The four metrics of AI visibility, defined precisely

Vendors in this category use the same four words for slightly different things, which makes numbers hard to compare between tools. Here is exactly what we mean by each, including the arithmetic, so you can check our work or reproduce it.

1. Visibility Score

The share of tracked answers in which your brand is mentioned at all.

visibility = answers_mentioning_you / total_answers_generated

The denominator is prompts multiplied by engines multiplied by runs. Ten prompts across six engines run daily is 60 answers a day, 1,800 a month. A single mention in 1,800 answers is noise; a stable 34% is a signal. This is why sample size and cadence are not a pricing detail — they are the difference between a measurement and a guess.

The failure mode

Visibility says nothing about competitive position. Every brand in a category can gain visibility at once as engines get better at answering the topic. Rising visibility with flat revenue usually means the whole category got more visible, not you.

2. Share of Voice

Of every brand mention across the tracked answers, the proportion that were yours.

share_of_voice = your_mentions / all_brand_mentions_in_same_answers

This is the competitive number, and it is the one that moves when a rival ships a good comparison page. It is also the one that catches a specific and common problem: being mentioned reliably, but always fourth in a list of five.

The failure mode

Share of Voice is sensitive to your competitor set. Track two competitors and you will look dominant. Define the set as the brands the engines actually name in your answers, not the ones on your internal battlecard — the model has its own opinion about who your competitors are, and that opinion is the one buyers see.

3. Average Position

When you are mentioned, the mean ordinal position of your first mention among the brands named in that answer.

avg_position = mean(index_of_your_first_mention_among_brands)

Position 1.4 means you are usually named first or second. Position 4.6 means you are usually the also-ran at the end of the list. Because users routinely act on the first suggestion, the gap between 1.4 and 2.4 is worth considerably more than the gap between 4.6 and 5.6.

The failure mode

It is computed only over answers where you appear, so it improves when you disappear from the answers where you were ranked last. A rising Average Position alongside a falling Visibility Score is not good news; it is survivorship bias.

4. Sentiment

The classification of each mention as positive, neutral or negative, based on the clause the model wrote about you — not on the overall tone of the answer.

Scoping matters here. An answer can be broadly enthusiastic about the category while describing you as 'the budget option with a steeper learning curve'. That mention is negative for you regardless of the paragraph's mood, and any sentiment score computed over the whole answer will miss it.

The failure mode

Sentiment is the metric most likely to be repeating something a model absorbed from a two-year-old review thread. When it dips, look for the source before you look at your product. Frequently the fix is a corroborating page that gives the engine something more current to quote.

Reading all four together

• Visibility up, Share of Voice down: the category is growing faster than you are. Expand prompt coverage.
• Visibility flat, Average Position improving: your positioning copy is landing. Keep going.
• Visibility up, Sentiment down: you are getting mentioned as the caveat. Find the source being quoted.
• All four flat for six weeks: the engines have nothing new to read about you. That is a content problem, not a measurement problem.

Fenn reports all four per engine rather than blended, because a blended score hides the engine you are losing. You can pull the same numbers into your own agent with the visibility_report MCP tool and skip the dashboard entirely.