AI Referral Traffic

AI referral traffic is the set of visits a website receives because an AI assistant or answer engine, such as ChatGPT Search, Perplexity, Gemini, Claude or Google AI Mode, cited, linked or recommended it. Only a fraction of that traffic is labelled as such in analytics. The rest arrives looking like brand search, direct visits or ordinary Google clicks.

The useful way to think about it is as three layers: clicks that identify themselves, visits that AI caused but another channel gets credit for, and machine reads that never become a human visit at all.

What Is Measurable

Tagged and referred clicks. When a user clicks a citation in an AI interface, the visit can carry a referrer or a campaign parameter. ChatGPT appends a utm_source parameter to cited links, and extended it to links in its "More" sources section on June 13, 2025, according to PPC Land; before that change those clicks showed up as direct traffic. These visits can be grouped into an AI channel in any analytics tool by matching the referring domains of the major assistants.

Crawler and fetch activity. Server logs show which AI user agents request pages. The distinction between agent types matters here. OpenAI's documentation separates OAI-SearchBot (search index), GPTBot (training) and ChatGPT-User, which fetches a page when a user's request calls for it. A ChatGPT-User or Claude-User hit means a person's question led an assistant to read the page at that moment, which makes it the closest log-level signal of real demand. Cloudflare's 2025 Year in Review reported that this user-action crawling grew more than 15 times during 2025. A fetch is not a visit, however, and should be counted separately. See AI crawlers.

What Is Hidden

Google's AI surfaces. Google states that sites appearing in AI Overviews and AI Mode "are included in the overall search traffic in Search Console" under the Web search type. There is no first-party breakdown, so clicks from Google's AI answers cannot be separated from ordinary organic clicks.

Recommendation followed by a search. The largest hidden component may be visits where the user never clicks the AI answer. Similarweb data reported by Search Engine Journal in June 2026 found that people shown a brand in a ChatGPT recommendation were 2.5 times more likely to visit that brand's site within seven days, and that 55.9% of those visits came through branded search. In analytics those visits are indistinguishable from any other brand search. The study is limited to US desktop users in finance, travel and beauty, and it is correlational, so the multiplier should not be applied to other sectors as if it were a constant. The direction is the point: referral reports capture a minority of AI-influenced visits. This is the other side of zero-click search.

Answers with no visit. When the assistant answers fully, nothing reaches the site's analytics. Influence of that kind can only be estimated by sampling the answers themselves, the subject of AI visibility measurement.

Self-Reported Figures

Several software companies have published their own numbers. They are worth knowing and worth labelling, because each is a single company's account, unaudited, and measured differently.

The two are not comparable. Tally's figure asks people how they heard of the product, which captures the recommendation-then-search path that referral data misses but depends on recall. Webflow's counts referred traffic, which its own growth lead said likely understates the full effect. Both come from self-serve software products with technically fluent audiences, and companies with striking numbers are the ones most likely to publish them. They show what is possible, not what is typical.

How to Instrument It

Define an AI channel. Group sessions whose referrer or utm_source matches the major assistants, and track it separately from organic search and referral.

Ask customers. A "how did you hear about us" question with an explicit AI assistant option, at signup or checkout, is the only direct way to capture the branded-search path. It is how Tally produced its number.

Read the logs by purpose. Separate training crawlers, search-index crawlers and user-initiated fetches rather than lumping them together as "bots". This reporting is becoming a platform feature rather than a log-parsing project: LightCMS, the CMS serving this site, reports which AI crawlers read a site and how many visits each AI assistant refers.

Watch brand demand. Track branded search impressions and direct traffic over time alongside the AI channel. They cannot be attributed to AI with confidence, but a rise in brand demand that tracks AI visibility is consistent with the Similarweb pattern.

Do not over-read small numbers. AI-labelled sessions are usually a small share of total traffic, and citation sources shift from day to day. Trends over months are more informative than week-to-week changes.