AI crawlers and bots

ChatGPT Business Hub clicks against your own counted ChatGPT visits

By Emily Parker · Nov 21, 2026 · 9 min read

What the hub reports, and what it leaves undefined

OpenAI opened Businesses in ChatGPT in beta on Nov 10, 2026. Its Business Hub shows a verified business the number of mentions and website clicks it receives. According to PPC Land's coverage, the hub has a Brand insights tab and an Agent insights tab, each with seven-day and 28-day views. The brand tiles are mentions, profile views and website clicks, and a business can claim a listing by text, phone call or DNS verification.

The help note on the clicks tile is the sentence to keep in front of you: "Number of recorded clicks on links to your website in ChatGPT. Clicks don't confirm website visits or conversions." OpenAI is telling you, on the tile itself, that the number is not a visit count.

Four things the reporting leaves open:

  • A mention has no definition. PPC Land notes that the announcement does not define one, or say whether repeated appearances in one conversation count once or several times.
  • The display thresholds are not stated. OpenAI says insights are only available when there are enough conversations or interactions. It does not say how many.
  • The data lags. Both live dashboards shared on launch day carried the label "Data through Nov 9 (UTC)", a day before the launch itself. PPC Land reads that as a reporting delay of at least a day.
  • The window may start before you did. PPC Land suggests the counts may include activity from before the business claimed its listing. That is an inference, not something OpenAI has confirmed.

OpenAI also says it does not share users' conversations with businesses. So the hub shows a count of clicks and some themes, and no question.

Peeko does not read the Business Hub. There is no connection to it, and nothing below is an import. The comparison is done by hand: you copy the hub's click number, read your own ChatGPT visit number for the same days, and divide.

Expect small numbers on both sides

A site that has just claimed a profile will probably see a single or double digit click count. Your own count will be about as small. That matters because a ratio built on ten clicks moves by 0.1 every time one visit is added or lost.

Ahrefs measured how small AI referral counts are on ordinary sites. In its March 2025 study of about 35,000 websites, sites receiving under 10,000 visits could expect 0.3% of their traffic to come from LLMs, against 0.1% for sites with more than 1 million visits.

0.3%0.1%Under 10K visitsOver 1M visits

Source: Ahrefs, AI Traffic Research: What the Data Tells Us, 2025-03-26. Retrieved Nov 17, 2026.

Read that chart with its limits. It is from March 2025, it covers every LLM source and not ChatGPT alone, and Ahrefs notes that AI traffic is often recorded as direct in analytics, so the true share is likely higher. It tells you the scale to expect and that smaller sites get a larger share, not a benchmark for your site. For why a small site's AI numbers are noisy, our post on checking your own AI-to-Google visit ratio goes through the same problem for one pair of sources.

Why the two numbers will not match

A click is not a visit

The tile says so itself. A click is recorded in ChatGPT at the moment someone selects a link. A visit is recorded on your site when the page loads and your counter runs. Between the two, the person can close the tab, lose signal or leave before the page finishes. A script based counter, Peeko included, also misses a visitor whose browser never sends the beacon. Each of those is a click with no visit.

The referrer can be missing, and then the visit is Direct

Peeko counts visits from ChatGPT, Claude, Perplexity, Copilot and Gemini from ordinary page views, by their referrer. No middleware is involved, and no cookie. The method has one hard limit: a visit that arrives with no referrer falls into Direct, and Peeko's count misses it as well. Ahrefs makes the same point about its own earlier study: "some AI traffic will pull through as 'direct'."

This is not a rare case. Search Engine Roundtable reported that ChatGPT stopped tagging outbound links with utm_source=chatgpt.com around the GPT-6 launch. Adobe's Cedric Huesler found that GPT-6 opens links with noopener,noreferrer, so the visits lost their source tag and their referrer and were filed as Direct. In his comparison of about 500 pages, ChatGPT visits fell by roughly 31,000 a day from Oct 9 to Oct 13 against the prior week, while Direct rose by roughly 26,000 a day. On Oct 15 he called it a bug and said UTM tags were returning. The article carried no statement from OpenAI. Our post on whether your ChatGPT visits moved to direct traffic shows how to look for that period in your own data. If your comparison window touches those dates, expect a low ratio that says nothing about the hub.

A shopfront with a guest book on a stand inside the open door, while across the street a person with a hand tally counter marks each passerby who points at the shop, and a few visitors enter by a side door that no one is watching.

