utm_source=chatgpt.com: how to see ChatGPT traffic in Google Analytics

What does utm_source=chatgpt.com mean?

utm_source=chatgpt.com is a label on a link. It says one thing: this link was followed from ChatGPT. Someone asked ChatGPT a question, the answer showed a link, and the link carried the label to the page it opened. UTM tags are the same labels marketers put on email and ad links so that analytics tools can tell where a visit came from.

What the tag does not tell you matters just as much.

• It says nothing about who wrote the page. The tag does not mean ChatGPT wrote it, or that it was written with AI at all.
• It is not an endorsement. An answer can link a page as the source of one sentence. The tag does not say the answer was favorable, or even correct.
• It does not prove the visitor used ChatGPT. The tag stays on the URL when the link is copied. If someone pastes it into a chat with a colleague, the colleague’s visit carries the same label.

If you received a link with this tag, you are only seeing where that link came from. On most sites, the page opens the same if you delete everything from the question mark onward. If you own the site and see it in your reports, someone reached you through an answer in ChatGPT.

Does ChatGPT still add the tag, and on which links?

Mostly yes, but read that as a behavior and not as a promise.

OpenAI’s help site had a Publishers and Developers FAQ. In the copy search engines still held on November 4, 2026, it says ChatGPT automatically adds utm_source=chatgpt.com to referral URLs, and that sites allowing OAI-SearchBot can follow these visits in tools such as Google Analytics. The page itself no longer opens: that day its address sent us to an unrelated help article. OpenAI’s developer page about its bots, read the same day, does not mention the tag. We found no current OpenAI page that lists which links get it, or that commits to keeping it.

One episode shows why that matters. Profound, a vendor of AI search analytics, published a timeline titled “ChatGPT’s missing UTM tags: what happened from October 8 to October 16”. It collected ChatGPT answers and counted how many of those with cited links had at least one tagged source link: about 99% before October 8, 2026, about 75% across October 9, generally 5% to 12% for roughly 27 hours after 19:00 UTC on October 14, and about 99% again on the morning of October 16.

These are Profound’s figures for the answers it collected, not visits, and Profound says they are no multiplier for correcting a traffic report. It also observed that a link shown directly in an answer could lack the tag while a link opened in a further panel still had it. Search Engine Roundtable’s report, updated October 15, 2026, quotes Adobe’s Cedric Huesler calling the gap a bug. It carries no statement from OpenAI, and we found none.

If your reported ChatGPT referrals fell and your direct number rose in those days, this may be why. Profound notes that the tag’s return does not recover earlier visits, so mark that week in your notes and leave it out of comparisons.

Where do you find ChatGPT traffic in GA4?

The menu names here come from Google’s Analytics Help, read on November 4, 2026. Menus move, so if one differs, look for the report’s name.

Open the report. Sign in to Google Analytics. From the left menu, select Reports, then Acquisition, then Traffic acquisition. If it is missing, Google says an Editor or above can add it back.

Change the table to sources. The report opens on Session default channel grouping, Google’s page says. Click the drop-down at the top left of the table and choose Session source / medium.

Search. Type chatgpt in the Search field between the chart and the table. Google’s help says this field uses a “contains” match and is not case sensitive.

Or filter for several assistants at once. Click + Add filter at the top of the report. In the Build filter dialog, leave Include selected. In the Dimension drop-down select Session source, and in the Dimension values drop-down tick the values you want. Click OK, then Apply. Google’s help describes picking values from a list here and does not mention regular expressions, so the pattern further down goes into a channel group instead.

You may see more than one row for the same source.

Google does not document how every combination of tag and referrer is sorted, so trust your own rows over this table. To add the rows together, switch the drop-down to Session source.

Does GA4 have an AI channel, or do you need your own?

GA4 has one. You will also hear this called LLM traffic or AI traffic: visits from people who clicked through from an assistant. Google’s “What’s new in Google Analytics” page, in its entry for May 13, 2026, announces an AI Assistant channel in the default channel group. To see it, leave the table on its default dimension and look for that row.

Its Default channel group page describes the channel as covering sources “like ChatGPT, Gemini, Deepseek, Copilot, or Grok”, and the release note names ChatGPT, Gemini and Claude. Neither gives the full list. The rule refers only to “a list of AI Assistants”, and the downloadable list of sources on that page, which we searched on November 4, 2026, covers search, social, shopping and video sites and contains none of these names. Google’s note also does not say whether older visits were moved.

So you cannot read exactly what the row counts. If this number will go in a monthly report, our suggestion is to add a channel you defined yourself. Google’s help says custom channel groups “can be applied to your reports retroactively”, which gives you history across May 2026.

Here is the pattern, for the Source field.

(.*\.)?(chatgpt\.com|gemini\.google\.com|claude\.ai|perplexity\.ai|copilot\.microsoft\.com)

We ran it in Node.js on November 4, 2026, requiring a match on the whole value. It matched the five addresses and their www versions, and none of 14 other values such as notchatgpt.com, mail.google.com and example.ai. Google’s help says GA4 regular expressions match the whole value by default, which is why this goes on Source and not on Session source / medium.

Microsoft’s Clarity documentation names chatgpt.com and copilot.microsoft.com as standalone sites and says AI sources are identified by known referral patterns. The other three are the public web addresses of those assistants and what site owners report seeing: we found no company page that lists them as referrers. If your report shows another spelling, add it.

These are the steps in Google’s Custom channel groups page. You need the Editor role or above on the property.

