ChatGPT now shows companies how often it links to their site. Who that leaves worse off
The news has already landed: OpenAI launched a dashboard where a company can see, from ChatGPT itself, how often it links to the company's site. It feels as if visibility in an AI answer will now be provided from the source. And it is not clear yet what this does to AI visibility monitoring tools: whether it gets easier for them, or whether they have to compete with a number from ChatGPT itself? But they do not have to compete, because the dashboard counts whether ChatGPT linked to the site, and a visibility tool counts whether the company was named in the answer. What gets worse is calling your own sample of prompts official ChatGPT statistics.
On 10 November 2026 the dashboard came out in beta under the name Business Hub. A company can create a profile there and set up a card: a description, a category, a website, and so on. If it sells products, it can hand over a feed with prices and availability. A qualifying online store or online service in the US can also launch an agent that answers customers about products and policies. The dashboard has a 28-day summary: mentions, link clicks, and broad topics (the screen also shows a shorter period and profile views).
The part that looks frightening for visibility tools is only the summary: as if the official visibility number now comes from the source, and nobody can dispute it. In fact the card, the feed, and the agent from ChatGPT do not compete with visibility tools.
What the Business Hub summary counts as a mention
A mention is a link to the site or a shown place
For an online business, a mention is counted when ChatGPT has referred to the site. A brand name in the answer is not part of that number. For a local place, a mention is when ChatGPT shows that place in the answer, even if the person never opens the card and never visits the site. The OpenAI help page says this directly: an online business is counted by a link, a local place by being shown. That help page opens only after a login from many networks. The open source for the launch is the OpenAI announcement.
The dashboard has no prompt and no competitors
Business Hub does not see individual questions or chats. It has no answers from other AI models. A blank mention count does not prove there were none: there may not have been enough conversations to pass the threshold, the data may still be collecting, or the report may not be available yet. Claiming the profile and filling it in does not promise that ChatGPT will recommend the company more often. Mentions in AI answers are still a body of work, not one profile.
What AI visibility tools count
A visibility tool shows the answer, the prompt, and the competitors
A visibility tool shows the answer itself, the wording of the question, who else was nearby, and what other models say. The OpenAI dashboard does not. For an online business it checks whether ChatGPT linked to the site. For a local place, whether it showed that place. Good tools still have something to compete on: who is more accurate, easier to use, and more honest about what they counted. What they lack is the right to call their sample of prompts official ChatGPT statistics.
Study: what 8 monitoring tools themselves call a mention
6 of 8 show the name and a link or source separately, 2 have no separate link metric, 0 count only a link or a shown place
On 16 November 2026 we opened the first page of Google for two queries: "AI visibility tool" and "AI visibility tracker". Ads, videos, Reddit, and "best tools" roundups were left out. Eight product sites remained.
None of these 8 reduces a mention to a link or a shown place, the way the OpenAI dashboard does. Six show the brand name in the answer and a link or source as separate figures: Semrush, Rankscale, Ahrefs, HubSpot, SE Ranking, and Track My Visibility. Two write about brand visibility or how models see the brand, and that page does not publish a separate link or source metric (SEO Review Tools and Amplitude). This is not a measurement of ChatGPT's answers and not official statistics. It is what the monitoring tools write themselves.
Comparison: what the ChatGPT dashboard shows and what visibility tools show
Take an abstract private kindergarten. Over a month, people asked ChatGPT the same question 12 times: "a good private kindergarten". In 4 answers the kindergarten was not named. In 8 it was. Of those 8, ChatGPT showed this kindergarten as a place in 5, and in 3 there was only the name. The dashboard will see 5 of 12: for a local place it counts a shown place, not a link to the site. The 3 answers with only a name do not exist for it. A visibility tool will see 8 of 12.
| What it shows | ChatGPT dashboard | Visibility tool |
|---|---|---|
| How often the kindergarten made the answers | 5 of 12 — the answers where the kindergarten was shown as a place. The 3 answers with only a name are not recorded | 8 of 12 — every answer where the kindergarten was named |
| What question was asked | It does not write the question. Instead of the question it may show a topic, such as "kindergarten nearby" | It writes the question in full: "a good private kindergarten" |
| Competitors in the same answer | It does not write them | It writes them: 2 other kindergartens |
| Other AI models | None. The dashboard only watches ChatGPT | You can ask other models too |
A shown place and a name are not the same thing
For an online store the dashboard counts a link, so the gap with a visibility tool will be smaller. The question, the competitors, and other models are still missing from the dashboard.
For a local place, the dashboard counts a shown place. Of the 12 answers, the number includes the 5 where ChatGPT showed this kindergarten. A name without a shown place does not enter the summary. A person could read the name and call the kindergarten, and the dashboard still will not count that answer.
For the recommendation itself, the name counts too. A visibility tool therefore writes 8 of 12: the kindergarten was named, even if the place was not shown. This is not a counting error and not a more honest score. It is an answer to a different question: did the kindergarten make the list a person heard.
If the job is to get a visit from the answer to the site, a link is closer to the point. If the job is to be named instead of the kindergarten next door, a name without a link did the job too. There is no familiar position here: the answer does not hand out place No. 1 or place No. 5. There are two facts — it was named, and it was shown.
Where to monitor mentions in AI answers
AI visibility tools, for example Ahrefs, show whether a company is named in ChatGPT, Gemini, and other models, and which competitors appear in the same answer.
What we recommend
Monitoring how a site shows up in AI answers is a good thing, but most companies are not ready for it yet. It is worth buying monitoring to record the starting point. But until your site can give information about itself in a form that AI systems can read, monitoring will often come back empty, and this is not a lack of demand. It makes sense to take the first measurement and prepare the site at the same time. About a month after the technical work is done, look at what changed against that measurement.
Prepare the site technically on your side
Preparing the site is not a purchase of monitoring. The site needs clear answers, company details, and markup that an AI system can read. What that takes is already covered in these articles: a readiness checklist for AI answers, schema.org in plain language, what to do yourself and what to hand off, how to get into AI answers, and how to get into ChatGPT answers.
Then check whether the site's code has what a mention needs
We, for example, have free online GEO and AEO audits and an SEO audit. If you want something more thorough, there is a product that gathers what matters most, and the fuller picture, for how ready the site is. The report gives you 70 checks, a priority score, and explanations that stay on the point, rather than a long dump of errors. You can try 20 days free (not a subscription, and no card to attach) and run a live query to the models at the same time. The report does not promise that you will start being cited. It shows whether the code has the parameters an AI system can use to read the site.
After the technical work — look at what changed
That is when you can compare with the first measurement and start competing: whether the company is named and whether a link is included. Before the site is ready, the same measurement is there so you have something to compare with. A zero in it means AI systems still have nothing on the site to read, not that there is no demand.
Check the site's readiness for visibility in AI answers
20 days free: not a subscription, and no card required. The report has 70 checks and a status by priority. You can also edit the report for the client and download the files without the mark.