Your direct traffic line keeps growing and nobody on the team can say why. Sign-ups from "unknown" sources creep up, the branded search numbers do not explain it, and no campaign was switched on.
One likely suspect is people clicking links inside AI assistants. They read an answer, tap a citation, land on your page, and your analytics files the visit under whatever bucket is nearest to hand. This post shows how to find those visits, why they hide, and how to report them as a channel of their own. If the basics are new, start with what web attribution is.
What is AI assistant traffic?
AI assistant traffic is website traffic that arrives from an AI assistant, such as ChatGPT, Perplexity, Claude or Gemini, when a visitor clicks a link or citation in its answer.
It is easy to confuse with two neighbours, so draw the lines early:
- AI crawlers are bots that fetch your pages to train models or build answers. They are not visitors and never convert.
- AI ads are paid placements inside an assistant. They carry campaign parameters and click IDs you control. Our guides on how ChatGPT Ads attribution works and on ChatGPT Ads installs showing as organic cover that paid side for apps.
- AI assistant traffic is the unpaid, human click from an answer. This post is about that one.
Why does AI traffic hide in direct and referral?
AI assistant visits often land in "direct" because the browser sends no referrer, or in "referral" because the assistant's domain is treated like any other website. Neither bucket tells you the visit came from an AI answer.
Common causes:
- Stripped referrers. Some apps and in-app browsers can fail to pass the referring page, so the visit has no source at all.
- Copied links. A visitor copies the URL from an answer and pastes it into a new tab. That is a direct visit by definition.
- In-app browsers and redirects. A link opened inside a mobile app, or sent through a redirect, can lose the original referrer on the way.
- Default channel rules. Most analytics tools group an assistant's domain under generic referral unless you tell them otherwise.
What the assistants pass, as of November 2026
Behaviour here changes without notice, so treat this as a dated snapshot and test your own data.
- ChatGPT: links often carry a
utm_source=chatgpt.comparameter. Public reports say that tag went missing for a period in October 2026 and later returned, with the cause unconfirmed. It is not on every link. - Perplexity: may pass a referrer. We could not confirm whether it appends UTM parameters.
- Claude, Gemini, Copilot and Grok: we could not confirm that any of them append UTM parameters. Assume nothing and check what arrives in your own reports.
How do you spot AI assistant visits?
Work through three signals, in this order.
- Referrer domain. Look for these in your referrer report. These are the domains most commonly listed for each assistant. Whether a given click actually passes one varies by app, browser and assistant, so confirm what appears in your own reports.
- ChatGPT:
chatgpt.com(older visits may showchat.openai.com) - Perplexity: public guides say it may pass a referrer; we could not confirm this. Domain:
perplexity.ai - Claude:
claude.ai - Gemini:
gemini.google.com - Copilot:
copilot.microsoft.com - Grok:
grok.com, and possibly another hostname
- ChatGPT:
- UTM parameters. Filter landing URLs for
utm_sourcevalues that name an assistant. Our guide to UTM naming conventions that scale helps you keep your own tags from colliding with these. - Landing page pattern. AI visits often land on deep, specific pages such as a comparison or a how-to, not the home page. A cluster of direct visits on those pages is worth a second look.
Quick manual check: open your referrer report for the last 28 days, search for each domain above, and note the landing pages. Then compare against direct visits to the same pages.
How do you report AI traffic as its own channel?
Report AI assistant traffic as its own channel by applying an ordered rule: match the referrer domain first, match the UTM source second, and send everything else to your existing buckets.
A simple rule set:
- If a referrer matches one of these domains, label it AI assistant.
- Else, if
utm_sourcenames an assistant, label it AI assistant. - Else, leave the visit in its usual bucket (direct, referral, organic or social). Do not guess.
Then measure what matters. Visits alone mislead, so compare each assistant on visitors, sign-ups, purchases and revenue. A channel with few visits and strong sign-up rates is a different story from one with many visits and none.
Where Linkrunner fits
Linkrunner is a mobile and web attribution platform, and its Web SDK (beta) classifies visits from six assistants (ChatGPT, Perplexity, Claude, Gemini, Copilot and Grok) as their own AI assistant class. It sorts other visits into traffic classes such as paid, organic, social, referral and direct. Because it also records sign-ups and purchases, you can compare conversions per assistant, and the same platform measures your app installs.
Two details help when you read the numbers:
- Last touch and first touch are both kept (see first touch vs last touch attribution for websites). Last-touch UTMs are held for 24 hours from the campaign click and last-touch click IDs for 90 days, while first-touch values stay until the visitor clears browser storage. That matters for any last-touch attribution report, because the attribution window decides whether a later visit steals credit.
- Setup is a script tag. The web attribution hub and the Web SDK docs cover the beta in detail.
What can AI traffic tell you, and what can't it?
AI assistant traffic can show which pages earn citations and whether those visitors convert, but it cannot show how many people saw your brand in an answer and never clicked.
Keep your claims small:
- Volumes are often small and noisy. One viral answer can swing a week. Do not build a story on a handful of visits.
- Quality varies by page. A visitor from a comparison answer is not the same as one from a definition answer. Segment by landing page before you compare.
- Zero-click is real. Many people read the answer and never visit. Their influence is invisible here.
- Your own site decides the size. We will not quote a typical share of traffic, because it depends on your content and audience. Measure yours for a month.
- It is one touch in a longer journey. Someone may meet an assistant, then search your brand later. A multi-touch attribution view and our guide to multi-touch versus single-touch models explain why a last-touch report can undersell it.
What should you do with the data?
Once AI visits have their own line, three actions follow:
- Find the pages that earn citations. List landing pages by AI visits and sign-ups. Those are your proven answers.
- Improve what is almost there. Pages that get AI visits but few sign-ups need a clearer next step. Pages that rank for the topic but get no AI visits may need a direct, quotable answer near the top.
- Tag links you control. Where you share a link yourself, add consistent UTMs so it never depends on what an assistant passes.
Re-run the manual check monthly. Assistant behaviour shifts, and last month's rule can quietly stop matching.
FAQ
How do I find traffic from ChatGPT in my analytics?
Filter your referrer or source report for chatgpt.com and chat.openai.com, then filter landing URLs for utm_source=chatgpt.com. Combine both, because neither is guaranteed on every click.
Why is AI traffic often counted as direct? Direct is where a visit lands when no referrer or campaign tag arrives. Copied links, stripped referrers and in-app browsers all remove the source, so some AI clicks cannot be separated from typed-in visits.
Is AI assistant traffic the same as AI crawler traffic? No. Crawlers are bots that fetch your pages and never convert. AI assistant traffic is a human visitor who clicked a link in an answer, so it belongs in your visitor and conversion reports.
Can I tag links so AI traffic is easier to track? Yes, for links you place yourself, such as in your own profiles and posts. You cannot tag the links an assistant generates, so you still need the referrer and UTM rule for those.
How big is AI traffic for most sites? It depends on your content, your audience and which assistants cite you, so no single figure is reliable. Measure your own over a few weeks, using visitors and sign-ups rather than visits alone.
Next step
Build the three-step rule, run the manual check on your last 28 days, and put sign-ups next to visits. If you also run an app, request a demo from Linkrunner to see web visits, sign-ups and installs in one place while the Web SDK is in beta.