Visibility signals
Brand mentions, owned-source citations and explicit recommendations, each with a documented definition.
Decision summary
AI search measurement should tell your team what was observed, how comparable it is and what decision it supports. TANTU AI separates brand mentions, owned-source citations, explicit recommendations and commercial outcomes, with recorded questions, interfaces, dates and valid sample sizes. A useful report also explains missing observations and changes in the question set. This lets you examine whether an apparent improvement reflects stronger representation, a different sample or collection conditions before changing your content budget or drawing a conclusion about revenue.
Keep a comparable panel when the program evolves
Questions may need to change when products or markets change, but a revised sample should not silently replace the baseline. Keep stable question IDs, record the reason for a wording change and mark additions or retirements. Report the current scope for operational coverage and, when useful, compare the subset collected under comparable conditions in both periods.
Consider an illustrative report with eight mentions from 30 valid observations, followed by eight from 24. The rate rises from 26.7% to 33.3%, although the mention count is unchanged. If the missing six observations belong to a difficult platform or question group, coverage can influence the apparent trend. The report should show the planned and valid counts, identify where observations are missing and avoid presenting the percentage change alone as progress. The example explains a denominator problem; it is not a customer result.
Use uncertainty to choose the next action
Some observations justify immediate action: an incorrect product fact can be checked against approved evidence and corrected. Others need investigation: a small change in mentions may reflect ordinary answer variation. We distinguish factual corrections, repeated patterns and isolated observations so the team does not redirect a program around one unusual answer.
For a suspected pattern, inspect the underlying answers, affected question groups, source URLs and collection conditions before proposing work. If a meaningful product change altered the question set, establish a new baseline for that group instead of forcing continuity. The review should end with a named action, the evidence supporting it and a condition for revisiting it. Revenue attribution remains a separate assessment using available site and sales records; a visibility pattern alone cannot close that gap.