ChatGPT Business Hub Mentions: What the Number Counts and Misses

OpenAI launched Businesses in ChatGPT in beta on 10 November 2026. One part of it, Business Hub, gives a company OpenAI's own count of how often ChatGPT mentioned it in the past 28 days. If you run a B2B company, that number is worth having. It is also narrower than its name suggests. For an online business, OpenAI counts a mention only when ChatGPT references your website. A response that names your brand and links nothing of yours is not counted.

This post explains what the Insights numbers cover, the four things they leave out, and how to check those four yourself. It is based on OpenAI's announcement and help article as read on 11 November 2026, not on a claimed profile. Where we use our own data, the instrument and the date are named.

The same reading rule applies here as in the 0 percent ChatGPT citation trap: find out what the instrument counts before you believe its number, including when the number is blank.

What Business Hub gives you#

Business Hub is where a company claims its business in ChatGPT, manages its profile, and reads its insights. According to OpenAI's help article, the path looks like this:

1. Open Business Hub from the ChatGPT sidebar and search for your business by name. Eligible online businesses can add a listing if none is found.
2. Verify the claim. The methods OpenAI lists include phone, business email, website domain (a DNS TXT record), and Shopify. OpenAI reviews the claim and emails a decision.
3. Complete identity verification. It is required before you can publish profile changes, connect a product feed, or view insights.
4. Edit the profile: website, description, product or service category, logo, cover photo, delivery policy, return policy.

Three insight metrics follow, each aggregated over the past 28 days.

Two limits are stated by OpenAI directly. Claiming does not guarantee more visibility or give control over what ChatGPT recommends independently. And OpenAI says its support team is unable to delete, unpublish, or hide a published profile as part of a business setup or ownership request, so check the details before you publish. OpenAI also notes that counts may be incomplete.

A mention in Business Hub is a website reference#

Most AI visibility writing, ours included, uses "mention" for the brand being named and "citation" for the website being referenced. Business Hub uses "mention" for the second one. In the vocabulary of what AI citation rate means, the Business Hub mention count is a citation count for one engine.

That matters because the two events come apart. We covered the mechanism in why a brand is mentioned but not linked. The question for Business Hub is how often they come apart, and our data says the answer depends heavily on the engine.

Instrument: 16 questions a buyer or team lead might ask around project management software, each asked 5 times on 3 engines, 240 answers, 0 errors. Run on 3 September 2026. The engines were Claude with web search, an OpenAI model with web search run through the Codex CLI, and the Perplexity Sonar API. We tracked 14 established brands (Jira, Asana, Linear, ClickUp and ten others) and checked, for every answer that named one, whether that brand's own domain appeared among the answer's sources. Docs and support subdomains count as the brand's site.

Five cautions on that table. The OpenAI row is the API-side model with web search, not the ChatGPT app that Business Hub reports on, so treat it as a neighbour of ChatGPT, not a reading of it. The brands are large companies with strong sites, and this data cannot say what happens for a smaller company. The brand-to-domain map is ours (Jira counts as atlassian.com, Trello as trello.com or atlassian.com). Sources are not collected the same way on every row: for Claude and the OpenAI model they are the URLs the model wrote into its answer after being asked to cite its sources, and for Perplexity they are the citation list the API returns. And 2 of the 16 questions name brands in the question itself, which moves the OpenAI row a great deal: without those 2 questions it is 5 of 140.

With those cautions, the direction holds. On the OpenAI model, 21 of 178 namings (11.8%) had no reference to the brand's own site. On the other two engines it was about half or more. Eight Claude answers listed no sources at all, and without them the Claude figure is 90 of 182.

Four things the number does not show#

1. Answers that name you without referencing your site. These are excluded by definition for an online business. A buyer can read your name in a ChatGPT answer and Business Hub will record nothing.

2. Whether a counted mention was a recommendation. A website reference means ChatGPT used your page. It does not mean ChatGPT recommended you. In the 80 Perplexity answers, Perplexity listed a tracked brand's site among its sources without naming that brand in 56 of 123 cases. In a separate single-question check on 8 November 2026, the top organic Google result for the question belonged to a vendor whose blog Perplexity used as a source in 5 of 5 answers, and none of the 5 named that vendor. The 5 answers used the same set of sources, so read that as one query on one day, repeated. Those are Perplexity readings, not ChatGPT readings. They show the pattern is possible: your page can be the source for an answer that recommends someone else.

3. Your competitors. OpenAI's pages describe insights for your own business only. Neither page describes a share figure or a view of who else was referenced for the same topics. A count of 400 means something different if the category leader has 4,000.

4. Every other engine. Business Hub reports on ChatGPT. Claude, Perplexity, Gemini and Google's AI results are outside it, and those are the engines where our naming-without-site gap was largest.

There is also a fifth point that is a reading rule more than a gap. A blank is not a zero. OpenAI states that a blank metric can mean the threshold was not reached, the data is still collecting, or reporting is unavailable. Record a blank as unmeasured.

How to use it this week#

Claim the profile. It is the only count of ChatGPT website references that comes from OpenAI itself, as far as we found, and the profile itself is what appears when a buyer clicks your business in a response. Then put the number in context:

1. Record the baseline on the day insights first appear. Write down mentions, website clicks, the topics shown, and the date. A 28-day rolling number is only useful against its own earlier readings.
2. Compare website clicks with your own analytics. Check the ChatGPT referral sessions your analytics tool records for the same 28 days. A large disagreement is a finding in its own right, as it was when we read a citation spike that sent zero visits.
3. Ask your buyers' questions by hand. Take 10 to 15 questions a buyer would ask before choosing a product like yours. Ask each one several times in ChatGPT and in at least one other engine. For each answer record three things: were you named, was your site referenced, and who was recommended. The manual method is written up in how to track brand visibility in ChatGPT and Perplexity.
4. Read discovery topics as themes. They tell you which subjects your business comes up under. They do not tell you which questions you lost.

Business Hub answers one question well: how often did ChatGPT reference our website, and how often did someone click. The questions a B2B company usually cares about more are the ones around it: were we named, were we the one recommended, and who was chosen in our place.

What to do next#

Website references depend first on whether an AI system can fetch and read your pages. The free AI readiness scan runs 18 checks against a URL and shows what blocks that. It does not measure citations. It checks the conditions for them.

Google has its own version of this first-party reporting, with its own blind spots. Chudi's read of it is on chudi.dev: Search Console's generative AI report and the 3 things it hides.