Methodology
How we decide what a company's verdict is, and how much to trust it.
What we cover
We record whether a person's AI agent (one that shops, books or banks for them) can use a company's consumer-facing services. Developer and enterprise access, such as cloud APIs or MCP connectors for admin consoles, is out of scope. A company is one row, with its consumer products listed as surfaces. Segments that are not consumer-facing, like Amazon's AWS, are noted but never scored.
What counts as evidence
- Our Check: we read the company's own robots.txt and record what it says about each AI agent.
- Company Document (Confirmed): the company's own terms, policy page or statement, with a quote.
- News Report (Reported): a reputable outlet reports the company's position. It sets a verdict, but we label it as reported until a primary source confirms it.
- Unverified Lead: a blog or aggregator claim. It is shown for context and never sets a verdict.
How a verdict is decided
- What a company says and what its robots.txt does are kept separate. If they point in opposite directions the verdict is Conflicting. We never average them.
- Agents are judged one by one. If a company blocks some agents and allows others, the verdict is Selective.
- A statement about one named agent, such as Meta Muse, applies only to that agent.
- No Policy Found means we read the site and found nothing about AI agents. It does not mean the company allows agents.
- Couldn't Check means we tried to read the site's robots.txt and could not, because it refused our checker or did not respond. We say why, try again every day, and never work around a refusal.
- A verdict from robots.txt covers only the agents the site names. The verdict shows how many (for example, "Covers 3 of 5 agents"), so a partial result is not read as a rule for every agent.
How we check robots.txt
Our checker reads each tracked site's robots.txt about once a day. It requests only that file, identifies itself, and never logs in, crawls pages or works around blocks. If a site refuses the checker, we record that and conclude nothing. Read about our checker bot for details and how to opt out.
Some vendors say their user-initiated agents may ignore robots.txt (for example Perplexity-User and Google-Agent). A robots.txt rule for those agents shows the site's preference, not proof the agent is stopped, and we mark those names in italics.
Meta Muse has no documented robots.txt token and presents as an ordinary Chrome browser, so only company statements can speak to it.
x402
A company is marked as supporting x402 only when an unpaid request to a host it owns returns a valid HTTP 402 payment response, which we store as evidence. We never mark a company as not supporting x402 based on probing.
Limits and corrections
Data can be stale or wrong. Each page shows when a site was last checked and links to the evidence behind every result. If something is wrong, tell us through the contact form with a link to the source.