A name appearing in an answer is the start of the investigation. Read the surrounding passage to see whether the brand is recommended, criticized, or mentioned as a comparison.
Track a consistent prompt set and keep answers separated by platform and market. If the sample changes, record the change before comparing the next period.
Worked example
Start with
Write the brand names and aliases you intend to count before collecting answers. Include the official name and genuine variants. Do not automatically count a common word merely because it resembles the brand: “Monday” in a scheduling answer may mean a day rather than a software product.
Read the evidence
For every potential mention, keep the surrounding passage. Decide whether it identifies the intended brand, whether the brand is recommended or merely listed, and whether the wording is positive, neutral, or negative. A product can be named without being recommended; a positive phrase can describe a competitor in the same sentence.
Make the decision
Count presence once per brand per answer when calculating share of voice. Preserve passage-level occurrences for inspection without adding them to that denominator. Mark ambiguous cases for review. Check a small sample of both detected mentions and apparent absences to find missed aliases and false matches.
Take this away
Group questions by the decision people are making, then compare like-for-like samples. Keep the original text so another person can audit a disputed count instead of trusting a colored label.
Reusable outline
Keep the context with the mention
- Track mentions by prompt cluster and assistant.
- Separate positive, neutral, and weakly relevant mentions.
- Review whether mentions are source-backed or unsupported.
- Tie mention changes to site, PR, and content actions.
Build a repeatable review
- Choose a small recurring prompt set first.
- Build a review cadence so teams do not only check when there is a crisis.
- Compare brand mentions with competitor mentions in the same prompts.