How to Track ChatGPT and AI Referral Traffic

For most websites in 2026, AI assistants send somewhere between a fraction of a percent and a few percent of total traffic, and ChatGPT is usually the largest single source. You track it the same way as any referral: by the referrer hostname (chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com) and by the utm_source=chatgpt.com parameter ChatGPT adds to links it cites. The catch is that part of AI-driven traffic arrives with no referrer and lands in “direct,” so the number you see is a floor, not the total.

How much traffic comes from ChatGPT?

Industry studies published through 2025 and 2026 consistently put AI referrals at roughly 0.1% to about 1% of total website visits on average, with large variation by sector. Developer tools, SaaS and B2B documentation sit at the high end; local businesses and entertainment sites at the low end. Among AI sources, ChatGPT sends the majority of referrals in most datasets, though its lead is shrinking: Similarweb's data shows ChatGPT's share of visits to generative AI platforms falling from about 76% in June 2025 to about 53% in May 2026 as Gemini and Claude grew.

Averages are not very useful for your own site. What matters is your trend line and what those visitors do, which is why the rest of this article is about measuring it yourself.

Which referrers identify AI assistants?

You can paste any referrer URL into our referrer channel checker to see how it is classified. Note that a referrer is only the origin (for example https://chatgpt.com/) in most cases, because browsers apply a strict-origin referrer policy by default; you will not see which conversation sent the click.

What does utm_source=chatgpt.com mean?

It means the visitor clicked a link in ChatGPT. According to OpenAI's publisher FAQ, ChatGPT automatically includes utm_source=chatgpt.com in referral URLs so publishers can track traffic from ChatGPT search. Because the tag is part of the URL, it survives cases where the referrer header is stripped, which makes it the more reliable of the two signals for ChatGPT.

Two side effects to know about. First, if you build canonical URLs or caches from the full request URL, the parameter can create duplicate cache entries; strip utm_* before caching. Second, some analytics setups count utm_source as a campaign; ChatGPT links then appear under campaigns rather than referrals. If you want a refresher on how the tags work, see UTM parameters.

Why does some AI traffic show up as direct?

A click only carries a referrer if the browser sends one. Several common AI paths don't:

• Native mobile and desktop apps open links in an in-app browser or the system browser, often without a referrer.
• Links rendered with rel="noreferrer" strip the header deliberately.
• People copy a URL out of an answer and paste it into a new tab.
• Someone reads an AI answer that names your product, then types your domain or searches your brand on Google. That visit is caused by AI but attributed to direct or branded search.

The last case is probably the largest and the least measurable. Two practical proxies help: watch direct traffic that lands on deep pages (people rarely type /docs/webhooks/retries by hand), and add “ChatGPT / AI assistant” as an option in a “How did you hear about us?” field at signup. Self-reported attribution is noisy, but it catches what referrers cannot.

How do you track AI referral traffic in GA4?

GA4 puts AI referrals in the generic Referral channel by default. To report them as their own channel, create a custom channel group:

1. 1.In GA4, go to Admin → Data display → Channel groups and create a new channel group (copy the default one so the other channels stay intact).
2. 2.Add a channel named “AI assistants” with the condition Source matches regex, using the pattern below.
3. 3.Move the new channel above Referral in the list. GA4 assigns traffic to the first matching channel, so order matters.
4. 4.Save, then use the new channel group in Traffic acquisition reports and explorations. It applies to historical data as well.
5. 5.Build an exploration with the AI channel, landing page and your key event (signup or purchase) to see which pages AI sends people to and whether they convert.

^(chatgpt\.com|chat\.openai\.com|(www\.)?perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|you\.com|chat\.deepseek\.com|chat\.mistral\.ai|grok\.com)$

Session source is set from utm_source when present, so ChatGPT clicks with the tag match on chatgpt.com either way. Re-check the list every few months; new assistants and new domains appear regularly.

How do privacy-first analytics tools handle AI referrals?

Most lightweight tools show the raw referrer host, so chatgpt.com and perplexity.ai appear in the sources list without configuration, and you can filter on them. Some go further and group them into an AI channel automatically. VisitTrack does: referrers from ChatGPT, Perplexity, Claude, Gemini, Copilot and You.com are classified as an AI channel next to search, social, referral and direct, and UTM sources such as chatgpt.com show in the campaigns view. The comparison pages for Plausible and Google Analytics describe how each tool reports sources.

How do you measure whether AI traffic converts?

Volume is the wrong scoreboard for AI referrals. Visitors who click a citation have usually read a summary of your page and chosen to go deeper, so they tend to arrive with clear intent. Measure:

Revenue per visitor is where small channels prove their worth. If you attribute payments to the source of each customer's first visit (revenue attribution), you can say “ChatGPT sent 1.2% of visitors and 4% of new revenue,” which is the sentence that justifies investing in it. Our revenue per visitor calculator helps compare channels on that basis.

How do crawler data and referral data fit together?

Referrals are only the clicks. Before a click, an assistant has to fetch or index your page. Combining both views tells you where you are in the funnel: user-triggered fetches (ChatGPT-User, Claude-User, Perplexity-User) without referrals mean you are being read but not clicked, and referrals rising on specific pages tell you which content earns citations. Setting up the crawler side is covered in how to track AI crawlers, and improving citation odds in GEO for SaaS.

Can you tell which question sent the visitor?

No. Unlike search engines, which at least share query data through tools like Search Console, AI assistants pass nothing about the conversation: the referrer is a bare origin and the UTM tag only names the source. The landing page is your best proxy for the question. A visitor from chatgpt.com landing on /pricing was probably comparing costs; one landing on a deep docs page was probably debugging something. Grouping AI landing pages by intent (comparison, pricing, how-to, reference) gives you a rough map of what people ask assistants about you.

You can complement that from the outside by running the questions your buyers ask through ChatGPT, Perplexity, Claude and Gemini every month, and noting whether you are cited and which URL is linked. It is manual, it is a sample rather than a census, and answers vary between runs, but it shows the other half of the picture: the answers where you are absent.

What does a useful weekly AI traffic report include?

Keep it to one screen. The goal is to notice change, not to produce a dashboard nobody opens:

• AI visits this week versus the four-week average, split by assistant. One line each.
• The top five landing pages for AI visitors, with their conversion rate next to the site-wide conversion rate.
• Signups and revenue whose first touch was an AI assistant, so the channel is judged on outcomes.
• Direct traffic to deep pages, as a rough indicator of AI-driven visits that lost their referrer.
• User-triggered crawler fetches on your most important pages, as a leading indicator for next week's clicks.

If you share links to your site in places AI assistants read (docs, GitHub READMEs, community answers), tag them with consistent UTM parameters using a UTM builder, so that downstream clicks keep their origin even when an assistant re-surfaces the link.