Article50ResearchBot

It loads one public page per site, the home page, the way a first-time visitor sees it, plus the first bytes of a few images on that page. It doesn't log in, fill in forms, open the chat or send it messages, or follow links to other pages.

It identifies itself in the User-Agent header. In the browser it adds its name to an ordinary Chrome user agent, so the page renders as it does for a visitor:

Before loading a page it reads the site's robots.txt and follows the rules for Article50ResearchBot, or for * if there are none. It checks again for the address a page redirects to, and before reading each image. If your robots.txt can't be fetched, it doesn't visit. To keep it out entirely, add:

It visits one site at a time, with a pause of at least 20 seconds between sites, and each visit is a single page load with the same time limits as our ordinary scanner. It visits each site once per study.

For each site it keeps a few yes/no flags and counts: whether a chat widget was detected, whether an AI disclosure was detected, and how many images it read and how many carried AI provenance data. It keeps no page text, screenshots or personal data. Results are published only as totals and percentages, never naming a site, and the per-site records are deleted when the study is done.

Add the robots.txt rule above, or email [email protected] with your domain. We'll remove it from the study, including any record already made, and won't visit it again.