The short answer
Log an appearance with its platform, exact prompt, date, settings and cited URL. Treat it as an observed response under those conditions, not a permanent rank. A reproducible record helps the team compare changes without claiming visibility across users, locations or repeated prompts it did not test.
A worked example
A fictional reviewer sees an article linked in one answer and absent in another. The log stores both prompts and timestamps, not just the flattering screenshot. The team can examine whether the answers addressed different tasks or cited different pages, while avoiding a claim that the brand 'owns' the topic in generative search.
A practical checklist
- Define a small, stable prompt set based on real reader tasks. Record the platform and test conditions before the first observation.
- Capture cited URLs and relevant response context, with a privacy-safe record of the test. Keep missing appearances in the dataset instead of deleting them.
- Compare repeated observations under similar conditions and report the sample size. Separate cited appearances from actual referral visits and business outcomes.
What to avoid
Avoid inventing a universal generative-search rank or interpreting a single response as market share. Do not store private user prompts unnecessarily. If a third-party monitoring tool supplies a score, retain its methodology and label it as that provider's measure.
A useful follow-up
Does one citation prove a permanent position?
No. It is one observed response; repeated, scoped measurement is needed to understand a pattern.
