Trust labels and disputes
Sill labels the AI agents that visit your site by how sure it is that each one is who it says it is. Every label comes with the evidence behind it, and you can report a label you think is wrong.
Where the labels appear
Section titled “Where the labels appear”Open Agents from the main nav, then the Agents tab. A trust label sits under each agent’s name in the table, next to the agent’s category. Open an agent to see the same label at the top of its page; there it replaces the older identity badge, and Who this is follows the label.
The same labels appear in the visits panel in Activity: next to an agent’s row when there is something to flag (Verified, Likely impersonator, Impersonated or Suspicious), and in the agent’s detail header, including Unverified, with Why?. A Mismatch row shows no label, because its mark already says the sender is not who it claims; a row with a Probing pill shows only a Likely impersonator label. That way a row never carries two signals that disagree.
What the three labels mean
Section titled “What the three labels mean”| Label | What it means |
|---|---|
| Verified | Sill confirmed the requests came from the agent they name. The proof is a valid signature from a signer the agent’s company declares, or an address inside the network the company publishes for that agent. |
| Unverified | Sill could not confirm who sent the requests. This is the neutral label, and it is the answer whenever proof is missing, out of date or unclear. It is not a sign of bad intent: many genuine agents publish no way to check them. |
| Likely impersonator | Requests used an agent’s name but came from someone else. Sill only uses this label when it has positive evidence of deception, never just because proof is missing. |
Two more markers can appear next to a label:
- Impersonated — the agent is Verified, and some other requests that used its name in the same period came from someone else.
- Suspicious — some requests behaved unlike the real agent does on your site (for example, a different network or a different set of request headers). Sill shows this only after it has seen enough verified traffic from that agent on your site to compare against, and it never turns this into Likely impersonator on its own.
Labels cover the last 7 days when the Agents view is set to Today or 7 days, and the last 30 days for 30 days or 90 days.
Read the evidence with Why?
Section titled “Read the evidence with Why?”Select Why? next to a label. The list shows:
- one line explaining the label;
- the evidence, grouped under Signs of impersonation, Unusual behaviour, Proof and Why nothing stronger, each with the number of requests it applied to;
- how many requests the label is based on, and when Sill last checked.
Evidence lines are plain sentences, for example:
- “Signed by chatgpt.com.”
- “From OpenAI’s published network.”
- “Address doesn’t belong to OpenAI’s network.”
- “Used a robots.txt-only name as its user agent. Real crawlers never send these names.”
- “Part of a burst of fake AI names coming from the same addresses.”
Evidence comes only from your own site’s traffic. Another merchant’s traffic never changes your labels.
Which installs can support a Likely impersonator label
Section titled “Which installs can support a Likely impersonator label”Sill only accepts evidence of impersonation from reports it can tie to your own install. That means the Sill Cloudflare Worker or WordPress plugin, deployed from your dashboard after install keys were added, so that it signs what it sends. Reports from older installs can still show Verified or Unverified, but never Likely impersonator or Suspicious.
Report a label you think is wrong
Section titled “Report a label you think is wrong”- In Agents, select Why? next to the label.
- Select This is wrong.
- If more than one label is shown, choose the one that is wrong.
- Add a note if it helps, for example “this is our own monitoring tool”. Notes can be up to 500 characters.
- Select Send report.
A person at Sill reviews every report. While it is open, Why? shows “Your report is under review.” If Sill agrees the label is wrong, it removes that label for that agent on your site, and Why? says so. The label can come back only after Sill changes how labels are worked out.
Owners, admins and reviewers can send reports; viewers can read labels and evidence. Each site can send up to 10 reports a day, and one report per label per agent can be open at a time. Each report appears in your account’s member activity.
What the labels do not do
Section titled “What the labels do not do”Trust labels are information. They do not block or allow any request, and they do not change what your rules decide. To block traffic that uses an agent’s name, add a rule in your own firewall, such as a Cloudflare WAF custom rule.