Define the question before the metric
Pick the audience, market and decision you care about. Use a small prompt set covering category discovery, brand understanding and announcement verification. Keep the exact wording so you can repeat it. Branded questions and unbranded category questions answer different research needs.
Keep mentions and citations separate
A mention means the answer names the brand. A citation or source link points to a page. Record both, including which claim a source supports. A source can be linked without naming your brand in the answer, and a brand can be mentioned without a source.
Record the conditions
Save the platform, exact prompt, date, region, account state, answer excerpt and source URLs. For Google searches, note whether an AI Overview appeared at all. If it did not, record that observation instead of marking it as a failed brand mention.
Do not infer an internal index
A sampled answer cannot tell you whether information is in a model’s training data or internal index. An absent mention is an observation for that query, not proof of absence from the system. The playground does not query private provider indexes.
Build a useful agency report
Show the prompts and evidence alongside your interpretation. Repeat a consistent set over time and disclose the sample size. Investigate changes in source pages and answer wording. Avoid a percentage score unless the denominator and collection method are explicit.