AI referral traffic from ChatGPT, Gemini and similar assistants can be separated out in your data, but only in part. Some visits arrive with a recognizable referring address or a UTM parameter, while others land in direct traffic with no information about where they came from. So instead of asking how many AI visits you get, ask what the people who definitely came from an assistant did next.
This guide measures visits that already reached your site, without guessing prompts or confusing Google impressions with clicks from AI apps. If you want to know how your content performs inside assistant answers themselves, read our piece on website visibility in AI answers.
How to recognize a visit from an AI assistant
Two signals can be verified without guessing: the referrer, meaning the page the browser came from, and a utm_source parameter added to the link by the tool that displayed it. If either points to an assistant's domain, attribute the visit to it; if neither is present, do not.
Scroll the table sideways to see all columns.
| Signal | What it shows | What it does not show |
|---|---|---|
| Referrer with an assistant domain | That the click came from the assistant's service | The conversation or the user's question |
| utm_source parameter in the URL | The source the tool assigned to the link | Whether the person read the answer or just clicked |
| Landing page | Which page the assistant linked to | Why it picked that page |
| No referrer and no UTM | That the origin of the visit is unknown | That it was not an AI assistant |
In its FAQ for publishers, OpenAI states that ChatGPT adds the utm_source=chatgpt.com parameter to referral URLs, which lets you track traffic from ChatGPT search results. It does not prove that every click from every app will arrive with that parameter. OpenAI, Publishers and Developers FAQ.
For Gemini and other assistants, we did not find an equally clear official statement about how links are tagged. Instead of assuming a format, check which referring domains actually appear in your data and add them to a list of sources. Update it when a new domain appears, not when someone feels a visit “probably came from AI”.
Why some visits end up in direct traffic
By default, browsers apply the strict-origin-when-cross-origin policy, so a cross-site click passes on only the origin domain, without the path or parameters. A service can also switch the referrer off entirely. So even a recognized visit lacks the specific chat address.
People cause further losses. Someone copies an address from an answer and pastes it into a new tab, opens the link inside a mobile app, or comes back the next day by typing your company name. Any of these paths can produce a visit with no source.
Google Analytics explains that traffic labeled (direct) / (none) consists of visits without a clear referral source, and that redirects can strip UTM parameters from the URL. The page does not discuss AI assistants, so it does not tell you what share of that traffic comes from ChatGPT. Google Analytics Help.
The practical takeaway: the number of visits attributed to assistants is a lower bound, not the full picture. Do not add an estimated slice of direct traffic to it, or you will lose the ability to verify the result. We cover broader measurement gaps, such as consent and blockers, in our article on what Google Analytics cannot show you.
Search Console shows Google, not ChatGPT
In June 2026, Google announced Search Console performance reports for generative AI features. They cover impressions of URLs from your site in features such as AI Overviews and AI Mode in Search, as well as in Discover, broken down by page, country, device and date. Useful data, but it describes visibility in Google products, not visits from the ChatGPT app or the Gemini app.
Google Search Central describes the new reports as a view of impressions within generative AI features in Search and Discover; according to a note dated August 31, 2026, they are available to all websites. The announcement lists impressions, pages, countries, devices and dates, so it does not show that the report includes clicks or queries. Google Search Central Blog.
Keep three numbers separate: impressions in Google's AI features, visits with a recognized assistant source, and inquiries after those visits. Adding them up into one “AI traffic” metric produces a figure nobody can break down later. When two tools show different numbers for the same period, our article on why the data does not match explains how to account for the gap.
How to check whether these visits lead to inquiries
Visit volume will not tell you whether this channel matters for sales; what happens after the visit will. This order gets you an answer on a small sample without a separate attribution model.
- List the recognized sources.Collect the referring domains and utm_source values that actually occurred in the chosen period. Do not add domains just in case.
- Isolate the landing pages.Write down the landing pages for these visits. Check whether assistants send people to a guide, a pricing page or your homepage, as that shows which content they treated as the answer.
- Follow the rest of the path.See whether the person moved on to your offer, contact page or a form, or left from the first page. Compare paths, not just visit durations.
- Connect the visit to a contact.Mark visits that were followed by a chat, a callback request or a ticket. That is your number of inquiries, not your number of visits.
- Read the first question.The customer's first message shows what they were looking for after leaving the assistant. It does not replace the original prompt, but it tells you what your page was missing.
The last step is often the most valuable. Questions like “do you work with businesses outside the city?” reveal that the assistant's answer was general and your page did not fill in the detail. We describe how to stitch one person's visits into a single story in our guide to the customer journey on your website.
Ask the customer instead of guessing the prompt
You will not see the assistant conversation in your data, and the landing page URL cannot reconstruct it. If the source matters to you, ask about it directly once the customer's request has been handled. Keep it short, optional and never in the way of service.
“Thanks for getting in touch. One more thing, entirely optional: how did you hear about us? For example a search engine, a recommendation, an AI assistant or something else.”
Store the answers as separate, self-reported information next to the technical source. When a customer says “from ChatGPT” but the visit has no referrer, do not overwrite the technical data; just note the mismatch. In Meetlead Analytics, each visit stores its traffic source, campaign and landing page, along with a link to the chat, callback request or ticket that followed it. So you can compare the customer's answer with what was actually recorded.
How to report traffic from AI assistants
A report should show the result and its limits. This checklist keeps the definition stable from month to month.
- State which domains and UTM values you count as AI assistants.
- Show visits with a recognized source separately from direct traffic.
- Include the share of visits with no recorded source in the same period.
- Name the landing pages that receive the most assistant visits.
- Count inquiries after the visit: chats, callback requests and tickets.
- Never merge Search Console impressions with visits from AI apps.
Compare this channel with its own results from earlier periods rather than with search. A visit from an assistant answer may start on a specific page with a different intent, just as social media traffic behaves differently from search visits. If assistants mostly send people to articles, our piece on whether a company blog generates inquiries shows how to judge their contribution.
Questions about ChatGPT and Gemini traffic
Can I see what someone asked ChatGPT before visiting my site? No. The referrer usually contains only the assistant's domain, and the UTM parameter the name of the source. You learn about intent from the landing page, the rest of the path, and the first question the customer asks in a chat or a form.
Does the generative AI report in Search Console show traffic from Gemini? According to Google's announcement, the report covers impressions in generative AI features in Search, such as AI Overviews and AI Mode, and in Discover. Look for Gemini app visits under a recognized referring domain in your own visit data; a missing domain means missing data, not missing traffic.
