AI NativeWhat is the mention rate and how is it calculated?
Mention rate is the share of answers, for a given set of prompts and engines, in which your brand was present. An answer counts as a mention if your brand name, a recognised alias, or a domain you own appears anywhere in the answer text or citations. If you ran 30 answers and your brand appeared in 21 of them, your mention rate for that set is 70 percent. Absent answers are the denominator together with present answers; there is no exclusion.
What is the recommendation rate?
Recommendation rate is the share of answers where your brand was put forward as a good choice. It is a subset of the mention rate: you cannot be recommended without being mentioned. An answer is counted as a recommendation when the AI judge classifies it as recommended or recommended_first on the visibility ladder. A passing reference or a neutral list entry does not count. Recommendation rate tells you how often the AI is actively choosing you, not just naming you.
What does the visibility ladder mean?
The visibility ladder is a five-rung outcome scale for individual answers: absent, mentioned, listed_option, recommended, and recommended_first. Each rung reflects a qualitatively different relationship between your brand and the answer. Absent means no presence at all. Mentioned means present but not treated as a choice. Listed means named among options. Recommended means put forward as a good choice. Recommended_first means the top recommendation. The ladder mean for a cell is the average ladder position across all answers in that cell, converted to a 0 to 1 scale. See The visibility ladder for the full breakdown.
What is AI share of voice?
Share of voice measures how much of the brand conversation in AI answers belongs to you relative to your competitors. If five brands appear across a set of answers, and your brand accounts for one-fifth of those appearances, your share is roughly twenty percent. It is a relative metric: your share of voice can fall even if your mention rate holds steady, because a competitor started appearing in the same answers at higher frequency. Share of voice is most useful as a competitive measure rather than an absolute target. See AI share of voice.
What does sentiment score mean?
Sentiment score is the average framing of your brand in the answers where you appear, on a scale from -1 to 1. Positive values indicate net-positive framing (the AI is saying good things about you). Negative values indicate net-negative framing. The four sentiment labels are recommended, neutral, hedged, and negative, where hedged means technically present but surrounded by caveats a buyer would read as discouraging. See Sentiment in AI answers for how the layers work.
What is the accuracy state in a branded answer?
Accuracy state records whether the answer made a checkable numeric claim about your product and, if it did, whether that claim was correct. The four states are accurate, inaccurate, unverified, and not applicable. Inaccurate means the answer stated something that directly contradicts a verified fact in your Brand Truth Studio. The check only runs against facts a strategist has confirmed, so unconfirmed candidates do not trigger an inaccurate verdict. See Accuracy and fact checks.
Why do the numbers change between scans when I have not changed anything?
AI answers are stochastic. The same question posed twice to the same engine can produce a different answer, a different set of citations, and a different brand mention pattern. AI Native handles this by running each prompt multiple times and averaging, but some variation between scans is expected even without any change on your side. A small movement in a rate is normal noise. A sustained directional move across multiple scans is signal worth acting on. See Why numbers vary between scans.
What is the difference between a demo scan result and a live scan result?
Demo results are deterministic and illustrative. They show what the screens look like and how the metrics work, and they cost nothing. Live results reflect what the AI engines actually say about your brand today. Demo numbers are not predictions and should not be treated as a baseline. Your first live scan is the real baseline. The platform always labels which mode a result came from.
What does the opportunity score tell me?
The opportunity score combines value (how much it matters), gap (how far you are from the top), and winnability (how much you can influence the outcome) into a single number that ranks which prompt cells to address first. A high score means a large commercial gap you can plausibly close. A low score means either the gap is already small, the cell is low-value, or the sources shaping the answer are too entrenched to influence easily. See The opportunity score.
What does value mean in the context of a prompt?
Value is a score that reflects the commercial weight of a prompt. It is computed from the head-term search demand the prompt inherits, a multiplier for the commercial intent of the question, a multiplier for the funnel stage the question sits in, and the business weight of the persona it belongs to. A transactional question asked by a high-priority buyer persona at the decision stage of the journey has higher value than an informational awareness question asked by a peripheral persona. See Demand and value scoring.
Can sentiment be positive on one engine and negative on another?
Yes, and this is common. The sentiment breakdown by engine shows you exactly that pattern. Different engines draw on different source sets and have different training data, so the framing around your brand can vary significantly by surface. A brand that is framed neutrally on ChatGPT might be framed with persistent caveats on a different engine. Identifying where the negative framing is concentrated is the first step to addressing it.
What does "negative and inaccurate" mean on the reputation panel?
When an answer is both negatively framed and factually wrong, the platform flags this combination separately. It is the highest-priority reputation issue because you are dealing with a wrong claim and a negative framing in the same answer. Fixing the factual error addresses both problems at once if the error is the cause of the negative framing, making these cells the most cost-effective to fix first.
How do I know if a change I made actually moved the numbers?
Run a new scan after your change and compare it directly against the scan you ran before. The before-and-after comparison in the scan view shows which cells moved and by how much. A single scan difference can include noise; if the movement is consistent across multiple prompts in the same cell and the direction holds on the next scan too, you are looking at a real signal, not variance.