Can AI write my listings and answer leads without breaking the rules?

A rental listing in New South Wales promised nearby schools. The schools did not exist. An LJ Hooker branch had used ChatGPT to write the listing, the listing went up as written, and the agency corrected it only after Guardian Australia got in touch.

That is the whole story of AI in a brokerage in one paragraph. The tool saved someone twenty minutes, and the agent still owned every word it wrote.

Agents have taken to it anyway. The National Association of REALTORS® found in its 2025 technology survey that 20% of its members use AI daily and another 22% weekly, while 32% had not used it in their business at all. ChatGPT was the most common tool, named by 58% of those who answered.

That last number matters. Nearly half of the agents asked said AI had made no noticeable difference, against 17% who called the effect significant. The survey covers American agents, not Ontario ones, and we found no Canadian survey of the same question. Read it as a sign of where the work is going, not a forecast for your office.

We picked the five jobs below because each one shows up in that survey, in reported cases, or in what Canadian regulators and boards have chosen to write rules about. For each one: what the AI does, what it costs, and where the line sits.

Drafting listing descriptions

Verdict: the safest place to start, if you check every fact.

This is the most common use, and the easiest to see working. You give the tool the room sizes, the upgrades, the neighbourhood and a tone, and it gives you a clean first draft in under a minute. Most agents then cut half of it, because AI copy runs long and leans on the same adjectives every other listing uses.

The risk is the LJ Hooker risk. A chatbot (a program that writes replies to whatever you type) does not know the house. When it is missing a fact, it tends to supply a plausible one: a school, a renovation date, a walkout that is really a window. BC's real estate regulator puts it bluntly in its AI guideline: AI tools "may end up telling you something that is not correct", and licensees "remain accountable for the accuracy of any real estate advertising they publish".

Ontario has no AI-specific rule, but it does not need one. RECO's bulletin on online advertising requires "current, clear, and accurate information" and no "false, misleading, or deceptive information", and makes the brokerage responsible for every agent's ads. The same bulletin says using "verbatim property descriptions created by someone else requires the written consent of the owner". So do not paste a competitor's listing into a chatbot and ask it to reword it.

Cost: a general chatbot subscription, or nothing extra if your CRM or listing software already includes a writing assistant. The real cost is ten minutes of checking per listing.

Editing photos and virtual staging

Verdict: useful, and the one most likely to get you a complaint.

AI photo tools can brighten a dark basement, replace a grey sky, or furnish an empty room. Virtual staging in particular saves the cost and the moving truck of real staging for a vacant property. It is also where the clearest problems have come from.

In Quebec, Real Estate Magazine reported a listing photo of a house in Terrebonne that had been edited with AI to add windows and a fence, enlarge the garage door and remove a door. The brokerage's director apologized and said the changes "exceed what is acceptable". The same article sets out the Toronto Regional Real Estate Board's position: colour correction, brightness and virtual staging are allowed if they do not materially alter the image's factual accuracy or misrepresent a feature, condition or view. TRREB usually gives members a chance to fix a problem photo first; if it happens again, the matter can go to a professional standards hearing.

The working rule is simple. Edit for clarity, never for condition. Furniture that is clearly not part of the sale is fine. A new window, a removed crack or a borrowed view is not. BC's regulator also tells licensees to label altered or enhanced photos. Ontario's rules do not say that in so many words, but a one-line "virtually staged" note costs nothing and answers the complaint before anyone makes it.

Cost: virtual staging services charge per photo, and the AI editors inside photo apps are often included in a subscription you already have. Compare it with what a real staging company would quote for the same vacant house.

Answering leads after hours

Verdict: worth it for busy teams, as long as people know it is a bot.

A buyer finds a listing at 11 p.m. and sends a message. If the reply comes the next morning, they may have already booked a showing with someone else. AI assistants that answer website chat, email or texts at any hour are sold as standalone services and as add-ons to CRMs (customer relationship management software, the database where you keep your leads).

Keep the bot to logistics: confirming the property, collecting a phone number, offering showing times, answering what is already in the listing. Do not let it discuss price, offers, or anything that sounds like advice. BC's guideline makes the transparency point directly: if you use an AI chatbot while you are unavailable, "you should inform your client of the situation". A line at the top of the chat that says "automated assistant" does the job.

Privacy is the other half. Whatever the buyer types goes to the vendor's servers. Under PIPEDA, the federal privacy law that covers most private businesses in Ontario, you stay responsible for that personal information even when a vendor holds it. Ask any vendor where the data is stored and whether it is used to train their models, and get the answer in writing.

Cost: usually a monthly add-on to your CRM or website. Price it against the leads you lose to slow replies now, which you can count from your own inbox timestamps.

Following up with past leads and clients

Verdict: the biggest time saver, and the one with a legal clock on it.

AI can draft a market update for past clients, a check-in for a buyer who went quiet, or a sequence of follow-ups after an open house. That is real time saved for an agent who would otherwise not send them at all.

The catch is Canada's Anti-Spam Legislation, CASL. CREA's guidance to members sets out the clock: you can rely on implied consent from an existing business relationship only if you did business with the person in the past two years, or they made an inquiry about your services in the past six months. Every commercial message also needs identification and an unsubscribe link. An AI tool that sends a cheerful follow-up to a ten-month-old lead who never closed is sending spam on your behalf.

The fix is in the software settings, not the writing. Tag each contact with when and how you got consent, and set the automation to stop at six months unless the person has agreed to stay on your list.

Cost: included in most CRMs and email tools. Setting up the consent dates properly takes an afternoon, once.

Summarizing documents and client files

Verdict: hold off unless the tool is built for it.

It is tempting to drop a condo status certificate, an inspection report or a client's offer history into a chatbot and ask for the three things that matter. The summary may be good. The problem is where the document went.

BC's regulator warns that "you should never assume that the data you input is safeguarded and private", that some AI platforms store what users type to train their systems, and that licensees should get clients' informed consent before putting their information into an AI tool. It also advises avoiding confidential or personal information entirely unless you are certain the tool does not store or use it. That guideline binds BC licensees, not Ontario ones, but the privacy reasoning behind it is the same under PIPEDA.

If your brokerage wants this, use a business plan whose terms say your data is not used for training, keep names and financial details out of what you upload, and read the original document before you tell a client anything.

Cost: business tiers of the main chatbots cost more than personal ones. That difference is the price of the data protection, and it is the part worth paying for.

The five side by side

Notice what is missing. None of the reported uses replace the agent's judgment on price, negotiation or advice, and the regulators have been clear that responsibility does not move to the software. In every case above, AI writes or answers faster. The licence, and the complaint, still have your name on them.

Where to start

1. Pick one job, most likely listing descriptions, and write a one-page brokerage rule for it: which tool, what goes in, and who checks every fact before it posts.
2. Open your CRM this week and check that each contact has a consent date. Turn off any automated sequence that ignores it.
3. Put "virtually staged" on every staged photo, starting with your current listings.

If you want a second pair of eyes on how AI fits your brokerage's workflow, you can book a call with Trimotech.