A useful perception report explains which claims deserve action and why.
Keep the actual wording
A single sentiment score compresses different judgments into one number. ‘Good for small teams but limited for complex reporting’ includes a strength, a limitation, and an audience boundary. Whether that description is a problem depends on your product and the customer’s requirements.
Retain the relevant passage and surrounding answer. Tag the attributes being discussed: price, usability, reliability, integrations, support, implementation, or other criteria important in your category. Include mixed and uncertain descriptions instead of forcing every answer into praise or criticism.
Separate accuracy from tone
A negative statement can be accurate; a positive one can be wrong. Review truth and sentiment as separate dimensions. An assistant incorrectly claiming you offer a required integration creates customer risk even though the statement sounds favorable.
Keep an approved fact register with a source, owner, and review date. Compare checkable claims with this register. For opinions, preserve the cited context and avoid labeling disagreement as misinformation. The right response may be to improve the product or explain a limitation more clearly.
Use severity to prioritize
Give priority to claims that affect a purchase or create a material misunderstanding. An obsolete price, unsupported compliance claim, or false availability statement matters more than a vague description like ‘less modern.’ Your team should agree on severity rules before scoring individual answers.
Trace recurring claims to cited sources when possible. Update owned facts, request evidence-backed corrections elsewhere, and record cases where no source is visible. Do not pretend a citation exists simply because the claim resembles an old article.
Review movement with examples
When reporting a perception change, show the underlying passages and the number of observations. Explain any change in prompts, scoring rules, or audience mix. A more positive score after replacing difficult questions is not evidence of improved perception on the original panel.
Close the loop with customer-facing teams. Do real buyers share the objections? Are the descriptions causing confusion on calls? An answer audit is one research channel. Combining it with customer evidence makes the resulting action more credible than optimizing for a score in isolation.
