Docs/ people/ Interpret Evidence Before Acting
guidev1stableVerified 2026-08-01

Interpret Evidence Before Acting

Separate direct observation from sampled and inferred findings before choosing a response.

Prerequisites

  • An evidence record that identifies the URL, platform, observation window, and collection method.
  • The Evidence Model open while you review the record.
  1. Label each result as observed, sampled, or inferred.
  2. Read crawler visits as proof that a classified request reached a URL, not as proof of citation, quotation, or referral.
  3. Treat sampled citations and quoted text as evidence tied to the documented sample and matching method.
  4. Treat detectable AI referrals as a lower bound because some clients remove referrer headers.
  5. Choose a next check that matches the evidence: access policy for request declines, page distribution for platform shifts, or samples for citation questions.

Expected result

You can state what the record supports, what it does not support, and the next observation needed before changing a page or policy.

Verify

In the dashboard or API response, confirm the evidence record includes its URL, platform, observation window, and collection method. For weekly request evidence, retrieve the same domain and range with:

curl "https://api.seellm.link/api/analytics/weekly-trend?range=90d&domain=example.com" \
  -H "Authorization: Bearer sk_live_your_api_key_here"

Troubleshooting

  • If a statement says “AI visibility increased,” replace it with the measured signal and its window.
  • If a page was crawled but absent from a citation sample, call it a crawled-not-cited gap rather than proof of never being cited.
  • If a referral total appears low, check referrer coverage before inferring user demand.