Keep a dated log of the exact buyer prompt, platform, market, answer, named brands, cited URLs, product accuracy, and available referral data. Measure mentions and citations separately; repeated observations and stated sample limits matter more than a single score.
Define the questions before collecting results
Choose prompts tied to buyer decisions, such as a task, budget, compatibility requirement, or comparison. Group them by intent and market. Save the exact wording rather than paraphrasing after seeing a result. If a platform personalizes an answer or changes its interface, note the conditions you can observe and avoid claiming a perfectly controlled experiment.
A useful log has one row per observed answer. Preserve the raw answer or a durable reference alongside the extracted fields. A later repeat belongs in a new row with its own date, even when the question is identical.
- Question ID, exact wording, intent, target country, platform, and date.
- Named brand and exact product model, with a note on the role given to each.
- Each cited domain and full URL, linked claim, and whether the page supports it.
- Approved product fact checked, issue severity, and reviewer note.
Three outcomes that should never be merged
A brand mention means the name or model appears in the answer. A citation means a link to a particular page appears as evidence. A shopping result or product card is another surface with merchant data and a destination. One observation can have any combination of these outcomes, so a single visibility percentage hides the real problem.
For example, a page may be cited while the brand is never named; a brand may be named but linked to a retailer with an outdated variant; or a shopping card may carry the right price but the answer discuss a different use case. The remedy differs in each situation.
Compare a baseline with a follow-up carefully
Record the number of prompts and observations behind any reported rate. If five of twenty observations named the brand, state “5 of 20 observations in this panel” and include the dates and question set. Do not call that a universal probability or a stable ranking. Repeat where variation is material and report the range rather than selecting the most favorable answer.
When pages change, record the approval and publication date in a separate change log. At follow-up, compare mention, citation, accuracy, and source patterns, while allowing for other causes. Referral visits and inquiries can be included when client analytics are available, but they should not be silently attributed to one content edit.
Turn measurement into a decision
For each gap, link to the underlying observations and name the next owner: product team for incorrect specifications, editorial team for an unanswered buyer question, retailer manager for stale listings, or analyst for a recurring platform discrepancy. A dashboard is useful only if its underlying rows and interpretation are reviewable.
Our sample audit report shows how the prompt panel, answer log, source review, and action list fit together. Use it as a starting structure and adapt the questions to your catalog.