Read the evidence with context.
A measurement describes a particular test. Keep its limits alongside the result when you decide what to do next.
AI observations
An observation records the provider, API interface, model, prompt revision, locale, time and returned sources. Official API output does not establish what a person saw in a consumer ChatGPT or Google interface. Repeated answers can differ.
Mention and citation summaries count successful samples only. Failed attempts remain visible and are excluded from that denominator. Branded prompts and neutral prompts are reported separately: naming a brand in the question changes what the test measures. Compare like prompts, models and locales before interpreting a trend.
Competitor share of voice is found by exact name search in saved answers, against the brand, aliases and competitors recorded when the run collected. Each rate shows a 95% Wilson range and its answer count. A prompt whose text names the client is treated as branded and left out of the headline, even if it was saved as neutral. List position is the order of first mention among tracked names; it is not a search ranking.
Page audits
An audit preserves the fetched page and the time it was observed. Missing headings, descriptions, canonical links and similar checks describe that saved response. They are evidence to inspect, rather than a promise of a ranking or traffic change. Robots rules, page budgets and failed requests can leave a crawl incomplete.
Domain verification permits the connected workflow; it does not establish ownership of every claim on the page. Review the actual source before approving it as a fact.
Search Console imports
A connected property provides query and page observations for the selected period. Clicks, impressions, click-through rate and average position belong to that property and reporting window. They are not a market-wide search-volume estimate. Row limits, aggregation, privacy thresholds and the available reporting period can affect coverage.
An imported query can suggest work to investigate. It does not prove that a particular edit will improve its performance.
Priorities and reports
Opportunity scores use stated heuristics and assumptions. Open the supporting observation, assign the work and record the decision. A client report places completed work alongside subsequent observations. Timing alone does not establish that the work caused a change.
Empty or disconnected integrations stay empty. Named QA fixtures demonstrate a workflow and are not customer results.
Content and publication
Approve facts and sources before generation. A draft still needs editorial and factual review, including any source attribution. Approval belongs to an exact revision; editing it requires another approval. CMS operations preserve the target and revision history, and an uncertain response requires reconciliation before another attempt.