A SkillAEO score is a summary of the checks and observations shown in a report. It is not a universal probability of being cited, a documented external ranking factor, or proof that a recommended change will cause an outcome.
Read the underlying questions, answers, mentions, citations, and technical findings before acting on the headline number.
Start with the report scope
The anonymous public audit currently asks Perplexity 5 brand-relevant questions, with up to 3 anonymous audits per day, 1 per domain, and daily capacity limits. Signed-in plans have separate allowances.
That scope does not represent every system, prompt, user, market, or future answer. Record the domain, date, exact questions, and system before comparing results.
Read the question-level evidence
For each question, check:
- whether it is relevant to the domain;
- the complete answer, not only a label;
- whether the brand is mentioned;
- whether a URL from the domain is cited;
- which other sources or competitors appear;
- whether the description is accurate and current.
An uncited answer may still mention a brand, and a citation may still support an inaccurate or incomplete summary. Keep those cases separate.
Interpret the score as a summary
The score helps compare findings inside the report. It does not prove a citation probability or business value. Avoid labels such as “excellent” if they imply a universal external benchmark.
Use it to ask better questions:
- Which observations contributed most?
- Are any questions irrelevant or ambiguous?
- Which errors can be verified on the site?
- Which findings involve an external system the site owner cannot control?
- What changed between two runs besides the website?
Review technical findings
Technical checks may cover public accessibility, response behavior, canonical or indexing signals, structured data, and optional files. Verify each finding against the actual built or live page.
llms.txt is an optional community proposal. agent.json is SkillAEO's experimental format. Their presence can affect an internal readiness check but does not guarantee external discovery, indexing, citations, or rankings.
Prioritize by risk
| Priority | Examples |
|---|---|
| Urgent | exposed private data, false public claims, broken sign-in or purchase flow |
| High | accidental noindex, wrong canonical URL, widespread broken pages, structured data contradicting visible content |
| Medium | unclear descriptions, stale profile data, missing source links |
| Low | optional formatting or file changes with no demonstrated user harm |
Do not prioritize solely by the number of points attached to a recommendation.
Compare runs responsibly
Save a documented baseline containing:
- exact prompts;
- system and mode;
- date and relevant location or account state;
- complete responses and cited URLs;
- website changes between runs;
- analytics with stated attribution limits.
AI answers vary. A changed score or citation may reflect retrieval, model, prompt, competition, timing, or indexing changes. It does not establish that one site edit caused the difference.
Measure business impact separately
Use analytics to measure attributed visits, trials, leads, or sales where possible. Do not assume a mention or citation created a fixed conversion rate.
If attribution is unavailable, report the business effect as unknown. The report can still be useful for finding inaccurate public information or organizing a review.
Frequently Asked Questions
Is the score an objective measure of all AI visibility?
No. It summarizes the report's defined checks and sampled answers. Inspect the underlying evidence and scope.
What does a low score mean?
It means the sampled report found more gaps or fewer positive observations under its method. It does not prove that every AI system ignores the site.
Will fixing every recommendation guarantee a higher external ranking?
No. Some recommendations improve site quality or the internal score, but external systems make their own changing decisions.
How often should I compare reports?
Use a cadence based on material site changes and business risk. Repeat the same prompt set and record all relevant context.
What should I do first?
Verify the evidence, then fix false facts, security or privacy issues, broken core journeys, and accidental access or indexing problems before optional optimizations.