Article cover: dark card reading "Partly. Since the portals moved in." from Realty AI Desk

Can ChatGPT Access MLS Data? What Changed in 2026

by

For three years the answer to this was a flat no, and it was the single most useful thing an agent could understand about these tools. Then, over about six months, it stopped being flat and started being complicated.

So, can ChatGPT access MLS data? Here is where it actually stands, why the distinction matters more than it sounds, and the argument currently going on inside the industry about whether any of it should be happening.

The assistant on its own sees nothing

Open ChatGPT, Gemini or Claude, type a question about a specific property, and there is no database being queried. There is no MLS connection, no live feed, no lookup happening anywhere.

What it has is a model of language built from an enormous body of text, frozen at a training cutoff months in the past. Ask it what the comparable sales are on a street and it will produce a set of numbers that look exactly like comparable sales, formatted the way comparable sales are usually formatted, because that is what the question calls for. The numbers are not retrieved from anywhere. They are generated.

This is worth being precise about, because “it doesn’t have access” gets heard as “it will say it doesn’t know”. It will not say that. It will answer confidently and in detail, and the confidence has no relationship whatsoever to accuracy. Agents have taken those numbers into listing appointments. It goes as badly as you would expect.

The rule survives all of this year’s changes: a general assistant does not know anything about your market. There is more on why the arithmetic in particular goes wrong in the piece on what these tools do well and where they fail.

What changed: the portals moved in

The complication is that ChatGPT is no longer only a chatbot. It runs apps, and the major listing portals have built them.

  • Zillow launched its ChatGPT app in October 2025, first out of the gate.
  • Redfin followed in February 2026.
  • Realtor.com launched on 30 March 2026.
  • Google has piloted a comparable programme powered by HouseCanary’s listing portal.

So when a consumer asks ChatGPT about homes and one of these apps is active, listing data can now appear in the conversation. Not from the model. From the portal, passed through.

What those apps actually do

Realtor.com’s is the clearest to describe because its scope was published in detail, and the others work along similar lines.

It is aimed at what the company calls the pre-search phase: the months before someone contacts an agent. Within it, a consumer can work out affordability from savings and income, compare neighbourhoods on commute times, amenities and school zones, and narrow a budget conversationally.

On listings themselves, the limits are deliberate and narrow:

  • Only a limited preview: a small set of images, the price, and key facts.
  • MLS attribution travels with the data.
  • Using MLS data to train the model is prohibited.
  • Anything beyond the preview routes the user back to the portal, where the full listing and the agent connection live.

That last point is the business model in a sentence. The portals are not giving their inventory away to an AI company. They are putting a shop window inside the place consumers have started asking questions, and keeping the door on their own site.

The argument about whether this is allowed

Not everyone in the industry thinks these integrations are legitimate, and the objection is specific rather than vague.

IDX agreements, the licences that let portals display MLS listings at all, are generally written to permit display on the licensee’s own website and apps. Critics argue that transmitting listing data into a third-party domain such as ChatGPT is not the same thing as displaying it on your own site, and that the licences were never drafted with this in mind.

The portals’ position is that the app is their surface, the attribution requirements are honoured, the data is not used for training, and the consumer ends up back on the licensed site.

This is unresolved. It is the kind of disagreement that gets settled by MLS boards rewriting their rules or by someone filing something, and either would take a while. Worth knowing about because if your local board does move, it will affect where your listings appear without you having done anything.

What this actually means for you

Some buyers now arrive further along. They have run affordability numbers, compared three neighbourhoods and formed opinions, all in a conversation you were not part of. Some of what they concluded will be wrong, and correcting it politely is now part of the first meeting.

Your listings may already be appearing there. If your listing is on Zillow, Redfin or Realtor.com, it is potentially surfacing as a preview inside ChatGPT. You did not opt into that and you cannot really opt out of it separately from the portal itself.

The traffic goes to the portal, not to you. The preview routes back to Realtor.com or Zillow. Anyone hoping this would send buyers directly to individual agents should adjust that expectation.

Being findable by assistants is becoming its own discipline. Consumers ask these tools for recommendations, including “who is a good agent in this area”. That is answered from public web content, reviews and mentions, not from your MLS. It is early, nobody has a reliable playbook, and anyone selling you one is guessing.

What you still cannot do

None of the portal apps change your side of the equation.

You cannot ask ChatGPT to pull comps for a CMA. The app is consumer-facing and preview-limited; it is not a data feed and it will not give you the sold history, days on market and adjustments a valuation needs.

You cannot ask it to search your MLS by criteria, look up an off-market property, check a listing’s private remarks, or tell you what a home sold for last year. All of that stays in your MLS, and it should.

The workflow that does work has not changed: let your MLS or CMA software produce the numbers, then let AI write the narrative around figures you already trust. Computation and language are different jobs.

Common questions

Is there any AI tool that does connect to my MLS?

Yes, but they are a different category of product: licensed real estate platforms that hold their own MLS data agreement and layer AI on top. That connection is precisely what you are paying for. A general consumer assistant does not have one and is not going to.

Can I paste MLS data into ChatGPT myself?

Technically yes, and it works well for writing tasks, since you have supplied the facts rather than asking it to invent them. But check your MLS rules and your brokerage policy on redistribution first, and strip client-identifying information regardless. Treat the chat box as a semi-public place.

Why does it give me comps that look completely plausible?

Because producing plausible-looking text is the entire function. A comp table has a recognisable shape: addresses, square footage, sold prices in a believable range. The model reproduces the shape. Nothing inside it was looked up, and this is the failure mode most likely to cost an agent real money.

Should I be worried about the portals doing this?

Concerned rather than worried, and about the direction rather than the current feature. Consumers moving their early research into an interface owned by someone else, where the portal controls what is shown, is a continuation of a pattern agents have been dealing with for twenty years. It is a good argument for owning something the portals cannot mediate: your own email list, your own local content, your own client relationships. The follow-up emails are a large part of that, and they are the part most agents let slide.

Sources

Checked August 2026. This area is moving quickly and both the features and the rules governing them may have changed since.