How do I detect AI agents on my website?

Signals in your logs

Start with what you already collect. Server logs are more honest than client-side analytics, because a lot of agent traffic never runs your tracking script.

• User-agent strings: many assistants and agent frameworks identify themselves, though this is voluntary and easily spoofed.
• Requests that skip your JavaScript entirely — visible in server logs, absent from analytics.
• Sessions with near-zero dwell time between actions, or perfectly regular intervals.
• Deep entries straight to a product or pricing URL with no referrer and no journey before it.
• Repeat visits from datacentre IP ranges rather than residential ones.

What the signals do not tell you

Detection tells you agents showed up. It does not tell you whether they succeeded — and that is usually the number that matters.

A page can be popular with agents and still fail every one of them at checkout. Volume without outcome is a vanity metric here.

The measurement worth having

Two things, tracked over time: how much of your traffic is non-human, and what share of agent journeys reach the intended end state. The second is the one that moves revenue.

PlateTale measures the second directly, by running the journeys itself rather than inferring them from logs.