Measuring mentions for real estate agents: one afternoon, seller and buyer

An afternoon finds wrong facts; only the panel reads change

One afternoon answers one question well: what are the assistants saying about me today, and is it true? It cannot say how often you are named, because the same question put twice can come back with different names, and one run is one draw. That is the job of the monthly panel for agents, with its fixed questions and a record kept month after month.

The afternoon's use comes earlier. A wrong brokerage, an old license number or a neighborhood you gave up, repeated in an answer today, is worth correcting before the panel takes its first reading, because a baseline taken while those pages are wrong measures the wrong pages.

Before the first question, write today's facts on one sheet: your brokerage, your license or registration number as your regulator's public lookup shows it where your jurisdiction keeps one, your team's name if you have one, and the neighborhoods and property types you take clients for. The real estate agents hub lists what every page about you should state; the sheet is your copy of it for the day.

The listing side: four questions a homeowner might type

Start with sellers. Write three questions the way a homeowner in one of your neighborhoods would type them, each naming the neighborhood and the kind of home, and a fourth that names you:

• "agent to sell a [property type] in [neighborhood A]"
• "which agents know the [property type] market in [neighborhood B]"
• "listing agent in [neighborhood C] who can sell before a move in [month]"
• "is [your name] a good agent to list with"

Ask each question in a new conversation on each of the four assistants, so that every answer is to that question alone. Note beside each answer whether web search was on and whether the answer listed its sources, since both limit what the record can tell you later.

The buying side: three buyers and one question about you

Then the buyers you want more of, each tied to a neighborhood and a property type you work, and one question that asks about you by name and brokerage:

• "agent for first time buyers looking at condos in [neighborhood A]"
• "relocating to [city], which agent knows [neighborhood B]"
• "buyer's agent for small multifamily properties in [neighborhood C]"
• "[your name] [brokerage] reviews"

Paste every answer whole into one document, with the assistant, the date and the question above it, and save each source it links. Read afterward, not while asking: an answer skimmed in the chat window loses the half you did not scroll to.

What to check in each answer, and what to hold it against

Open your own pages and profiles beside the answers: your site, your brokerage's roster page for you, each portal agent profile you have, your Business Profile if you keep one, and the regulator's lookup. Then read every answer four times, once for each of the things below, and mark each reading right, wrong, out of date or missing.

Missing is a finding too. An answer that names you for a seller question and attaches no neighborhood gives that seller nothing to act on, and it may mean your pages give an answer nothing about that neighborhood to repeat. A mention with the wrong brokerage beside it sends a caller to the wrong office, so it is marked both named and wrong.

When a portal's agent profile stands where your site should

A portal can enter an answer in three ways, and each is marked differently. The answer can name you and cite your profile on a portal: you are named, and the page cited is the portal's. It can send the reader to a portal instead of naming any agent: no agent is named, and the portal goes into the record with the other names. Or it can name you with a fact that only an old portal profile still states, such as a brokerage you left, which points to that profile as the page to fix.

Our audit counted names, not the pages behind them. In the probe that went with our audit of 9,483 small business websites, one buyer question per business put to Claude with web retrieval on, 99.4% of the names given belonged to businesses and 0.6% to directories or aggregators. That covered all 40 trades, with no real estate cut, and recorded no source citations, so nothing in it says how often a portal profile stands behind an agent's name. Your afternoon shows that for you, on one day.

Read a cited profile only for whether its facts are right today. Being cited says nothing good or bad about a portal, and the afternoon ranks none of them.

Tracing each kind of wrong answer to the page that still states it

A wrong fact in an answer either sits on some page or was produced without one. Where the answer lists sources, open them first. Where it lists none, search the web for the wrong fact itself beside your name, in quotation marks, and see which pages still carry it. An answer that states something no page says is a hallucination; record it as one, and read your own pages once more in case one of them can be read two ways.

Correct your own site first. It is the one page you can change today, and the others can be checked against it. Then the pages you can edit yourself, then the requests in writing, each with the date it was sent. You cannot edit an answer; you can edit pages, and the next reading shows whether the answers moved. The general procedure for a wrong fact is in when AI gets facts about your business wrong.

If your own site never appears among the sources, check what the crawlers may fetch before rewriting anything. OpenAI's crawler documentation says a site that disallows OAI-SearchBot is not shown in ChatGPT's search answers, though it can still appear as a navigational link. The free crawler access check reads your robots.txt for that token and eighteen others. It reads the file only, so a challenge page your host serves to bots is a question for the host.

What to carry into next month

Keep the document: the questions, the answers pasted whole, the four readings, and the list of pages you corrected, with dates. The questions can become lines in your panel, kept word for word, and the afternoon's answers stand as a record of the state before the corrections.

Then leave it. Asking again the next morning shows the run-to-run variation that makes one afternoon a snapshot; it does not show whether a correction reached anything. Read change on the monthly panel, scored the way our methodology sets it out, where a difference that holds across questions and months can be told from noise.

For a first look done by someone else, the free visibility check takes your website and an email address, and a person sends a written result within 48 hours.