For a branded question, "is the answer positive?" comes second. The first question is "is it right?". An answer can sound glowing and still describe the wrong product, repeat a rule that no longer applies, or come from an engine that doesn't recognise your brand at all. The old Reputation page tagged answers like these as positive, because it only looked at tone.
Brand facts
Each Reputation brand now has a short list of facts: what it is and the other names it goes by, who it is for, regulatory or safety status, availability, pricing. I draft them from your website, and you edit and save them. Nothing is checked until you save, and the page never treats a drafted fact as verified.
What the page shows
- How many answers have a problem, split by type, and how many have none.
- Problems per engine, and the engine behind most of them.
- Answers to fix, most serious first. Each one shows the exact sentence, the fact it conflicts with, the prompt, the engine and the sources it cites. You can export the list as a CSV for medical or legal review.
- Every answer at a glance: your prompts × the 4 engines.
- Tone, measured only on answers with no problem.
- What changed since your previous run, and the sources behind the answers.
The limits, stated up front
An answer is only checked against the facts you saved. "No problem found" means nothing in it conflicts with them, not that it has been verified. Accuracy checks are on paid plans, because each answer needs 1 extra model call. On the free plan, the page shows tone and sources.
If your agent reads your data, GET /api/v1/brands/:id/reputation returns the same numbers.