To check if ChatGPT recommends your business, give four assistants one unbranded buying task on two days. Record who they name, who they would buy from and how far they get. Business Hub shows no rivals.
Does ChatGPT recommend my business? How to run an AI competitor check, and what Business Hub does not show
Published November 20, 2026 · Last reviewed November 20, 2026
You find out whether ChatGPT recommends your business the way a buyer would: ask it to choose. Give ChatGPT, Claude, Gemini and Perplexity the buying task your customers bring, without naming any company. Run it twice on different days, and write down five things: who each assistant names, whether it opens your site, whether it gets your prices right, who it would buy from, and whether it reaches your booking or checkout. You need no tracker for that, only a few hours spread over two days. The dashboards measure something else. Since November 10, ChatGPT’s Business Hub can show a verified business its own mentions and clicks, but no prompts and no competitors.1 AI visibility trackers run prompts at scale and estimate share of voice against rivals. Neither is built to follow the buyer’s AI to the point where it books or buys. Below: the three data sources, a protocol to copy, and how to read what you find.
For the concept behind it (cited, used or closed), see agent readiness vs AEO vs GEO; this piece is the hands-on half.
Disclosure: Tobira publishes this blog and sells two products this article touches. Tobira’s AI Visibility & Outcome Report gives a buyer’s task to ChatGPT, Claude, Gemini and Perplexity and follows each one through a business’s site and its competitors’ sites. Each run stops before payment, so it tests whether a sale could close, not whether one did.2 Tobira Site Agent is an AI agent that answers and sells to people and their AI agents on a business website.3 The protocol below adapts the method published for that report, and it works with no Tobira product at all.
Three data sources, three different questions
| Source | What it answers | What it shows | What it does not show |
|---|---|---|---|
| ChatGPT Business Hub insights (first-party, ChatGPT only) | How often did ChatGPT point to my site, and did anyone click? | Mentions (for an online business, only where ChatGPT references your website), profile views, website clicks, top discovery topics | Prompts, chats, competitors, share of voice, what happened after the click |
| AI visibility trackers (sampled, many assistants) | How often am I named next to my rivals, and for which prompts? | Mention and citation rates across a prompt set you choose, share of voice, cited sources | Usually not whether an assistant can finish a booking or checkout step on your site |
| An AI competitor check (a buying task you run yourself, or a service) | When a buyer’s AI shops my category, does it pick me and get to the order? | Who is named without a brand, who the assistant would buy from and why, where it gets stuck on your site | How often you are named across many prompts; any purchase that actually happened |
The first two are about being cited: Business Hub counts it inside ChatGPT, and trackers estimate it across assistants, next to your competitors. The third is about being used: whether the assistant can do the buyer’s job on your site, and whether it picks you when it does.
Business Hub is not the only first-party source on the cited side. Bing Webmaster Tools has shown since February how often Copilot and Bing’s AI summaries cite your pages, with a sample of the queries behind them, and since October a beta list of queries where other sites are cited and yours is not.4 Google Search Console has shown impressions in AI Overviews and AI Mode for every site since August 31.5 None of them follows the buyer onto your site.
A tracker can tell you that you appear in, say, 15% of answers. A competitor check can show that the assistant named you, found no price on your pricing page and said it would book a demo with your competitor. Then you know which page to fix.
Business Hub insights: a first-party baseline, not a competitor view
OpenAI opened Businesses in ChatGPT in beta on November 10, 2026.6 Once a business claims its profile and passes a Persona identity check, its Insights tab shows mentions, profile views, website clicks and discovery topics over a rolling 7- or 28-day window.78 The definitions matter more than the totals. As Digiday reported them, with an OpenAI spokesperson’s comments:1
- A mention is not a name. For an online business, a mention counts when ChatGPT references the business’s website; for a local business, when ChatGPT shows a result for that location. “Just naming the brand doesn’t count,” the spokesperson said.
