# Agent Directory

> A directory of how companies treat consumer AI agents: whether a person's AI agent can shop, book or bank on a company's consumer-facing services, with the evidence behind each verdict. Developer and enterprise access (cloud APIs, MCP connectors for consoles) is out of scope.

Each company has a verdict (allowed, conditional, blocked, selective, conflicting, or no policy found), per-agent results for ChatGPT, Claude, Perplexity, Gemini, Meta AI and Meta Muse, a fee field and an x402 field. Every claim carries a source link, a tier and a date.

## Read this first

- "No policy found" does not mean allowed. It means no evidence has been found yet.
- Evidence tiers: T1 is a check of the company's own robots.txt, T2 a company document with a quote (confirmed), T3 a reputable report (reported), T4 an unverified lead that never sets a verdict.
- A robots.txt rule for an agent the vendor says may ignore robots.txt (for example Perplexity-User, Google-Agent) shows the site's preference, not enforcement.
- Data is checked daily at most and can be stale; cite the observed date.

## Data

- [One company](https://agentaccepted.com/api/v1/companies/amazon): full evidence, per-agent claims, robots.txt checks (rate limited; no bulk endpoint)
- [API documentation](https://agentaccepted.com/api)

## Pages

- [Directory](https://agentaccepted.com/): searchable list of companies with a finding
- [News](https://agentaccepted.com/news): dated changes in how companies treat AI agents ([Atom feed](https://agentaccepted.com/news.xml))
- [Methodology](https://agentaccepted.com/methodology): how verdicts are derived
- [About the checker bot](https://agentaccepted.com/bot): what our crawler requests and how to opt out
- [Contact](https://agentaccepted.com/contact): corrections and opt-outs