Counting mentions is not the same as controlling the answer.
On November 6, 2026, screenshots of a "Claim your business in ChatGPT" console started circulating. They show a business claiming its own record inside ChatGPT, watching how often it is mentioned, editing its product details, and answering customer questions through an agent it owns.
The screens carry an unreleased or beta label. OpenAI has announced nothing. No launch date, no price, no country list, and no help page. That gap is the most useful fact in the whole story, and most coverage skips it on the way to the opportunity.
What the reporting actually establishes
Juozas Kaziukenas of Marketplace Pulse published the screens on LinkedIn on November 6, and said he had seen early pieces of this in August. Search Engine Roundtable and Search Engine Land covered it the same day. Three independent outlets, one set of screenshots, no confirmation from the company that would have to ship it.
What the screens reportedly show:
- How often the business is mentioned in ChatGPT answers, charted day by day.
- How many people click through from an answer to the business site.
- Editing of company and product details.
- A business agent that answers customer questions from context the business controls.
- A claim step that requires verified ownership. The example in the screens is connecting a Shopify store.
What they do not show: the prompts that produced a mention, user level data, or a competitor ranking. Those absences are not a detail. They are the whole margin between a scoreboard and a decision.
A count without the question is a number you cannot act on
Two numbers answer how often and how many. Neither answers why.
A mention happens because somebody asked something specific. Which something? A click happens because a particular sentence convinced someone. Which sentence? An answer leans on particular pages. Which pages?
Read the count alone and you know that a business question moved. You do not know which question, which source, or which rival was quoting the page you should have written. The reporter's own framing said it plainly: the prompts are the part marketers need most, and they are the part that is missing.
There is a second gap, larger than the first. Someone asking a business question in one assistant and asking the same question in another assistant can get two different answers, with two different sets of sources. A console inside one vendor can only ever see its own half. The customer does not live inside one vendor. They ask everywhere.
The claim step is an identity record
Look past the dashboard for a moment. Claiming a business is a data entry step, and the record it creates becomes the context an assistant answers from.
Name. Services. Products. Whatever the agent is allowed to say.
Whoever writes that record first defines what gets said about the business. A company that has spent ten years being described three different ways across three systems now has one more system asking which version is true, and it arrives with a verified ownership check attached.
That is why the prep work is mostly boring and mostly clerical. One name. Current products. A site the crawler is allowed to read. One named owner of the description. None of it is a marketing campaign, and all of it is what a claim screen would ask for anyway.
A readiness checklist that needs no tool
Five things worth settling now, because none of them can be fixed from inside a console that does not exist yet.
- One name, used everywhere. Legal name, trading name, and the name on the site should agree. An assistant reading two versions has to pick one, and it will not pick the one you prefer.
- Current product and company details, on your own site first. If you would type a fact into the console, make the same fact true on the page the crawler reads.
- A site that is citable. OpenAI's crawler documentation states that sites blocking OAI-SearchBot are not shown in search answers, except as navigational links. A firewall rule can make a business invisible long before anyone gets to claim a record.
- One owner for the record. One person accountable, in one place. Shared ownership is how a business ends up described three ways.
- A baseline of the answers you get today. Ask the same buyer question in two assistants, save what came back with its sources. That record is the only honest way to tell, later, whether any of this changed anything.
What closes the gap
Measuring appearances and producing a correct, checkable answer are two different layers. The second one has to work wherever the customer asks, not inside one vendor's console.
CorpusIQ connects the assistant to the systems that already hold the answer: the accounting system, the store, the CRM, the analytics, the inbox. Ask a business question in plain English and the answer is assembled from those systems at the moment you ask, with the sources attached so it can be checked.
- Read-only access. External source connectors retrieve without writing back.
- Authorized per tool, through each system's own login screen, revocable on its own.
- Cross assistant by design, so the same question can be asked wherever the customer happens to be.
- Answers carry their sources, which is what makes a figure usable in a decision instead of only in a draft.
That is the layer a count cannot reach. A number tells you something moved. Sources and reproducibility tell you what moved, and whether it is still true tomorrow.
What to watch out for
Three honest limits.
The feature may change before it ships, ship in a narrower form, or never ship. The screens are evidence of a test, not a commitment.
A count can move for reasons that have nothing to do with work you did: a model update, a crawl change, a competitor leaving a category. Treat any movement as a question, not a result.
And a claimed record is not a ranking. Being the business the assistant describes accurately is one thing. Being the business it recommends is another, and the second one is decided by the sources the model chooses to lean on.
See also
FAQ
Is the ChatGPT business claim screen available today?
No. The screens carry an unreleased or beta label, OpenAI has announced nothing, and no help page exists. Everything published about it so far is reporting on a test.
What do the mention and click counts measure?
Per the reporting, one is how often the business appears in answers and the other is how many people click through to its site, both charted by day. Neither reports the questions behind the mentions or the sources the answers used.
Does CorpusIQ store my business data?
Direct MCP does not retain raw customer files or full connector response payloads. Operational logs may persist for up to 30 days. CorpusIQ does not train models on customer data; the selected AI client's plan and settings govern conversation handling.
What can I do before this launches?
Settle one name for the business, keep product details current on your own site, confirm your firewall does not block OAI-SearchBot, name one owner of the record, and save the answers you get today as a baseline.