• In Admin, under Data display, click Channel groups.
• Click Create new channel group, which starts from a copy of the default group, and give it a name.
• Click Add new channel and name it, for example, AI assistants.
• Click + Add condition group, choose Source, choose “matches regex” and paste the pattern. Click Save channel.
• Click Reorder and drag the channel above Referral, since traffic goes to the first channel it matches. Click Apply, then Save group.

To read it, open Traffic acquisition and pick your group’s name in the same drop-down. A standard property allows 2 custom groups, so spend one with care.

Google’s page prints a broader example of its own. Run the same way, it also matches any source ending in “ai”, such as example.ai. The narrow list misses assistants you did not name; the broad one counts sites you did not mean.

Why does ChatGPT traffic run lower in GA4 than it really is?

Every number so far is a floor. These are the visits that never reach the chatgpt.com row.

Apps and missing referrers. Google’s page on direct traffic defines “(direct) / (none)” as traffic without a clear referral source, and names links without UTM parameters, redirects that strip them and ad blockers among the reasons. Clarity’s documentation says that “If the source is hidden, the traffic might appear as Direct”, and site owners report that visits from assistants’ mobile apps can arrive this way. We found no page from the assistants’ companies on how their apps open links, so that part is an observation.

Consent banners. In Google’s consent mode documentation, basic consent mode blocks Google tags completely when a visitor does not consent. With the advanced setup, a declined visitor sends cookieless pings that GA4 uses for modeling. Either way, that visit does not appear the way a consenting visit does.

Google’s own AI answers. GA4 gives them no source of their own: the Default channel group page says Organic Search includes AI Overviews and AI Mode, and the AI Assistant channel excludes them. Search Console has a separate Generative AI performance report, announced June 3, 2026 and rolled out to all websites as of August 31, 2026. Its help page describes impressions by page, country, device and date, and names no click figure. So you can see that you were shown, not how many people clicked from that answer.

Bing and Copilot. The AI Performance report in Bing Webmaster Tools, in public preview since February 10, 2026, counts citations and cited pages for Microsoft Copilot, AI-generated summaries in Bing and select partners. Microsoft’s help page says a citation does not represent traffic, so it will not match any row in GA4.

People who never click. An answer can name you, and the reader can act on it, with no visit at all. To learn whether assistants name you, you have to ask them, which is what tracking AI visibility is about.

Are bot visits counted as ChatGPT traffic?

You should not expect them to be.

The wording is from OpenAI’s bots page on developers.openai.com, read November 4, 2026. On the GA4 side, Google’s Known bot-traffic exclusion page says traffic from known bots and spiders is automatically excluded. Neither company names these two bots in that context, so this is what to expect and not a written guarantee.

The use is in the gap. Many ChatGPT-User lines for a page with few human visits suggest that ChatGPT opened the page for people who then did not click. That is our reading, and OpenAI publishes no ratio between the two. For which bots to allow and how to tell a real one from a fake, see the AI crawlers guide.

Does the tag create duplicate URLs or hurt SEO?

It creates a second address for the same page, as any tracking parameter does. Google’s canonical documentation uses the same kind of example: a URL ending in ?gclid=ABCD next to the clean address. It calls specifying a canonical “generally not critical”. Redirects and rel=”canonical” annotations are strong signals, and sitemap inclusion a weak one.

Should you strip the tag? No. It is the measurement. A redirect or plugin that removes it costs you the label and gains nothing a canonical does not already give. Google’s direct-traffic page warns that redirects can strip UTM parameters, and asks site owners to keep them intact on arrival.

A check you can do today:

• Open one of your pages with ?utm_source=chatgpt.com added to the address. The page should open normally, and the tag should still be in the address bar once it has loaded.
• View the page source and search for “canonical”. It should point to the clean address, without the tag.
• Check that your own menus, emails and sitemap use clean addresses. Google asks you to link to the canonical URL within your site, and a tagged link in your newsletter would count as ChatGPT.

Your test adds one visit to your own report, so run it once.

What should you do with the number?

The rest of this section is our suggestion, not a rule from Google or OpenAI.

See which pages get these visits. Google’s Landing page help gives the recipe: in Traffic acquisition, change the primary dimension to Session source / medium, then add “Landing page + query string” as a secondary dimension. Search for chatgpt. These are the pages visitors from answers arrive on.

Compare with organic on the same pages and dates. The report has the columns: Sessions, Engaged sessions, Engagement rate, Key events and Session key event rate. Google counts a session as engaged if it lasts longer than 10 seconds, has a key event, or has 2 or more page views. Key events are what GA4 calls the conversions you have marked, so if you have marked none, that column stays empty.

Do not treat it as your AI visibility. This number counts people who clicked and were labeled. It does not tell you how often assistants name you, or what they say. And a rise can mean more visits or simply better labeling, as October 2026 showed.

What the MWY checker does and does not do about AI traffic

MWY’s AI visibility checker does not read your Google Analytics, see your server logs or count visits from assistants.

It covers the step before the click. It reads your public pages as raw HTML, skips anything your robots.txt closes to MWYGeoCheck, and scores 24 items in three groups: can AI read it, does AI know who you are, and can AI quote it. Two items connect directly to this topic. One checks whether robots.txt lets AI search bots in, and OpenAI says sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though they can still appear as navigational links. Another checks that no page has noindex and that canonicals point to the right address. The free check gives your score, the three groups and the five items that matter most, in a few seconds, with no sign-up.

Pro opens soon. As planned, it adds all 24 items with fixes, and asks 20 buyer questions to ChatGPT, Gemini, Claude and Perplexity every week through their APIs with web search on, showing which answers name your site and which pages they cite. Those answers can differ from what one person sees in the app, and they measure being named, not clicks.

MWY does not promise that an assistant will name your business or send you visitors. Nobody outside the AI companies controls that.