How do I know if AI assistants are sending me traffic?
Identifying referrals from ChatGPT, Perplexity and Claude in your analytics — and why most AI-driven visits are invisible.
Look for referrer hosts like chatgpt.com and perplexity.ai in your analytics. But expect the number to understate reality substantially: a great many people read an assistant's answer, then search for your brand or type your URL directly, and that arrives as organic or direct traffic.
The referrers to look for
chatgpt.comandchat.openai.comperplexity.aiclaude.aicopilot.microsoft.comgemini.google.com
In most analytics tools these appear under referral traffic. If you keep raw logs, the referrer field carries the same information and is not subject to whatever sampling your analytics applies.
Why the number is lower than reality
This is the part worth internalising before you conclude AI sends you nothing.
- The answer often suffices. Someone asks a question, gets your information inside the answer, and never clicks. You were useful and invisible.
- Brand search follows. A person sees your name in an answer and searches for it, arriving as organic traffic with no trace of the assistant.
- Referrers get stripped. Privacy settings, some app contexts and certain link handling remove the referrer entirely — the visit arrives as direct.
- Mobile apps frequently pass nothing. An in-app browser may send no referrer at all.
Better signals than the referral count
| Signal | What it suggests |
|---|---|
| Direct traffic rising with no campaign | People are learning your name somewhere that does not pass a referrer |
| Brand searches rising in Search Console | Your name is appearing in answers |
ChatGPT-User in your logs | People are pasting your URLs into conversations — a strong direct signal |
| Long-tail question queries appearing | Your pages match how people phrase things to assistants |
| Traffic to specific answer-shaped pages | Those pages are being surfaced as sources |
The third row is the most concrete of these, because it is a request in your own logs rather than an inference from a chart.
Testing it directly
Rather than only measuring, go and look:
- Ask several assistants a question your content genuinely answers well.
- Note whether you are cited, and who is cited instead.
- Look at what the cited pages do differently — usually they answer the specific question near the top rather than covering the topic broadly.
- Repeat monthly. Answer engines change, and so does who they favour.
This takes fifteen minutes and tells you more than a referral report, because it shows you the competitive set rather than a number.
Before optimising, confirm the mechanics
If you are getting no AI traffic and no crawler visits, the problem is almost certainly access rather than content. Confirm three things first: that robots.txt permits the search crawlers specifically, that your content exists in the delivered HTML rather than being built by JavaScript, and that your edge is not challenging automated clients. Rewriting content before checking these is effort spent on the wrong constraint.
What our audit reports about this
Every item below is measured directly, not inferred. Run it against your own site and the result names the exact rule or header responsible.
- Whether each AI search crawler is permitted, naming the exact rule that decides it.
- Whether your content is present in the delivered HTML, which determines whether a crawl produces anything usable.
- Whether pages are structured so a passage can be extracted and attributed — headings, question forms, clear authorship.
- Whether structured data and author attribution give an engine something to cite responsibly.
For agents and scripts, the same measurement is at
/api/v1/ai?url=yoursite.com —
see the API documentation.
Related questions
Why is my AI referral traffic so low?
Partly measurement. Answers that satisfy the question produce no click, brand searches that follow arrive as organic, and stripped referrers arrive as direct. The measured figure is a floor rather than a total.
Should I optimise for AI referrals specifically?
The mechanics are worth fixing deliberately — crawler access and server-rendered content. Beyond that, what gets cited is the same thing that has always worked: answering a specific question directly, with attribution, as a primary source.
Can I track it better than this?
You can tag links you control, which does not help for citations you do not control. Watching direct traffic, brand search volume and ChatGPT-User requests in your logs together gives a better picture than the referral number alone.
Is AI traffic worth less than search traffic?
It arrives with more context — the assistant has usually explained what you do before the person clicks — and it is generally lower in volume. Per visitor it tends to be at least as valuable; in aggregate it is smaller for most sites today.
Read next
Which AI crawlers have visited my site?
Finding GPTBot, ClaudeBot, PerplexityBot and the rest in your own logs — and what their presence or absence actually tells you.
ReadWill AI replace search traffic to my site?
An honest look at what changes when answers appear without clicks, which content is most exposed, and what actually holds up.
ReadHow do I get my site cited by AI assistants?
The mechanics that must be right, then the content decisions that actually earn the citation — in the order worth doing them.
ReadWhat is generative engine optimization (GEO)?
Optimising to be quoted inside an AI answer rather than ranked in a list of links — what genuinely differs from SEO, and what is the same work under a new name.
Read