Skip to content
The metrics

The five metrics every brand should track in AI answers

Presence, preference, accuracy, source share, and movement. Five numbers that turn 'how are we doing in AI?' into something you can act on.

7 min read · Updated 2026-07-08

If you track one thing in AI answers, track AI Presence, the metric much of the industry still calls AI Share of Voice. If you track five, you can actually manage your discovery. The five that matter are AI Presence (are you in the answer), recommendation rate (are you the pick), accuracy (is the answer correct), source share (which cited domains are yours), and movement (did it change after you acted). Each answers a different question, and together they tell you not just where you stand but what to do next.

1. AI Presence: are you present

AI Presence, formerly and still widely called AI Share of Voice, is the share of answers about your category that name or recommend you, weighted by how prominently you appear, measured per engine and blended. It is the headline number, the one you move over time. A low AI Presence means buyers rarely hear your name when they ask an assistant about your space.

Because engines are non-deterministic, AI Presence is sampled across several runs per engine, not read from a single answer. Watch the trend over weeks, not the wiggle between two runs.

2. Recommendation rate: are you preferred

Presence is not preference. Recommendation rate is the share of answers where you are the pick or named first, not merely listed. Tracking it separately from AI Presence stops a pile of low-value mentions from masking the fact that a competitor owns the actual recommendation.

The gap between your AI Presence and your recommendation rate is a to-do list: those are the questions where you are in the answer but losing it.

3. Accuracy: is the answer right

An answer that recommends you but misstates your fee, your coverage, or a guarantee is worse than silence, especially in a regulated category. Accuracy is the share of answers about you that match your approved facts. Every inaccurate answer is both a lost sale and, sometimes, a compliance exposure.

Accuracy is measurable: sample what engines say about you and compare it against a set of approved, source-backed facts. What fails the check is your highest-priority fix, because it is actively working against you.

4. Source share: which sources are yours

Answers cite sources. Source share is how many of the domains an engine cites, for your questions, are yours versus third parties versus competitors. It tells you whether the answer is built on ground you control or ground you need to influence.

This is the lever behind the other four. If the sources shaping an answer are all third-party pages that ignore you, presence and preference will not move until you earn a place among them. Trace the sources and you find the work.

5. Movement: did it change

The only metric that proves the work matters. Movement is the before-and-after: did AI Presence, recommendation rate, or accuracy change after you published or corrected something? Without it, you are optimizing blind and cannot tell a real gain from run-to-run noise.

Tie movement to what you shipped. A recommendation rate that rises the week after you published a comparison page, and holds, is signal. Re-measurement closes the loop and tells you which actions to repeat.

In Ansyra

All five, per prompt and per engine

Ansyra reports AI Presence, recommendation and prominence, accuracy against your approved facts, citation and source share, and before-and-after movement after you publish, so every number points to a next action rather than sitting on a dashboard.

Straight answers

Frequently asked

Isn't AI Presence enough on its own?
It is the best single number, but on its own it hides too much. It cannot tell you whether a mention was a recommendation, whether the answer was accurate, or whether anything you did moved it. The other four turn one score into a diagnosis.
How is AI Presence (AI Share of Voice) calculated?
It is the prominence-weighted share of sampled answers that name or recommend you, computed per engine and blended. Because answers vary run to run, it is sampled across several runs rather than read from one, so it reflects a typical answer, not a lucky or unlucky single response.
Which metric should I fix first?
Accuracy, if any answers are wrong, because a confident wrong answer actively costs you. After that, work the gap between presence and preference on your highest-intent questions, and use source share to find the pages to influence.

See where you stand in AI answers

Run your real questions across ChatGPT, Gemini and Google's AI answers, free for 14 days. No connections required to start; all five AI engines on every plan.