- A click is not a visit. The count covers clicks on links in ChatGPT, not arrivals on your site or purchases, and OpenAI may hold back small counts for privacy.
- Discovery topics are themes, not prompts. They group the conversations where ChatGPT cites your site. You see no individual prompts or chats, and share of voice is not part of the beta.
Early readers found the same limits. Profound’s Josh Blyskal wrote on November 11 that the Hub “tells you what happened in ChatGPT” but not why, since it has no prompts and no competitors.9 Peec AI’s Malte Landwehr noted on November 17 that it shows only your top five discovery topics, always as global data, and compares it to Google Search Console: a first-party baseline, not a diagnosis.8 Both companies sell prompt tracking, and Tobira sells the report disclosed above. Lily Ray, VP of SEO strategy and research at the agency Amsive, recorded four clicks from 25 mentions on her consultancy’s dashboard. She called the first-party numbers valuable as a point of comparison, and told Digiday: “It’s not going to replace any of the tools I’m using.”10
Claim the profile if your business is eligible, after reading the terms: OpenAI revised them on November 18 after lawyers raised concerns, and one lawyer told Digiday the new version still waives certain claims tied to Business Hub and Business Agents.11 Claiming is free for now. OpenAI’s VP of Monetization, Benji Shomair, said it “should not have an impact on how frequently it appears”, though a clearer profile may help ChatGPT match a business to more relevant conversations.1
How to check if ChatGPT recommends your business: one task, four assistants, two runs
Hold everything constant except the assistant, so runs compare with each other and month to month.
1. Write one buying task per category you sell in. Phrase it the way a buyer would, with a need and a constraint, and without your brand or any competitor’s. “Which project management tool should a 12-person agency use if we bill clients by the hour?” is a task. “Is Acme good?” is not. If you have Business Hub access, start from its discovery topics. Otherwise, start from the questions sales hears on first calls, or from how buyers’ AI agents shortlist vendors.
2. Ask the assistant to choose, then to go to the site. Paste a prompt like this and fill in the brackets:
We are [who the buyer is] and need [what you sell] for [the need], with [one constraint]. Without me naming a company, list five you would consider, then tell me which one you would pick if it were your money, and why. Next, go to [your website]: what does [your main offer] cost, how would I buy or book it there, and what, if anything, would stop you?
Then run it once with a rival’s website in place of yours, to see what the assistant finds there, and stop before any form.
3. Use ChatGPT, Claude, Gemini and Perplexity, each in its own app. Add Copilot or another assistant if your buyers use it. Where an assistant offers an agent or browsing mode, run that too, because it is the mode that opens your site and tries the form. Plain chat runs work on free plans, but most of those modes need a paid one: ChatGPT’s cloud browser, Claude in Chrome and Chrome’s auto browse among them.12 Use the apps, not the developer APIs: a university audit found that ChatGPT’s and Gemini’s APIs showed different sources from their apps for the same questions.13
4. Run in a clean session. Log out, or use a fresh account with memory and personalization switched off, and keep custom instructions empty.
5. Run each task at least twice, on different days. Answers vary between runs, and a single run tells you very little.
6. Stop before payment. Let the assistant reach the checkout or the booking request, then stop. Do not create orders or accounts on your own site or a competitor’s, and do not submit a competitor’s lead form under a false name.
7. Save everything. Keep the shared link or a screenshot of every run, with the date, assistant, mode and prompt, so next month you can tell your fix from the model’s drift.
