-minute walk to the village grocery store
- Asking price: $389,000 (illustrative)

Write: a 120-word description for the main listing site, then a
40-word version for social media. Describe the property, not the kind
of buyer it suits. Plain, warm tone, no superlatives like "stunning."

Two lines in that prompt do the compliance work. "Use only the facts below" limits invention. "Describe the property, not the kind of buyer" guards against phrases such as "perfect for young families" or "ideal for a retired couple," which can signal a preference based on family status or age. In the United States, the Fair Housing Act bans discrimination based on race, color, national origin, religion, sex, disability and familial status, and on May 2, 2024 the Department of Housing and Urban Development (HUD) issued guidance confirming the law applies to housing ads even when AI or algorithms write or target them. Canadian provinces have their own rules: in Québec, for example, brokers' advertising falls under the OACIQ, the province's real estate brokerage regulator. Check with your own regulator or association for what applies to you; this article is not legal advice.

A worked market-summary prompt

Here are last quarter's sales figures for single-family homes in our
town, from our board's statistics report: [paste table].
Explain them for a first-time seller in 150 words: the median price
change, typical days on market, and what that means for pricing.
Use only these numbers. Do not predict future prices.

The model is good at explaining a table; it is not a source of market data. If you ask it "what are prices doing in my town?" without pasting figures, it may invent them or rely on old training data.

Keep client information private

Before pasting anything into a chatbot, remove names, phone numbers, financial details and anything a client told you in confidence. Consumer chat tools may use conversations to improve their models unless you change the setting; in ChatGPT, for example, OpenAI added a control in April 2023 that stops new conversations from being used for training. Business versions of these tools generally offer stronger data terms, which is worth checking with your brokerage before adopting one.

What claims in a listing must a person verify?

Every factual claim. In practice, the riskiest ones fall into a few groups, and each needs a document or a direct check rather than the model's word:

Then read the whole text once more for anything the model added on its own: "quiet street," "recently renovated," "walking distance to schools." If you cannot point to the document or observation behind a phrase, delete it. Finally, read for who the ad is aimed at. A description should help any qualified buyer picture the home, without suggesting who should or should not live there.