# [ChatGPT traffic: the half your analytics cannot see](https://rebilder.com/solutions/chatgpt-traffic) > ChatGPT sends you two kinds of traffic: agent fetches and referred humans. Analytics records one and structurally cannot record the other. How to see both. - **Updated:** 2026-08-17 Your analytics can show you people who clicked through from an assistant. It cannot show you the assistant itself, and the second number is usually the larger one. ## What you get - Both kinds of AI traffic, separated and counted - Agent fetches by platform, from your own server logs - Referred human sessions, and how they convert - The answer rate: what you served against what you missed ## Two kinds of traffic, one name When people say ChatGPT traffic they usually mean one of two things, and the distinction is not pedantic: they arrive by different mechanisms, are recorded by different systems, and are worth different amounts. | | Agent fetches | Referred humans | | --- | --- | --- | | What it is | The assistant requesting your page | A person clicking a link in an answer | | Runs JavaScript | No | Yes | | Visible in analytics | Never | Yes, as a referrer | | Where it is visible | Your server logs | Your analytics | | Typical volume | Higher | Lower | | What it is worth | It decides whether you get recommended | Direct conversion | > **Why analytics cannot be fixed to show this** Page-view tools work by running a script in a browser. Agent fetches run no JavaScript and fire no beacon, so no configuration change makes them appear. It is not a setup problem; the measurement mechanism is absent. ## Seeing the agent half It comes from your own request log, which makes it exact rather than sampled. What you want out of it is four things, and all four are properties of your infrastructure rather than estimates of somebody else’s. - **Which platforms fetched you**, classified from real header samples rather than guesswork. - **What you served each one**: the format, the size and the render time. - **Your answer rate**: how many of those requests produced something useful. - **The miss log**: the URLs they asked for and did not get, ranked by frequency. The miss log is the one worth acting on first. It is a list of things the market asked you for, in the order it asked, which is a rarer artefact than it sounds. ## The referred half, and why it converts People arriving from an assistant have usually had their question answered already and are arriving to act rather than to browse. Retail measurement over Cyber Week 2025 put the conversion rate of AI-referred visitors around 38% above search-referred ones. That is the commercial argument for the agent half: the fetch is what decides whether the recommendation happens, and the referral is what you collect afterwards. Optimising the second without the first is collecting from a funnel you never filled. ## What this does not do - It does not watch assistants answer questions on your behalf. What it measures is your own server: who asked, what you served, and what you missed. - It cannot make an assistant mention you. Assistants are not deterministic and no vendor controls their output, so a guaranteed citation is not a thing anyone can sell. - It does not tell you what any assistant said about you. It tells you what was asked of your server and what your server returned. - It does not deanonymise visitors. Agent classification reads request headers, and captured queries are screened for personal data before storage. ## How much traffic does ChatGPT send? It depends entirely on your category, and any single published figure is describing somebody else’s site. The measurable version is your own: count agent fetches in your server logs and referred sessions in your analytics, and treat them as two separate numbers. ## Why does my analytics show no ChatGPT traffic? Because agent fetches run no JavaScript and fire no beacon, so a page-view tool cannot record them under any configuration. Referred humans do appear as a referrer; the agent half only ever shows up in your server logs. ## Is AI traffic worth having? The referred half converts well: retail measurement over Cyber Week 2025 put AI-referred visitors around 38% above search-referred visitors. The agent half is what determines whether that referral happens at all. ## Should I block AI crawlers? That is a real decision with real trade-offs, and it should be made deliberately rather than inherited from an old anti-scraping rule. Most sites blocking assistants today never chose to; they are simply caught by a filter written for a different threat. ## Related - [GPTBot](https://rebilder.com/solutions/gptbot) - [ChatGPT SEO](https://rebilder.com/solutions/chatgpt-seo) - [AI visibility tracking](https://rebilder.com/solutions/ai-visibility-tracking) - [AI agent user agents in your logs](https://rebilder.com/learn/ai-agent-user-agents) - [How the gateway classifies agent traffic](https://rebilder.com/docs/classification)