What to record at each of the five steps
Split every run into five steps. The names are the ones on Tobira’s report page: finds you, opens your site, reads your prices, chooses you, buys or books.2 A failure at an early step hides every step after it, so record where each run stops, not just whether you were named.
| Step | What to record | What a failure looks like |
|---|---|---|
| Finds you | Is your business among the companies named for the unbranded task? Which rivals are named in most runs? | You are missing, or named only when the buyer adds a detail that favours you |
| Opens your site | Did the assistant or its agent mode load your pages, or answer from other sources? | A block, a captcha or a blank JavaScript page; the answer quotes a review site or a directory instead of you |
| Reads your prices | Did it state your price, plan limits and terms correctly? Where did it get them? | ”Pricing not available”, an old price, a competitor’s price model applied to you, a figure from a third party |
| Chooses you | Which company would it buy from, and what reason did it give? | It picks a competitor for a reason you can fix, such as a public price, a free trial or a clearer fit |
| Buys or books | Did it reach your checkout, demo booking or request form, and could it complete the step up to payment? | A required phone field, a calendar it cannot read, a “contact sales” form with no next step |
Prices are a common first fix. In a study Profound and Kyle Poyar’s Growth Unhinged published on September 3, 2026, six AI engines evaluated the pricing of Cloud 100 companies; the companies’ own pricing pages appeared in 46% of about 7,600 responses but came first in only 12%.14 It is vendor research on large software companies, and it measured citations, not purchases. Still, if your page has no price an assistant can read, the price in the answer comes from somewhere else: in that study, the most cited outside sources were Vendr, Reddit and G2. For the page itself, see why agents still bail at your pricing page.
Reading the results without fooling yourself
AI answers are noisy, and a small test can mislead you in both directions.
Count, do not average. Write results as counts: “named in 3 of 8 runs”, not “38% visibility”. With two runs on four assistants, one changed answer moves a percentage by more than ten points.
Know what two runs can and cannot show. When 600 volunteers ran 12 prompts through ChatGPT, Claude and Google’s AI 2,961 times for SparkToro and Gumshoe.ai, the lists of brands almost never repeated, and the order was close to random. How often a brand appeared was steadier, so Rand Fishkin concluded that the share of runs naming a brand is probably a reasonable metric, but only across dozens to hundreds of runs of the same prompt (he suggests at least 60 to 100).15 Two runs will not give you that share; a tracker can. They can show where a buyer’s AI gets stuck on your site, because a fixed obstacle, such as a captcha on the pricing page or a price that is not on the page, tends to show up again in the next run. Ignore positions: “named third” means little when the order keeps changing.
Separate the assistants. ChatGPT naming you in both runs while Gemini never does is a finding, and a blended score hides it. Record a missing pick too. In the university audit, ChatGPT stated a preference of its own in 79% of product recommendations and Gemini in 7%, so write down “no pick” when that happens.13
Expect drift, and re-test monthly. Profound tracked 883,000 cited pages across seven AI engines for twelve months and found that the median page loses half its citation share within 11 days of its peak.16 Run the same tasks on the same schedule, and change one thing at a time between runs.
Line the check up with first-party data. Write down Business Hub mentions, clicks and topics on a fixed day each week, since the window rolls and there is no documented export, and compare them with ChatGPT referrals in your analytics.9 Expect gaps: analytics relies on a tag ChatGPT adds to its links, and ChatGPT can drop it. Profound found the tag in about 99% of ChatGPT answers with cited links before October 8, in only about 5% to 12% on October 14, and in about 99% again by October 16. If the Hub and your runs disagree for weeks, check your tasks before you trust either.
Do not read the check as revenue. A check that stops before payment, Tobira’s included, cannot show a closed sale; that lives in your CRM, tagged by the source each lead gives.
What to fix first, and what a check cannot tell you
Fix failures in the order of the five steps: a late fix does nothing until assistants get past the early ones.
- Let the assistants in. If runs stop at “opens your site”, check your bot rules, your CDN’s bot settings and any captcha on pricing or signup pages. OpenAI’s developer docs warn that a firewall may block or interrupt ChatGPT’s cloud browser when someone asks ChatGPT to use your site, and suggest adding its agent profile to the allowlist; Akamai, Cloudflare, HUMAN and Vercel verify its signatures automatically.17
- Put the facts in plain text. Prices or a pricing model, plan limits, what is included, delivery or onboarding time, refund terms. If a price depends on scope, publish a range and what drives it. An assistant cannot compare a “Contact us” against a competitor’s number.