The two windows are not the same window

The hub labels its data in UTC and lags by at least a day. Your analytics reports in whatever time zone the report uses. A click at 11 p.m. in New York on Nov 8 belongs to Nov 9 in UTC. Over seven days this shifts a few counts from one day to the next, which is enough to matter at small numbers. Pick the days from the hub's own "Data through" label, then pull your own counts for exactly those UTC days.

Nobody knows how repeats and test clicks are treated

The sources do not say whether the hub counts a second click by the same person, or whether it filters clicks made by the site owner. Your side has its own rules: if you excluded your own visits from analytics, a test click you made from ChatGPT is gone from your count and may still sit in the hub's. Our post on keeping your own visits out of site analytics covers the exclusion side.

The sets of links may differ

The tile says "links to your website in ChatGPT". Your referrer count covers every ChatGPT link that sent someone to any page on your site. We cannot say from the published material whether the two sets are identical. If the hub counts a narrower set, your ratio will run above 1. If it counts a wider one, below.

Which number to trust

Trust each one for the thing it measures, and neither as a conversion count.

Your counted visits are a floor for ChatGPT visits that carried a referrer. They come from your own server, they follow the visitor onto the page, and they connect to everything after it: a goal, a funnel, a sale. The hub cannot see any of that. When you decide which page to rewrite or whether a ChatGPT answer led anyone to sign up, use your own data. Our post on following an AI referral through to a signup shows how to do that without a cookie.

The hub's clicks are OpenAI's own record of what happened inside ChatGPT, which your server cannot see. They are the only place that tells you people were shown your link and selected it. That is useful as a sign that the profile is working. It is not a visit count, and OpenAI's help note says as much.

Do not treat the hub as an upper bound on your visits either. Because the hub does not define its terms, you cannot assume your count is capped by it.

One more distinction. A crawler fetch made while ChatGPT was answering someone is a different measurement, and Peeko files those under AI answers in its crawler tab, apart from visits. It says a page was read during a reply. The question itself never reaches your server, so nobody can report it. Do not add fetches to clicks or visits.

Compute the ratio in five steps

  1. Check the page counts at all. Before comparing anything, confirm the pages ChatGPT links to run a counter. The free analytics checker shows which of 18 analytics tools a page loads, with no signup. It reads the HTML your server sends, so it cannot see a tag added later by a tag manager.
  2. Fix the window. Read the "Data through" date on the hub. Take the seven days ending on that date, using the seven day view.
  3. Copy the hub's clicks. Record the website clicks number, the view you used and the date you read it. The figure can change as the lag catches up.
  4. Count your ChatGPT visits for the same UTC days. Match the referrer host chatgpt.com exactly, or its subdomains. Do not match on a substring. Our post on counting visits from AI assistants by referrer gives the matching rule. Count Direct visits for the same days and for the matching days a week earlier, so you can see whether Direct moved.
  5. Divide. Counted visits over reported clicks. Over several windows, add the visits and add the clicks first, then divide once. Averaging daily ratios lets one quiet day with 1 click and 3 visits dominate.

A worked example, with made up numbers so the arithmetic is visible. The hub shows 40 clicks for the seven days ending Nov 9. Your count for those UTC days is 22 ChatGPT visits. The ratio is 22 divided by 40, or 0.55. Written down in full: "22 referrer counted visits to 40 hub clicks, Nov 3 to Nov 9 UTC, hub read on Nov 12, Direct unchanged against the prior week." That sentence is the output. The 0.55 alone is not.

Reading the ratio

A ratio below 1 is the expected result. It says some clicks did not become counted visits. It does not say which ones. Abandoned loads and missing referrers both produce it. If Direct rose on the same days against the week before, missing referrers are a likely part of it. If Direct stayed flat, they are less likely. Neither one proves the cause, and with small counts a difference of a few visits is inside the noise.

A ratio above 1 is possible. It means your count holds ChatGPT visits the hub did not record as website clicks. The sources give no reason to rule that out, and we cannot say which visits those would be.

A ratio that moves from window to window is a reason to keep measuring, not to conclude. Re-read the hub after a few days too, since the figures may revise as the lag closes. If the ratio stays in a narrow band over several weeks of windows, you have a working estimate of how much of what the hub reports reaches your counter. If it swings, say so rather than quoting one figure.

What this does not settle

The comparison tells you nothing about how many people ChatGPT showed your business to. Mentions are undefined, so there is nothing to divide. It tells you nothing about visits that arrive with no referrer, since those are in Direct for every referrer based tool. And it cannot tell you whether a click led to a sale. For that, use a goal or revenue on your own side, where the visit is something you recorded.

Keep the hub as one more input on whether the profile is doing anything, and keep your own count as the record of what reached your site.