What do mentions in ChatGPT Business Hub actually count?

OpenAI’s help article for Business Hub defines a mention differently for two kinds of business. For an online business, a mention is a time ChatGPT referenced the business’s website, and the article is explicit that the brand name alone does not count. For a local business, a mention is a time ChatGPT showed a result for that location, even when nobody clicked it. All Insights figures cover the past 28 days and are reported in aggregate.

That definition matters more than it looks. A B2B software company that ChatGPT names in a shortlist, without linking or citing its site, may see no mention at all for that answer. A company whose pricing page is cited as a source in an answer about costs gets a mention even if the answer never recommends it. If you have not claimed the profile yet, start with our guide to claiming your business in ChatGPT, then come back to read the numbers.

Launch reporting from LLM Pulse and Sanbi, both vendors in this space, describes a 7-day and 28-day toggle and a table of the top five discovery topics. OpenAI’s help article describes the past 28 days. Expect the interface to change while the product is in beta.

A popular take after launch said Business Hub proves brand mentions now matter more than links. For local results that is close to true: a mention is a location result being shown. For online businesses, OpenAI counts the opposite of a pure brand mention. The metric only moves when ChatGPT references your website, which means your own pages have to be the source it reaches for.

The gap between being named and being cited is real. SPIRITT, an agency, published its own tracking from November 2 to 7, 2026: across 1,216 AI answers to 58 buyer questions, ChatGPT cited a SPIRITT page in 10.4% of answers but named SPIRITT in the text in only 2.3%. That is one agency’s sample, not a benchmark, but it shows why a Business Hub mention and “ChatGPT recommended us” are different events. We cover how the two signals interact in our piece on brand mentions versus backlinks.

Our view: both matter, for different reasons. Third-party mentions help ChatGPT decide who belongs on a shortlist. Clear, citable pages on your own site decide whether your website is the reference it uses, and that is what this metric counts for online businesses.

Every reporting source we read agrees on the limits. These are the gaps to explain before anyone puts the number in a board deck.

• Prompts: businesses never receive individual prompts or conversations, including chats with a business agent.
• Competitors: there is no competitor set, ranking or share of voice. Search Engine Roundtable quoted Juozas Kaziukėnas, who first reported the tool: it does not show prompt data, user data or any competition benchmark.
• Position and framing: a mention does not say whether you were the first recommendation, a footnote or a counterexample.
• The cited page: the metric does not tell you which URL ChatGPT referenced.
• Other assistants: Insights covers ChatGPT only, not Gemini, Perplexity, Claude, Copilot or Google AI Overviews.
• Causation: a count shows that something happened, not why it rose or fell.

OpenAI says a blank metric can mean activity has not reached the reporting threshold, data is still being collected or reporting is unavailable. It does not publish the thresholds. If a metric stays blank, the help article points you to OpenAI support.

For a mid-market B2B company with a narrow category, blanks will be common at first. Do not tell leadership “we are invisible in ChatGPT” because a tile is empty, and do not tell them “we are winning” because a small number appeared. Hans van Gent, a consultant who documented the launch, recommends copying Insights into your own sheet every week because the 28-day window is short and history is not kept for you. We agree, and we add a column for what changed on the site that week.

This is the setup we use with clients. It keeps OpenAI’s self-reported counts and our sampled measurements apart, so nobody averages two different kinds of evidence.

1. Record a baseline first: before claiming, run your fixed set of buyer questions in ChatGPT and save the answers, so you can tell later what changed.
2. Log Insights weekly: copy mentions, profile views, website clicks and the top discovery topics into a sheet with the date range shown.
3. Keep sampled tracking separate: report share of answer from your own prompt set on its own line, with its sample size.
4. Join clicks to analytics: compare Business Hub website clicks with ChatGPT referral sessions in GA4 and with leads in the CRM.
5. Annotate changes: note profile edits, new pages, PR coverage and review activity next to each week so movements have candidate explanations.

For the sampling side, use our guide to measuring share of answer. For the analytics side, our GA4 guide for AI referrals shows how to separate ChatGPT traffic from other sources.

Read a change in three steps. First, check whether it is real: compare the 28-day figure with your weekly log rather than one screenshot. Second, look at discovery topics: a new theme often explains a rise better than any site change. Third, run your prompt set for the affected theme and see which pages ChatGPT cites now.

If mentions fall for an online business, look at the pages ChatGPT used to cite. Were they moved, merged, blocked or left out of date? If they rise, find the page that is being used and make sure it answers the follow-up questions a buyer asks next. Our generative engine optimization service runs this loop monthly for B2B teams, and we report it as decisions, not just as a chart.

If you want a second pair of eyes on your first month of data, talk to our team.