How to measure traffic from AI assistants

People increasingly ask an assistant instead of a search engine. When the answer cites your page and they click, you get a visit. Measuring it takes two things apart that often get mixed up: visits from AI assistants and requests from AI crawlers.

Visits: someone clicked a link in an answer

These are real people. They arrive with the assistant as the referrer (chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com), and ChatGPT also adds utm_source=chatgpt.com to many of its links.

In Statsdot they are grouped in the AI assistants channel, so you can see the trend at a glance and click it to filter the whole dashboard: which pages they land on, whether they convert, which countries they come from.

Some apps pass no referrer, so part of this traffic shows up as direct. The real share is somewhat higher than what you see.

Requests: a crawler read your page

AI companies also send bots: to train models, to build their search indexes, or to fetch a page when someone asks about it. None of these run JavaScript, so no analytics script sees them. Statsdot can record them from your server: see AI crawler tracking. The report splits them into answers, search indexes and training.

The two together tell a story: crawlers reading a page, then visits arriving from assistants to that same page.

What to watch

• Which pages get AI visits. Usually guides, comparisons and pages that answer one question clearly.
• Conversion of AI visitors compared with search visitors. They often arrive later in their decision.
• Trend, not size. For most sites AI assistants are still a small channel; the direction matters more than the share.

Whether to let AI crawlers in is a separate decision: see GPTBot, ClaudeBot and other AI crawlers: should you block them?.