Markazi Panel

How to see traffic from ChatGPT and other AI assistants

Most traffic from AI assistants does not arrive labelled. When someone taps a link inside ChatGPT, Perplexity or an assistant built into a phone, the referrer is frequently stripped — so your analytics sees a visitor with no source and files them under “direct”. The traffic is real, it converts, and it is sitting inside a bucket most owners have learned to ignore.

Why it hides in “direct”

Analytics decides where a visit came from by reading the referrer the browser sends. An ordinary search result sends one. An assistant often does not: the link may open in an in-app browser, be handed to the system browser without the header, or be rewritten through a redirector that drops it. No referrer means no source, and no source means direct.

Google Analytics added a native AI Assistant channel in 2026, which catches the platforms it recognises. It is worth knowing two things about it: the channel only helps where a referrer survived in the first place, and the list of platforms it recognises is not the list of platforms that exist. Perplexity, in particular, is worth checking in your own property rather than assuming — validate it before you report a number to anyone.

Two different questions, often confused

“Am I getting AI traffic?” is really two questions, and the answers can point in opposite directions.

Is an assistant sending me people?

This is a visit: a human arrives on your site having been pointed at it. It shows up, if the referrer survives, as a session from an assistant’s domain. It is the one that turns into orders and enquiries.

Is an assistant reading me at all?

This is a crawl: a bot fetches your pages so the model behind it can answer questions about them later. No human is involved and no session is recorded, which is why a site can be quoted regularly by an assistant while its analytics shows almost no AI referrals. If the crawlers never come, you cannot be recommended — so this is the leading indicator, and almost nobody measures it.

How to check both, today

For visits, in the analytics you already have

Look in your traffic report for sessions whose source contains chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com or claude.ai. In GA4 you can group these into a custom channel and place it above Referral so they stop being scattered. This catches the share that kept a referrer — treat it as a floor, not a total.

A rough sense of the rest: watch whether “direct” grows on pages nobody types the URL of. Direct traffic to a deep guide page is not somebody remembering the address.

For crawls, in your server logs

Crawlers identify themselves in the user-agent. The ones worth looking for are GPTBot and OAI-SearchBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot, Google-Extended and Applebot-Extended. If your host gives you access logs, grep for those names. If it does not, this is the part that needs a tool — a JavaScript analytics tag cannot see a crawler that never runs JavaScript.

Check you are not blocking them by accident

Before optimising anything, read your own robots.txt. A theme, a plugin or a security setting may be refusing the exact crawlers you want. Being listed by an assistant requires having been readable by it.

The decision is genuinely yours: some publishers block AI crawlers deliberately, because being summarised costs them the visit. For a business that wants to be recommended to someone asking “who makes X in Lahore”, blocking them is self-harm. If you want them in, say so explicitly rather than relying on a default:

User-Agent: GPTBot
Allow: /

User-Agent: OAI-SearchBot
Allow: /

User-Agent: ClaudeBot
Allow: /

User-Agent: PerplexityBot
Allow: /

User-Agent: Google-Extended
Allow: /

What makes a page quotable

Once the crawlers are in, what gets you cited is not a trick. It is the same discipline that makes a page useful to a person in a hurry:

  • Answer in the first paragraph. An assistant lifts a passage that answers the question on its own. A page that builds to its point over four paragraphs offers nothing liftable.
  • Use the question as the heading. Headings phrased the way people ask are the anchors a model matches against.
  • Serve real HTML. If your content only exists after JavaScript runs, some crawlers will never see it. Check by viewing source, not by looking at the page.
  • Be specific and be attributable. Concrete facts about your own business — where you are, what you actually do, what you charge — are what an assistant can repeat without risk. Marketing adjectives are not.

Do not expect the numbers to be large yet

For most Pakistani businesses this is a small stream today, and anyone selling you a transformation is ahead of the evidence. What makes it worth watching now is the shape rather than the size: it is growing, it is invisible in the default reports, and the work that earns a citation — answering questions plainly on your own pages — is the same work that earns ordinary search traffic. You are not choosing between them.

Common questions about installing and measuring covers how this is counted without adding another script to your site.

Written by Markazi Panel, which does the arithmetic in this guide for you — see a dashboard with a month of data in it, or ask anything at hello@markazipanel.com.

Read next