Visibility, average position, share of voice and sentiment explained

The four numbers at a glance

Visibility counts how often AI mentions you, Average position measures how high, Share of voice sets you against tracked competitors, and Sentiment rates the tone. Read the four together: alone, each one has a blind spot the others cover.

Two supporting numbers sit alongside these four: citation share, which counts the domains AI links, and the sample line under every figure. Any percentage you cannot multiply back to a count of answers is a number waiting to mislead you.

What Visibility measures

The share of answers that mention your brand. Prompts you mark as more important count for more.

An answer either names your brand or it does not, so Visibility is the bluntest of the four and the one a manager reads first. It is also the one that moves when something real happens: a page gets crawled, a review site enters the sources, a competitor drops out of the answer entirely.

What it cannot tell you: whether the mention recommended you or passed over you (Average position), whether it was praise or a grudging footnote (Sentiment), who else was in the answer (Share of voice), and whether anyone clicked anything after. A blended Visibility figure also hides which prompt is dragging: the total can rise while your core category prompt stays unmentioned for a month.

What Average position measures

How high your brand appears in the lists AI gives, on average. 1 is the top, so lower is better.

This is what separates “mentioned” from “recommended first”, which is the actual difference between a footnote and a shortlist. Buyers reading an answer rarely get to the fifth name.

What it cannot tell you: anything about the answers where you did not appear, which is Visibility's job, and nothing about the distance between you and the brand above you (a gap of one place and a gap of four both read as “second”). It also swings on few answers, because it only has placements to average. Read its sample line before calling a movement a trend.

What Share of voice measures

Your brand's mentions as a share of all tracked brands and competitors in the same answers.

The denominator here is tracked mentions, not answers. One answer can contribute several: a list of four tracked brands gives the set four mentions and you one of them. That makes Share of voice a comparison metric rather than a coverage metric.

What it cannot tell you: why the share moved. It is zero-sum, so it climbs when a rival falls out of the answers while your own mentions hold still, and it falls when a competitor is having a good month and you are exactly where you were. It only counts the brands you chose to track, and like Visibility it carries no information about tone or rank.

What Sentiment measures

How positively answers talk about your brand, from 0 to 100, where 50 is neutral.

Sentiment is computed over mentions rather than answers, so its sample line keeps the real unit: based on 174 mentions. It scores the language around your brand in the answer, which is the only one of the four that reacts to what is actually said about you.

advisible shows the interval on perception scores such as Sentiment and keeps the score hidden until the sample is large enough to carry it, which is why a brand-new workspace shows no number at all rather than a confident 40 from four answers.

What it cannot tell you: what was said. A flat 50 can be twelve neutral mentions, or six glowing and six scathing cancelling each other out. It also scores the answer's wording, not customer satisfaction, so a complaint AI picked up from one review site will dent it without any of your customers changing their mind. Read the mentions behind a Sentiment move before you report one.

What citation share adds

Citation share is your domain's share of all the sources AI links in the answers you track. It is the same arithmetic as Share of voice, applied to cited links instead of brand mentions.

Visibility says you were in the answer. Citation share says your page was the source behind it, which is the number that matters to whoever owns the content roadmap. It is also the metric with the wildest spread across engines: Perplexity cites by default, while plain ChatGPT and Claude chats often link nothing at all, so a blended figure mixes places that cite with places that do not.

What it cannot tell you: whether you were cited as the answer or as the cautionary example. A page can be linked in an answer that tells buyers to avoid you, and the share count looks identical. That judgement needs the sentence around the link, which is why the raw answers stay one click from the chart.

One brand, one week: reading all four together

Invented scenario: Northwind CRM, 16 prompts on 5 engines for one week, 560 answers, four tracked brands. Here is what the four numbers said together, and what a careful reader concludes.

The plain read: you were mentioned in more answers, you sat in the same spot when you were listed, and your slice of tracked mentions grew. That last gain needs a second look, because a rival's mentions halved this week after it dropped out of the comparison prompts. Your share rose without one extra mention of your own.

Sentiment slipped four points, and the intervals overlap, so there is no movement to explain. Two answers do repeat a pricing complaint, which is worth reading for its own sake: the number did not find a trend, but it pointed at two answers that did.

Citation share stayed put, which tempers the Visibility win. More answers mentioned you, none of them started sourcing you. None of the four numbers alone would have caught all of that.

What to report to your boss

Visibility and Share of voice for the trend, Average position and Sentiment for quality, citation share when the question is whether AI sources you. Same window on both sides, a sample line under every figure, and two raw answers attached so the reader can judge the parser.

1. Headline. Visibility this 30 days against the previous 30, with “based on N answers” on the line. One number, one comparison, one caveat.
2. Quality. Average position and Sentiment with their intervals, and one sentence on what moved. Say plainly when the ranges overlap and nothing moved.
3. Sources. Citation share plus the domains AI cited most for your prompts, so the content owner knows which external pages are doing the work.
4. Evidence. Paste two answers verbatim: the best one and the one that got you wrong. A number with a real answer under it survives a skeptical room; a bare chart does not.

Leave out the week's smallest wiggles. A four-point swing inside two overlapping intervals is a question for the answers, not a slide, and the arithmetic for telling the two apart is in how many answers a number needs. The prompt set behind the report deserves the same care, which is choosing prompts to track.