- Answer the reason you lost. If a competitor won on a free trial, a per-seat price or a clearer fit, that is a positioning decision first and a page edit second.
- Make the last step work for an agent. Remove fields a buyer’s AI cannot fill, such as a required phone number on a trial form. Offer a booking link with readable times, and a way to send a request with the buyer’s needs attached.
- Give the buyer’s AI someone to ask. When the transcripts show an assistant stuck on a question your pages do not answer, a site agent can take it. Tobira’s Site Agent, for example, gives AI agents that come on a buyer’s behalf a direct line to the site’s own agent, where they can get exact facts, book or leave a request.3 Agent-readable vs agent-addressable explains the difference between a site an agent can read and one it can ask.
A check has limits too. It cannot show how many real buyers asked about your category, or prove that a fix caused a later change in mentions, since the models change too. And no file, profile or agent can make an assistant recommend you.
What to remember
- To learn whether ChatGPT recommends your business, ask it to choose: one unbranded buying task, four assistants, two runs on different days, in a clean session.
- Record five steps for every run (finds you, opens your site, reads your prices, chooses you, buys or books) and note where each run stops.
- Business Hub, Bing and Search Console report how often AI points to you, and trackers measure how often you are named. A competitor check shows where a buyer’s AI drops off on your site.
- Report counts, not percentages or positions, and re-test monthly.
When an assistant picks a competitor in your first run, what reason will it give, and does your site answer it today?
FAQ
How do I know if ChatGPT recommends my business? Ask it the way a buyer would, without your brand: describe the need, ask which companies it would consider and which one it would buy from. Repeat in a logged-out or fresh session on two different days, and do the same in Claude, Gemini and Perplexity. If you have claimed and verified your profile, ChatGPT’s Business Hub adds first-party counts of mentions; for an online business, a mention means ChatGPT pointed to your website, not just named you.
Does ChatGPT Business Hub show my competitors? No. In this beta, Business Hub shows your own mentions, profile views, website clicks and discovery topics, with no prompts, chats or share of voice. For rivals, use a tracker or run a competitor check; for Copilot and Bing, Bing Webmaster Tools has a beta list of queries where other sites are cited and yours is not.
What is the difference between an AI visibility tracker and an AI competitor check? A tracker runs many prompts many times and reports how often each brand is named or cited, so it shows share of voice over time. A competitor check gives a few buying tasks to real assistants and follows each one to the end: who gets named, whether the assistant reads your prices, who it would buy from and whether it reaches your booking or checkout. The first measures how often; the second shows where a buyer’s AI gets stuck.
Why test logged out instead of in my own ChatGPT account? Because your own account may already know your company. Memory, past chats and custom instructions can put your brand into answers a new buyer would never see. Log out, or use a fresh account with memory and personalization switched off and no custom instructions.
How often should I run the check? Monthly, with the same tasks and the same assistants, changing one thing on your site between runs. Citations move fast: Profound found that the median cited page loses half its citation share within 11 days of its peak.
