In this post 9 sections
OpenAI's Business Hub, launched in beta on 10 November 2026, gives verified businesses first-party counts of mentions and website clicks in ChatGPT. Together with Bing's AI Performance report and Google's Search Console generative AI report, owners now have platform-side numbers from three major surfaces. None of them reports share of voice, and third-party prompt-panel versions of that metric move a lot between runs. Current as of 13 November 2026.
Key Takeaways
- Business Hub shows aggregated mentions, website clicks and discovery topics, only when activity meets thresholds OpenAI has not quantified. OpenAI says clicks do not confirm website visits or conversions (Tomorrow's Publisher, PPC Land).
- "Mention" is undefined in the coverage I read. PPC Land notes OpenAI does not say whether repeat appearances in one conversation count more than once.
- Google's Search Console generative AI report counts link impressions only, and had a logging error from 9 to 11 November 2026 that lowered impressions (Search Engine Roundtable).
- Bing's AI Performance report is the most detailed of the three: citations, grounding queries, Citation Share, Compare and, since 13 October 2026, a Topics Portfolio beta. Microsoft calls its figures directional.
- Third-party share-of-voice numbers rest on prompt panels, and repeat runs disagree: Muck Rack's same-day test shows ChatGPT repeating 41% of cited domains, Gemini 45%, Perplexity 94%.
- My view: report AI visibility as ranges from repeated runs, triangulated against first-party reports, and label it an estimate.
What launched, side by side
| Source | Metrics | Launch or update | What it does not tell you |
|---|---|---|---|
| OpenAI Business Hub | Mentions, website clicks, discovery topics; windows reported as 28-day (Tomorrow's Publisher) or 7- and 28-day (PPC Land) | Beta, 10 Nov 2026 | Mention definition, thresholds, visits or conversions, individual prompts |
| Bing Webmaster Tools AI Performance | Citations, grounding queries, Citation Share, Compare, Topics Portfolio | Public preview 10 Feb 2026; extra views 16 Jun; Topics Portfolio 13 Oct | Clicks, competitor domains, causes of change |
| Search Console generative AI report | Impressions of links in AI Overviews and AI Mode, by page, country, date, device | Introduced in 2026 | Clicks, queries, position, which feature, the answer text |
Sources for the table: Microsoft's Topics Portfolio post, Microsoft's June post and a Search Console report breakdown written by an agency that sells AI search services. That breakdown's "what it leaves out" list is the author's reading of Google's help page, not a Google statement, so verify it against Google's documentation. I could not open OpenAI's own announcement; both OpenAI pages returned an access error.
OpenAI Business Hub in detail
PPC Land reports the Business Hub is part of "Businesses in ChatGPT," with Brand insights and Agent insights tabs. Verified businesses see mentions, website clicks and the top five discovery topics. For online businesses, Tomorrow's Publisher says a mention counts a reference to the business's website, not the brand name alone. Discovery topics summarize themes and do not reveal individual prompts or directly measure customer intent. A blank metric may mean insufficient activity, not zero.
Things I would watch:
- Eligibility and verification. Claiming is open to eligible businesses where ChatGPT is available, with verification and review steps. Business Agents are limited to eligible US ecommerce and online service businesses. PPC Land notes publishers are not addressed.
- Data terms. Tomorrow's Publisher says that under OpenAI's updated terms, OpenAI owns the Insight Data and restricts disclosure and commercial exploitation unless it authorizes it in writing. Service providers acting for a business may access the data for permitted purposes, under equivalent restrictions. If you report to clients or publish case studies, read those terms first.
- Unknowns. PPC Land flags that it is unclear whether claiming a profile changes how often a business is mentioned or ranked. I have no data either way.
The early dashboard numbers PPC Land shows (one live example of 690 mentions and 24 clicks over 28 days) are anecdotes from one account. Do not turn them into a benchmark click rate.
The wobble in share-of-voice measurement
Share of voice in AI answers, as I define it in what is share of voice in AI search, is your share of brand mentions across a set of prompts. It is only as stable as the sampling underneath.
| Evidence | Finding | Implication |
|---|---|---|
| Muck Rack, same prompt run twice in one day | ChatGPT repeated 41% of cited domains, Gemini 45%, Perplexity 94% | A single run on ChatGPT or Gemini is mostly noise |
| Orbit Media, questions rerun two days apart | Claude kept 30% of URLs, Gemini 38% of domains, Perplexity 65% | Weekly snapshots can show movement that is not real |
| Orbit Media, mentions versus citations | Claude 70% mentioned versus 40% cited; ChatGPT 56% versus 47% | Mention-based and citation-based share of voice give different answers |
| Orbit Media, brand size | Semrush had a 50-point gap between mentions and citations; two smaller sites had gaps of 5 and 4 | The gap depends on brand fame; Orbit labels the training-data explanation a theory |
| Tannenbaum, arXiv | Prior mention plus current exposure gave about 81 to 84% next-run mention rates; prior non-mention with no exposure about 2% | Visibility is sticky, so one-off tests overstate or understate your position |
Orbit's sample was 72 prompts across three brands, 12 to 40 per brand, which the author calls directional. The Tannenbaum model is predictive and observational. I discuss the broader pattern in share of model.
