Answer Radar
Reading the Visibility Metrics
Answer Radar has no single visibility score. It reports six defined metrics per segment, each computed only over admissible samples and each carrying its definition and a disclosure block. This article explains five of them as a reader of the Overview and the metrics tools meets them; mean position has its own article.
Quick Answer
Presence rate is the share of admissible answers that named or cited you. Recommendation rate is the share that recommended you, always lower or equal. Share of voice divides vendor visibility among products, one vendor once per answer. Sentiment mix and citation share describe tone and sources. A segment with nothing admissible reports usable false and no numbers.
- Metrics are per segment: engine, model, tool version, geography, language. Never averaged across engines.
- Only grounded samples count. Errors, ungrounded and thin samples are excluded and reported by grade.
- Every figure travels with its definition, sample counts and version stamps.
On this page
Presence rate
The share of admissible samples in which the buyer's brand was named in the answer text or cited by URL. Presence is an observation about the answer, not a judgement about ranking or endorsement.
Named or cited: an answer that cites your page without saying your name still counts you present, matching how a gap is decided not to exist. Presence is a whole-word match against the answer text and a domain match against citations; nothing about vendor grouping can move it.
Recommendation rate
The share of admissible samples in which the buyer's brand was not merely named but recommended, as judged by the mention extractor. Always lower than or equal to presence rate.
This is the metric the gaps are built on. A buying question where the engine named you but recommended someone else is a gap; a question where it recommended you is not. The judgement carries its own version, mentions-v4 on the day, and rows judged under an older version are excluded and counted.
Admissibility and the refusal to report
Every sample is graded before it is counted. A provider failure is an error. A sample where the engine did not search, or where grounding is unknown, is ungrounded. A sample that searched but returned fewer than one citation, fewer than 500 characters, or named no product at all is thin. Everything else is grounded, and only grounded samples feed the metrics.
The disclosure block with every metric set states total samples, admissible samples, the count excluded under each grade and under each other judgement version, the prompt count, the window and every version string. A segment with no admissible samples reports usable false and no metrics at all, because a zero would turn a failed measurement into a finding.
- A rate over 5 of 8 samples is a different claim from a rate over 8 of 8; the block shows which.
- Numbers stamped with different versions are not comparable; the versions are in the block.
- Metrics filtered by topic use only the prompts carrying that topic.
FAQ
Why do I see numbers for ChatGPT but none for one of the other engines?
That engine's segment had no admissible samples in the window, usually because its samples failed or did not search. It is reported as usable false rather than as zero visibility. Check the run's Answers view for the grade reasons.
Why is recommendation rate lower than presence rate?
By definition. Recommendation requires the engine to put you forward, not just name you. It is always lower than or equal to presence.
Can I compare my share of voice with a competitor's tool score?
No. Every vendor's score is a sample of its own prompt set on its own engines. Our numbers compare with themselves across time when the versions and segment match. The methodology page has the full rule.
Related help pages
Answer Radar
Reading the Mean Position Metric
What mean ordinal position measures, why it comes from the answer text rather than the model's ranking, why absence is not zero, and why it is never subtracted.
Answer Radar
Segments and Topics: How Results Are Grouped
Why every metric is reported per segment and never averaged across engines, what the five axes are, what the instrument token is, and how topics group prompts.
Answer Radar
The Audit Run and What It Costs in Credits
What Run audit does: the estimate, the reservation, the fan-out per question per engine per sample, the run states, and why one credit is one measured sample.
Answer Radar
What Answer Radar Does Not Do
The honest list: no prompt volume, no traffic attribution, no causal claims, no consumer-app reproduction, no written briefs, no publishing, no model ids,.