Skillaeo

SkillAEO Data Sources and Evidence Record

What SkillAEO collects from public webpages and provider runs, what it stores, and what each source can and cannot establish.
Aug 17, 2026

SkillAEO uses two primary source groups: public webpage observations and model responses requested through OpenRouter. Each source is limited to the audit being run.

Public webpage observations

For a submitted domain, the auditor may request publicly accessible pages and record URLs, response status, page titles, descriptions, headings, visible text measurements, canonical links, robots rules, sitemap availability, and visible structured data. It does not sign in, bypass access controls, or claim access to private analytics.

Website content may change after the timestamp shown in a report. A successful crawl does not prove search indexing or AI use.

Provider run evidence

For every model request, the evidence record can include:

  • exact prompt and prompt type;
  • provider and model identifier;
  • product surface and web-search mode when known;
  • locale, country, and language context;
  • start time, completion timestamp, and repeat index;
  • raw answer and parsed classification;
  • citation source URLs returned with the answer;
  • provider failure, parsing failure, or valid not-mentioned status;
  • evaluator version.

OpenRouter routes requests to model providers. Prompt, domain, brand, competitor, or page information sent for a check can be identifiable service data; it is not described as fully anonymous. See OpenRouter's current provider logging documentation for provider-specific handling.

User and account inputs

Users may provide a domain, business description, site content, features, audience, email, and audit choices. Skills Pack drafts label whether fields came from user input, AI assistance, audit evidence, or a template. Unknown values and placeholders require review.

What the data cannot prove

An observed answer cannot prove market share, all-user behavior, future inclusion, ranking, or causation. A missing citation in one valid run does not establish that a brand is never cited. A failed run does not count as a brand absence.

See Methodology, Privacy Policy, and Subprocessors.