Vendor tools add a transparency problem. Amplitude's AI Visibility page, for example, lists an AI visibility score and share of voice, currently for ChatGPT and Google AI Overview only, but does not explain whether prompts are synthetic or drawn from real traffic, or how often results are collected. That is one vendor's page; the point is a checklist. Ask any vendor four questions: where do the prompts come from, how many runs per prompt, which surface (app or API), and how do you define a mention. My comparison of options is in AI visibility tools compared, and whether they justify the cost is in are AI visibility tools worth it.
An honest way to measure AI share of voice
- Fix the panel. Pick 50 to 100 buyer questions and freeze them for a quarter. The 50-prompt panel method works.
- Repeat every prompt. At least three runs per engine per measurement date. Given the repeat figures above, one run is not a measurement.
- Report ranges, not points. "Mentioned in 30 to 45% of runs on ChatGPT" is honest. "38%" is not.
- Separate mention from citation. Track both, because Orbit shows they diverge by up to 30 points for some engines.
- Triangulate with first-party data. Use Bing AI Performance and Business Hub for direction and your own analytics for visits. Treat the Search Console impressions as exposure only.
- Log the calendar. Note platform changes and data bugs. A Search Console logging error from 9 to 11 November overlapped with the November 2026 core update that began on 10 November, so impressions dips in that window need care.
- Say it is an estimate. For executives, use the format in reporting AI search impact to your CEO.
What this means for marketers and founders
- This week: claim your Business Hub profile only if you are eligible and have read the data terms. Export your Bing AI Performance data. Do not treat Search Console AI impressions from 9 to 11 November as a trend.
- Do not buy a tool on the share-of-voice number alone. Ask the four questions above first.
- Do not set KPIs on AI share of voice until you have at least two months of repeated-run baselines. A target built on one run will chase noise.
- Keep citation counts separate from sales. Citations are not recommendations, which is the argument of citations are not recommendations.
- Watch for definition changes. The "mention" definition in Business Hub could change while it is in beta.
For the Google side of reporting, see Search Console's AI report: what it shows and doesn't.
What we do not know yet
- How OpenAI defines and counts a mention, and the reporting thresholds.
- Whether Business Hub data will be exportable or available through an API. Neither source I read says.
- Whether Search Console corrected the 9 to 11 November impressions retroactively. Google's note says the issue affects data logging only.
- Whether first-party numbers from different platforms can be summed. They count different events, so I would not.
FAQ
What is OpenAI's Business Hub?
It is part of Businesses in ChatGPT, which opened in beta on 10 November 2026. Verified businesses can see aggregated mentions, website clicks and discovery topics, subject to reporting thresholds.
Does ChatGPT Business Hub show my share of voice?
No. It shows counts for your own business, not your share versus competitors. For share of voice you still need a prompt panel or a vendor tool.
Do Business Hub clicks mean visits to my site?
Not necessarily. PPC Land reports OpenAI's help note says clicks do not confirm website visits or conversions. Compare them against your analytics before trusting them.
Does Google Search Console show AI Overview clicks?
No. The generative AI performance report counts impressions of links in AI features by page, country, date and device. Clicks are not part of that report according to the breakdown I read, and its impressions are already included in your Web totals.
Why do AI share of voice numbers change every week?
Because AI answers vary between runs. Muck Rack found ChatGPT repeated 41% of cited domains in a same-day test, and Orbit found Claude kept 30% of URLs two days apart.
How many prompts do I need to measure AI visibility?
There is no verified threshold. Orbit's 12 to 40 prompts per brand was labeled directional by its author. I would start with 50 to 100 prompts, three runs each, and report ranges.
Is Bing's AI Performance report worth using?
Yes, for direction. It shows citations and grounding queries from Microsoft's own AI surfaces, and Microsoft says its Topics Portfolio figures are directional signals, not precise counts.
Want help with this?
I'm Younus Fardeen, an organic growth, SEO and AEO strategist with 4+ years of marketing experience across edtech and startup teams. If you want help setting up an AI visibility measurement method that reports ranges and ties to first-party data, see my work and reach me through the contact form at younusfardeen.com.
Found this useful? Pass it on.