A real estate agent calls and says a prospect just told them something strange. The prospect ran the agent's name through ChatGPT and got an answer that was partly right and partly from two years ago. The old brokerage. The old phone number. In one case, the wrong city.
The first instinct is almost always the same. Open ChatGPT, type a correction, and expect the next person to get the updated answer. That is not how any of these systems work, and it is the reason the wrong information often lingers for months.
The way to change what ChatGPT, Perplexity, Google AI Mode, and Gemini say about you is to change the sources they are reading. The chatbot is a reflection. Arguing with the reflection does not change the thing being reflected.
What an AI assistant is actually doing when someone asks about you
Each major AI assistant does something different in the details, and the general shape is consistent enough to name.
When someone types a question about a real estate agent, the assistant decides whether to answer from what it was trained on or to search the live web. For a question about a specific person in a specific city, most of them search the live web, because the trained model does not have reliably current information about a working agent.
The search pulls back a short list of pages the assistant thinks are relevant. Perplexity's help documentation describes this plainly: the model reads the pages, synthesizes an answer, and shows the citations it used. OpenAI's description of ChatGPT search and Google's own documentation of its AI search features describe a similar pattern on their respective systems.
The important part for an agent trying to fix a wrong answer is this: the pages the assistant picks are usually the same ones that would rank at the top of a Google search for the same query. If your old brokerage bio outranks your current site on your own name in Google, there is a good chance it is also the page the AI assistant reads first.
The first thing to do is find out what the assistants are actually saying
Before trying to fix anything, run the actual queries. The questions that matter are the ones a prospect would ask, not the ones you would use to show off the site.
- Your full name.
- Your full name plus your city or metro area.
- Your full name plus "real estate agent" or "realtor".
- "Best real estate agent in [your city]".
- Specific questions you know you have earned the answer to, like "[your city] luxury real estate agent" or "[your niche] real estate agent [your city]".
Run each of these in ChatGPT, Perplexity, Google AI Mode, and Gemini. Save a screenshot or copy the answer and the sources into a document. Two things to look at in each answer: which facts are wrong, and which sources are cited. Perplexity shows citations inline on every answer. ChatGPT in search mode shows sources below the response. Google AI Mode shows the pages it drew from. Gemini sometimes shows sources and sometimes does not.
A pattern will emerge in the first ten minutes of doing this. The wrong answers almost always trace back to two or three specific pages that keep showing up as sources.
The sources an AI assistant trusts for a real estate agent
For a working real estate agent, the pages AI assistants trust most are, roughly in order:
- The agent's current personal or team website, when it is well structured and has been live for enough time to earn ranking.
- The brokerage's agent profile page for the firm the agent currently works at, because brokerage sites typically have strong domain authority.
- Zillow and Realtor.com agent profiles, because both sites rank heavily on real estate searches and the assistants treat them as authoritative directories.
- Google Business Profile, especially on Google AI Mode and Google AI Overviews.
- LinkedIn, which the assistants use for career and credential information.
- Local news coverage, if any exists.
Older versions of these same sources are often what cause wrong answers. An old brokerage bio from a firm you left. An old Zillow profile pointing at an old phone number. A LinkedIn profile that still lists an outdated title. The pages have not been removed, they are still indexed, and the AI assistants read them as if they were current because nothing about the page itself says they are not.
Cleaning up the sources, in order
The fix is slower than typing a correction into a chat, and it is the only one that lasts.
One, update the sources you directly control. This is Zillow, Realtor.com, LinkedIn, your Google Business Profile, your own website bio page, and any personal profile on sites you have logins to. For each one, confirm your current brokerage, phone number, email, service areas, and any specialties are accurate. This usually takes an afternoon.
Two, ask former brokerages to remove or update old bios. A polite email to the marketing contact at a firm you have left is the right route. Most brokerages have a process for this, though the speed varies. If the firm will not remove the page, ask them to add a note that you are no longer with the firm and to link to your new site, which both prevents the AI assistants from presenting the page as current and passes some of the page's authority to your new site.
Three, strengthen your current site as the primary source. This is where Person schema markup matters. A current bio page with Person schema, with sameAs links to your authoritative profiles (current brokerage, Zillow, Realtor.com, LinkedIn, Google Business Profile), tells search engines and AI assistants explicitly that these all describe the same person, and which page is the primary one. The sameAs property is defined for exactly this purpose: linking the scattered references to one identity.
Four, make sure the site ranks on your name. This sounds obvious. It is also where many agents stop, because the site exists and is live, and they assume ranking is automatic. It is not. A current bio page that ranks below an old brokerage bio on your own name is still losing the AI citation to the old page. The fix is usually a combination of the page being thin or missing the right structured data, both of which are addressable in a rebuild.
Five, wait for the AI systems to reindex. Perplexity and ChatGPT search usually reflect changes within days to a few weeks because they query the live web for most responses. Google AI Mode and AI Overviews track Google's index, so changes show up as Google re-crawls the updated pages. A repeat of the same test questions, a few weeks after the cleanup, usually shows the answers moving in the right direction.
The hard case: wrong information on a page you cannot change
Sometimes the wrong information is in a news article, a blog post by someone else, or a review site that will not cooperate with a correction request. You cannot change the source. You can change what competes with it.
Publish a current, comprehensive page on your own site that covers the same ground as the problematic page, in more detail and with explicit, dated facts. Link to it from your Google Business Profile posts, from your LinkedIn, and from any outreach that mentions your work. Over a few months, the AI assistants should begin pulling from the newer, more comprehensive page because it is more directly responsive to the query and more current than the old article.
This approach does not delete the wrong page. It changes which page ranks above it, and therefore which page the AI assistants read first. For most queries about a working agent, that is enough.
Tracking whether it is working
The same questions that found the problem are the ones that confirm the fix. Run them once a month in each of the main AI assistants. Note whether the facts in each answer are correct, whether the sources cited have shifted toward pages you control, and which old pages still keep showing up.
A good month looks like this: ChatGPT now cites your current brokerage, Perplexity pulls your current bio page, Google AI Mode uses your Google Business Profile, and Gemini is close to correct with one lingering wrong detail traceable to a specific old page. That last detail becomes the next task on the list.
This is the same tracking approach we cover in the AI visibility gap post, applied to a specific reputation question rather than a general visibility question.
The takeaway
An AI assistant that says the wrong thing about you is not expressing an opinion. It is reflecting the sources it can find, weighted roughly the way Google would weight them. Correcting the chatbot does nothing for the next person who asks. Correcting the sources changes the answer.
The work is not glamorous: updating directory profiles, asking old brokerages to update pages, and making sure the current site on your current name has the right Person schema and ranks on queries about you. A website rebuild handles the parts that happen on your own site, including the schema and the bio page structure, and the free preview can show how it would look before you commit. The agent bio and AI search post covers the bio page in more depth, and the contact page is a good place to start if the specific wrong answer is one you want help tracing.
