How we check answers
These results are samples of model answers. Answers in ChatGPT, Claude, and Gemini apps can differ. Changes in these results do not prove that a particular edit caused them.
What is covered
Your saved questions, models, language, market and response count define the coverage. Automated checks use provider APIs with web search. They do not observe all customer conversations or the brand’s whole market. A local search area requests context; it does not guarantee every recommendation is inside that area.
Daily checks cover selected priority questions on a named model. Weekly checks cover the full list, including those priorities. On Mondays, the priority answers from the daily model are included in the weekly check once.
What the percentages mean
Brand mentioned means the brand name or a saved alias appears in an assessed answer. Brand recommended means the answer endorses that brand. The denominator is the successful, assessed answers in the displayed selection. An answer can be useful without naming the brand, especially on a general educational question.
Missing answers and answers awaiting assessment are shown separately and excluded from rates. Human review is distinct from automated assessment; approval is a recorded decision. A completed check does not mean every claim has been verified by a person.
How comparisons work
Reports keep weekly results separate from daily observations. More frequent checks do not give priority questions more weight in the weekly scorecard. We calculate changes only for matching question identities and versions, models, language, market, response count and measurement method. A change to daily membership starts a new daily comparison group.
Repeated observations and confirmation checks can help establish whether an ordinary change persists. A possible consequential factual error should reach human review promptly. The evidence supports a decision; it does not guarantee improved answers.