Know which dataset answers which question
The dashboard brings several evidence layers together. They are useful because they illuminate different parts of the customer journey; they are not interchangeable. Before using a view, read its input requirement, source, date window, units and missing-data state. If the required import has not been supplied, an empty view is a setup task, not evidence of zero performance.
Topic demand and Search & Demand
Topic-demand estimates use a disclosed model and supporting search-related evidence. They are not observed counts of private ChatGPT or other assistant prompts. Search & Demand can use supplied search-platform exports. Impressions, clicks, click-through rate and average position describe the supplied search data, not the number of AI conversations.
When importing Search Console or Bing evidence, choose a dated window, preserve query and landing-page dimensions and inspect parsing results. Map evidence to the relevant buyer question and page. Avoid summing overlapping matches as though each represents new demand. Keep modelled suggestions distinct from measured imported rows.
Market discovery and markets/devices
Market discovery draws on a separate third-party corpus with its own coverage, rather than replaying your controlled question set. Its current supported Google AI Overviews and ChatGPT corpus is scoped to US English. Read the collection label and supported coverage shown in the product; it does not prove equivalent coverage in another country or language.
Markets and devices are dimensions of the evidence that actually carries them. Do not infer a mobile result from desktop data or a regional trend from one market. A missing combination is unmeasured. Compare the same evidence type and window when making a geographic or device decision.
AI referrals require analytics evidence
Supply an appropriate GA4 export when you want to examine identifiable AI referral traffic. Useful fields include landing page, session source or source/medium, sessions, engaged sessions, key events or conversions and revenue, with the date window retained. Check that the import recognises the expected columns and that totals agree with the source export.
Identified referral sessions are not all AI-influenced visitors. Referrers may be missing, and an answer mention does not imply a click. Revenue recorded in an identified session is attribution within that supplied data; it is not proof that a specific citation or content change caused a sale.
Crawler activity and declared crawlability
Crawl insights uses supplied log evidence or supported configured events to show recorded bot requests and outcomes. Crawlability examines declared access rules. A request, an allowed policy and an answer citation are three separate signals. Keep their dates and denominators separate when presenting them together.
Answer outcomes and business outcomes
Answer outcomes follows the recorded collection and usable answers through observed mention, citation and recommendation states. Those answer states can overlap. They should not be multiplied into an invented linear conversion funnel. A recommendation is not necessarily followed by a citation, visit or sale.
When business outcome evidence is supplied, join it only at defensible keys such as the exact landing page and period. Show what is measured, modelled and unavailable. A joined register can reveal an investigation priority without proving a universal chain from demand to revenue.
Your next step
Choose one advanced view, document its required input and verify one total against the original source.
Continue to the next guide → · All Academy lessons