Sources
Footnotes
-
Krystal Scanlon, Digiday, “OpenAI’s new business agents give brands a free way into the conversation it’s also selling as an ad,” November 10, 2026, read November 20, 2026: how mentions are counted for online and local businesses (Digiday’s reporting); an OpenAI spokesperson on clicks, discovery topics and what the beta leaves out (no individual prompts, chats or share of voice); Benji Shomair, OpenAI’s VP of Monetization, on claiming a business and on clearer information helping ChatGPT surface it in more relevant conversations; the free, organic route for businesses. https://digiday.com/marketing/openais-new-business-agents-give-brands-a-free-way-into-the-conversation-its-also-selling-as-an-ad/ ↩ ↩2 ↩3
-
Tobira, AI Visibility & Outcome Report page, read November 20, 2026: the buyer’s task given to ChatGPT, Claude, Gemini and Perplexity on a business’s site and its competitors’; the five steps (Finds you, Opens your site, Reads your prices, Chooses you, Buys or books); each run stops before payment, with no orders, accounts or payments; examples on the page are labelled illustrative. https://tobira.ai/ai-visibility-report . The method (version 1.2, November 2026: each assistant runs the task twice, on different days, from a clean session) is on page 12 of the sample report offered on that page, whose figures are labelled illustrative. ↩ ↩2
-
Tobira, Tobira Site Agent product page, read November 20, 2026: a direct line for ChatGPT, Claude and other agents that come on a client’s behalf; an agent asking on a client’s behalf gets exact facts and books the demo; per the page’s FAQ, the agent also publishes llms.txt, agent.json and AGENTS.md for visiting agents. https://tobira.ai/agentic-site/ ↩ ↩2
-
Krishna Madhavan, Meenaz Merchant, Fabrice Canel and Saral Nigam, Bing Webmaster Blog, “Introducing AI Performance in Bing Webmaster Tools Public Preview,” February 10, 2026: citations in Microsoft Copilot, AI-generated summaries in Bing and select partner integrations, cited pages, and grounding queries shown as a sample of citation activity. https://blogs.bing.com/webmaster/February-2026/Introducing-AI-Performance-in-Bing-Webmaster-Tools-Public-Preview ; Krishna Madhavan, Saral Nigam and Vivek Jain, “Introducing Topics Portfolio (Beta) in AI Performance,” October 13, 2026: Top Opportunity Queries, where other sites appear and yours does not, shipped with a beta label. https://blogs.bing.com/webmaster/2026/10/introducing-topics-portfolio-beta-in-ai-performance/ Both read November 20, 2026. ↩
-
Google Search Central Blog, “Introducing Search Generative AI performance reports in Search Console,” June 3, 2026, with the note that the reports reached all websites worldwide on August 31, 2026, read November 20, 2026: impressions, pages, countries, devices and dates for generative AI features in Search, such as AI Overviews and AI Mode. https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports ↩
-
OpenAI, “Introducing Businesses in ChatGPT,” November 10, 2026, https://openai.com/index/introducing-businesses-chatgpt/ (the page returned HTTP 403 to our fetch; read through OpenAI’s announcement thread in the OpenAI Developer Community, https://community.openai.com/t/introducing-businesses-in-chatgpt-in-beta/1414865). ↩
-
OpenAI Help Center, “Setting up your business on ChatGPT”: claiming options and the Persona identity check required before publishing profile changes, viewing insights, connecting a product feed or creating a Business Agent. https://help.openai.com/en/articles/20001635-setting-up-your-business-on-chatgpt (HTTP 403 to automated fetches; read through search-index text, November 18 and 20, 2026). ↩
-
Malte Landwehr, Peec AI, “ChatGPT Business Hub: what the numbers tell you,” November 17, 2026, read November 20, 2026: a rolling 7- or 28-day window; only the top five discovery topics; global data with no country or language filter; click-through rates under 1% for the brands he checked; the comparison with Google Search Console. Vendor post. https://peec.ai/blog/chatgpt-business-hub ↩ ↩2
-
Josh Blyskal, Profound, “ChatGPT Business Hub: What its analytics actually measure,” November 11, 2026, read November 20, 2026: no prompts, conversation text, competitor data or share of voice, and no documented export or API; the advice to log the numbers on a fixed day each week; the share of ChatGPT answers whose cited links carried the utm_source=chatgpt.com tag (about 99% before October 8, about 5% to 12% on October 14, back to about 99% by October 16). Vendor post. https://www.tryprofound.com/blog/chatgpt-business-hub ↩ ↩2
