Do AI retrieval crawlers execute JavaScript?

Nobody has published an answer. Every measurement in circulation is of training crawlers — GPTBot and ClaudeBot. The bots that actually decide what gets cited in an AI answer — OAI-SearchBot, Claude-SearchBot, PerplexityBot — have not been measured, and OpenAI's own documentation says nothing about JavaScript execution for any of its bots. Public claims about them are inference from a sibling bot.

This page is a measurement, run in the open by Aram Labs. Results will be published here, including if they contradict what we expected.

How it works

This page carries three markers. One is present no matter how you read it. One exists only in the server's raw response and is removed once JavaScript runs. One does not exist until JavaScript runs.

An agent that reports the raw-only marker read the unrendered HTML. One that reports the rendered-only marker executed JavaScript. One that cannot report the control marker did not read this page at all, and its answer about the other two means nothing.

This is not cloaking

Every requester — human, browser, crawler, agent — receives byte-identical bytes from this URL. Nothing is varied by user-agent, IP or referrer. The only difference is between the raw response and the rendered DOM, which is simply what JavaScript does on every website that uses it. We are measuring that ordinary difference, not manufacturing a special one.

Currently unmeasured, which is why this exists