|

Quickstart

Get AI bot retrieval showing up in your dashboard in about five minutes. This is the Cloudflare path - the quickest to set up, and it works on every Cloudflare plan including the free one. On a different stack? See reporting paths - you can also proxy through a free Cloudflare account, or report from your origin server directly.

What you get: every AI crawler or assistant that fetches a page on your site becomes a content_retrieved event in your dashboard (source_role: edge) - user agent, bot category, ASN, country, response status. No visitor IPs, no cookies. The page load is untouched - telemetry is sent after the response, asynchronously.


1

Register your domain

Sign up as a content owner and add your domain. You verify ownership by serving a .well-known/openattribution.json file (or a DNS TXT record). Once verified you get an API key that looks like oat_pub_..., scoped to write telemetry for that domain. Keep it handy for step 2.

The same key, with read scope, is what powers your dashboard - so you query your own telemetry through the API too.


2

Deploy the Worker

The open source Worker detects AI bot requests at the edge and reports them. Apache 2.0, single file, no build step.

bash
git clone https://github.com/openattribution-org/cloudflare-worker
cd cloudflare-worker
npm install

# Copy the template - you edit this next
cp wrangler.example.toml wrangler.toml

In wrangler.toml, set routes to the hostnames the Worker should run on and the zone id from your Cloudflare dashboard:

wrangler.toml
routes = [
  { pattern = "example.com/*", zone_id = "your-zone-id" }
]

[vars]
OA_TELEMETRY_ENDPOINT = "https://telemetry.openattribution.org/events"

Then store your API key as a secret (never in the toml) and deploy:

bash
npx wrangler secret put OA_API_KEY     # paste your oat_pub_ key

npx wrangler deploy
Not on Cloudflare?
You can put any site behind a free Cloudflare account - it adds sub-50ms of latency and immediately unlocks bot classification, then you deploy the Worker on top. WordPress, Vercel, Netlify, Fastly, Akamai and CloudFront integrations are in progress - see the integrations status page, or get in touch for early access.

3

Check it works

Hit one of your own pages with a bot user agent, then open your dashboard:

bash
curl -A "GPTBot/1.2 (+https://openai.com/gptbot)" https://example.com/

Within a few seconds you should see a content_retrieved event for that URL, with source_role: edge. After that, real AI crawlers and assistants show up on their own as they visit. Low-traffic sites see a trickle, not a stream - give it days, not minutes.

What gets detected
On Free and Pro plans the Worker matches against ~40 known AI bot user agents - GPTBot, ClaudeBot, ChatGPT-User, PerplexityBot, OAI-SearchBot and the rest. Some bots can't be caught by user agent alone (OpenAI Operator disguises as Chrome, Grok as Safari, Google's AI training uses the Googlebot UA) - Enterprise Bot Management catches those via Cloudflare's verified bot classification. See the Cloudflare integration page for the full detection model and event payload.

Next steps

  • Cloudflare integration - detection tiers, enrichment signals, access gating, the Zaraz alternative, and the API-token deployment if you'd rather we run the Worker for you.
  • The .well-known file - what the manifest declares and how agents, CDNs and marketplaces use it.
  • API reference - read your telemetry, manage delegations, or report events from your origin server without the Worker.
  • Reporting paths - how edge, marketplace and agent-side reporting fit together.
Building an AI agent, not running a site?
Different path: request agent access for a key, then see agent getting started and the telemetry spec.