Spec / Agent listings
Verifying agents on closed platforms
AI agents on platforms like ChatGPT, Gemini, Claude, Poe and Character.AI live behind walled gardens — there's no DNS to point at. AI Identity verifies them anyway, by reading the agent's own public listing.
The pattern
- You start a verification challenge for your Custom GPT (or Gemini Gem, Claude Project, Poe bot, etc.).
- We give you a short token, e.g.
aii-9k3xQ7m2p4. - You paste it anywhere visible in the agent's description, instructions, or website field.
- You click Verify. We fetch the public listing JSON or HTML, look for the token, and snapshot the entire response as permanent evidence.
- You can immediately remove the token. The proof is point-in-time — only the agent's owner could have placed it.
Two sub-modes
Paste & remove
Paste a short token in the description. We snapshot proof, you remove it. Trust score: +5. Re-verifies nightly only to refresh the snapshot — never penalises.
Permanent pointer
Set the agent's "Website" field to aiidentity.org/p/<your-id> (the Passport URL). Trust score: +10. Ongoing proof. Free verified backlink to your AI Identity profile.
Supported platforms
| Platform | Method | How we read it |
|---|---|---|
| ChatGPT (Custom GPT) | chatgpt_gpt | Public gizmo JSON endpoint |
| Gemini Gems | gemini_gem | Public Gem URL scrape |
| Claude Projects | claude_project | Public project URL scrape |
| Poe bots | poe_bot | poe.com/<bot> page scrape |
| Character.AI | character_ai | Character page scrape |
| Hugging Face Spaces | huggingface_space | HF API (api/spaces/owner/name) |
| Replicate Models | replicate_model | Replicate API (v1/models/owner/name) |
| Anything else | generic_agent_listing | Any URL — we look in the response |
Why this is honest
None of these platforms (yet) issue cryptographic ownership attestations. So we never claim we can. What we can prove is: at this moment, the public listing for this agent contained our token — therefore the person who controls the agent placed it there. We snapshot the full response so the proof is auditable forever, even after you remove the token.
When OpenAI, Google or Anthropic eventually publish official agent-ownership attestations, we light up a green "Platform-attested" badge automatically. Until then, listing-scrape is the best honest evidence available — and it's good enough to defeat the impersonation attacks that matter.
Re-verification
Every night, our cron re-checks link-mode listings. If the pointer URL has been removed, the badge dims to "stale" and the owner gets an email. Token-mode listings just refresh their snapshot timestamp — the original proof always stands.
Try it
Open your dashboard, pick an identity, choose "ChatGPT — Custom GPT" from the verification picker, and follow the prompt.