Tracking AI referral traffic: what you can and can't see

When an AI assistant cites your page and someone clicks through, that visit is AI referral traffic. It is the most direct link between being recommended and getting a visitor — and it is only partly visible in your analytics. This guide explains how AI referrals are detected, where they go missing, and how to combine the signals you do have into a picture you can trust.

In short

• AI referrals are detected from the referrer of a visit and, sometimes, a utm_source parameter on the link.
• A meaningful share of AI-influenced visits arrives without either and is counted as direct traffic.
• Many buyers read an answer and never click — they search your name later. Referral data can't see that.
• Treat AI referral traffic as a trend and a landing-page signal, alongside visibility and self-reported attribution.

How AI referral traffic is detected

When a browser follows a link, it usually tells the destination where it came from in the Referer header (available to scripts as document.referrer). If someone clicks a cited link in ChatGPT's web interface, the referrer is typically a ChatGPT hostname; from Perplexity, a Perplexity one; and so on.

Detection is therefore mostly a lookup: does the referrer's hostname belong to a known assistant? Common ones include:

• chatgpt.com and chat.openai.com — ChatGPT
• perplexity.ai — Perplexity
• claude.ai — Claude
• gemini.google.com — Gemini
• copilot.microsoft.com — Microsoft Copilot

Some assistants also add a utm_source parameter to outbound links (for example utm_source=chatgpt.com). That is useful as a fallback when the referrer is missing, because the parameter survives in the landing URL.

Hostnames and parameters change as products evolve, so any detection list needs occasional maintenance.

Why AI visits go missing

Even with good detection, a share of real AI-driven visits can't be attributed. The main reasons:

Referrers get stripped

Browsers and websites can limit what referrer information is sent. Some apps open links in embedded browsers that send no referrer at all. Privacy tools and strict referrer policies can remove it too. Those visits look like direct traffic.

Desktop and mobile apps

Clicks from native apps don't always behave like clicks in a web browser. Depending on the platform and app, the referrer may be missing or generic.

Copying instead of clicking

People often copy a URL or a product name out of an answer and paste it into a new tab. No referrer, no parameter — just a direct visit, or a branded search.

Influence without a click

This is the biggest gap. A buyer asks an assistant for recommendations, reads the shortlist, and later types your brand name into a search engine or straight into the address bar. The assistant shaped the decision, but the visit is attributed to branded search or direct. No referral tracker can see this.

Setting up detection in your web analytics

Most analytics tools let you group sessions by source. A practical approach:

1. Create a dedicated channel or segment for AI assistants, so they don't disappear inside "referral".
2. Match the session source against a list of assistant domains. A regular expression like the one below is a reasonable starting point.
3. Check utm_source too, for links that carry it.
4. Review the list every few months as assistants launch new domains or change how they link.

chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com

The exact steps depend on your analytics product, so follow its documentation for custom channel groups or segments.

How Mentioned measures AI referrals

Mentioned includes a small tracking snippet you can add to your site:

<script async src="https://YOUR-MENTIONED-DOMAIN/t.js" data-key="YOUR_TRACKING_KEY"></script>

It is deliberately minimal:

• Cookieless and storage-free — it sets no cookies and writes nothing to the visitor's device.
• Silent for everyone else — it runs once per page load and only reports a visit when the referrer, or the utm_source on the landing URL, matches a known AI assistant.
• No personal data — it sends your tracking key, the assistant it detected and the page path. Never the query string, and nothing about the visitor.

It recognises ChatGPT, Perplexity, Claude, Gemini, Microsoft Copilot, DeepSeek, Grok and Meta AI. The dashboard shows visits by assistant and your top landing pages, so you can compare them with your visibility trend and see whether being recommended more often coincides with more visits.

It has the same blind spots as any referral-based method: if the referrer is stripped and there's no utm_source, the visit can't be attributed.

Reading the data

AI referral numbers are often small compared with search traffic, and they are noisy week to week. A few ways to get value from them anyway:

Watch the trend, not the total

A steady rise in AI referrals while your visibility score rises is a meaningful signal, even if the absolute numbers are modest.

Look at landing pages

The pages that receive AI referrals are pages assistants cite. Compare them with the sources you see in your answer data. If assistants link to an outdated blog post instead of your pricing or comparison page, that's a content opportunity.

Compare engines

Different assistants cite and link differently. A source that drives referrals from Perplexity may never appear in answers from another engine. Per-engine views prevent you from drawing one conclusion for all of them.

Add self-reported attribution

Add "ChatGPT or another AI assistant" as an option to your "How did you hear about us?" question on sign-up or demo forms. It's imperfect, but it is the only way to capture influence that never produced a click.

Watch branded search

If your visibility in AI answers rises, a later rise in branded search and direct traffic is consistent with buyers discovering you in an answer and coming back. It isn't proof — many things drive branded search — but it belongs in the same conversation.

Putting the signals together

No single metric captures AI's influence on your pipeline. A useful picture combines:

1. Visibility — how often and how prominently assistants recommend you (how to measure it).
2. AI referral traffic — the clicks you can attribute, by assistant and landing page.
3. Self-reported attribution — what new customers tell you.
4. Branded search and direct traffic — the indirect footprint.

When these move together, you can be reasonably confident the effect is real. When they don't, the answers behind the visibility score are the best place to start investigating.

Checklist

• An AI assistant channel or segment in your analytics
• A maintained list of assistant domains and utm_source values
• A privacy-friendly snippet for AI referrals, if you want them next to visibility
• "AI assistant" added to your "how did you hear about us?" field
• Landing pages compared with cited sources each month

Curious which assistants mention you in the first place? Start with the free audit.