How are you separating AI browser agents from real users in GA4?

by•

I started noticing weird spikes in our docs and pricing pages recently sessions that last three seconds hit four pages in a row and never scroll.

Turns out a chunk of that is browser based agents running tasks for users.

Since they use standard browser headers GA4 just lumps them in with normal traffic which completely skews our bounce rate and conversion metrics. Honestly I just want them labeled cleanly so I know what humans are actually doing versus automated runs.

Has anyone found a reliable way to tag these sessions without blocking real customers?

14 views

Add a comment

Replies

Best

noticed this in our analytics last week too. since they look like normal browsers, GA4 misses them. we started tagging sessions that jump through 0r4 pages in a few seconds with zeros clicks, then filtering them out in our reports

There are two kinds, and only one can be labelled reliably.

Agents that run in a cloud browser, like ChatGPT agent, sign every request. They send a Signature-Agent header set to "", and Cloudflare, Akamai and Vercel already verify it for you. They also come from data center IPs. Agents that drive the person's own browser, like Comet, use that person's IP and browser, and I haven't seen a clean way to separate those. Arguably that's still your customer, just with help.

GA4 never sees headers or IPs, so do the labelling at the edge. Have your CDN or server check for the signature, set a small cookie or response value the page can read, and send it from GTM as a user-scoped custom dimension. Then compare with and without it in reports and explorations instead of deleting anything. Start soon, because GA4 can't add that label to data it already collected.

The timing rule Norma mentioned is a fine backup, but a person opening four tabs from a search results page can look the same.