-
Sara Guaglione, Digiday, “Media Briefing: ChatGPT’s new business hub is an early ‘search console moment’ for publishers,” November 19, 2026, read November 20, 2026: Lily Ray (Amsive) on four clicks from 25 mentions for her consultancy. https://digiday.com/media/media-briefing-chatgpts-new-business-hub-is-an-early-search-console-moment-for-publishers/ ↩
-
Sara Guaglione, Digiday, “OpenAI rewrites ChatGPT Business Hub terms after legal language raises concerns,” November 20, 2026, read November 20, 2026: OpenAI confirmed it updated the terms on November 18; Gary Kibel (Davis+Gilbert) on the revised terms still waiving certain legal claims related to Business Hub and Business Agents. https://digiday.com/media/openai-rewrites-chatgpt-business-hub-terms-after-legal-language-raises-concerns/ ↩
-
Read November 20, 2026: OpenAI Help Center, “Using cloud browser in ChatGPT” (cloud browser on paid plans, not Free or Go; HTTP 403 to automated fetches, read through search-index text; https://help.openai.com/en/articles/20001280-using-cloud-browser-in-chatgpt); Google AI plans (https://one.google.com/about/google-ai-plans/; Chrome auto browse for Google AI Plus, Pro and Ultra users in the US and India); Anthropic, “Claude in Chrome is generally available,” August 26, 2026 (every paid Claude plan; https://claude.com/resources/articles/claude-in-chrome-generally-available). ↩
-
Lucas G. Uberti-Bona Marin and seven co-authors (Maastricht University, Utrecht University, University of Zurich), “‘If I Had to Buy Just ONE: Galaxy S26 Ultra’: Auditing AI-Generated Product Recommendations,” arXiv:2609.18729, September 16, 2026, read November 20, 2026. Preprint, not peer-reviewed: 1,536 responses to product queries from ChatGPT and Gemini (apps and APIs) and Google’s AI Overviews; recommended products often changed across repeated requests; ChatGPT expressed a first-person product preference in 79% of product-recommending responses, Gemini in 7%; APIs shared 12.0% (ChatGPT) and 14.8% (Gemini) of displayed domains with their apps. https://arxiv.org/abs/2609.18729 ↩ ↩2
-
Kyle Poyar (Growth Unhinged) and Nikolas Laskaris (Profound), “What AI agents really think about your pricing,” September 3, 2026, read November 20, 2026: about 7,600 responses to “Evaluate Company X on pricing” across ChatGPT, Google AI Mode, Google AI Overviews, Gemini, Copilot and Perplexity for Cloud 100 companies, collected July 29 to August 10, 2026; the most cited outside sources were Vendr, Reddit and G2. Vendor research. https://www.tryprofound.com/blog/what-ai-agents-really-think-about-your-pricing ↩
-
Rand Fishkin, SparkToro, “NEW Research: AIs are highly inconsistent when recommending brands or products; marketers should take care when tracking AI visibility,” January 27, 2026, read November 20, 2026, research run with Patrick O’Donnell of Gumshoe.ai, which sells AI tracking: 600 volunteers ran 12 prompts through ChatGPT, Claude and Google’s AI Overviews or AI Mode 2,961 times; less than a 1 in 100 chance that ChatGPT or Google’s AI gives the same list of brands in any two of 100 responses, and lists in the same order rarer still; Fishkin’s conclusion that the share of runs naming a brand is a reasonable metric when a prompt is asked many times. https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/ ↩
-
Profound, “The half-life of an AI citation is 11 days,” September 30, 2026, read November 20, 2026: 883,000 cited pages, 1.19 million page-and-engine lifecycles across seven engines over twelve months. Vendor research. https://www.tryprofound.com/blog/the-half-life-of-an-ai-citation-is-11-days ↩
-
OpenAI Developers, Agentic Commerce, “Help ensure cloud browser access to your site,” read November 20, 2026: the cloud browser can visit your site, but a firewall may block or interrupt the visit when someone asks ChatGPT to use the site on their behalf; the advice to add ChatGPT’s agent profile to the firewall’s allowlist; Akamai, Cloudflare, HUMAN and Vercel verify ChatGPT request signatures automatically. https://developers.openai.com/commerce/agent-access.md ↩