How each figure is calculated

Every headline figure comes from the same set of answers: the answers in the checks you selected, on the engines you selected, with branded prompts and failed calls left out. This page gives the formula for each figure and one worked example.

Demo data. Every example below uses the demo project, a fictional brand (Fernhill Garden Rooms) with synthetic answers. The numbers are what the app works out from that demo data. They are not real measurements.

At a glance

A brand counts once per answer, however many times the answer names it.

The demo example

The worked examples use one engine, so every number is a plain count: the demo project's latest run, on ChatGPT only, with branded prompts left out. That is 28 prompts asked 3 times each, so 84 answers.

Visibility

Definition. The share of AI answers that name your brand.

Formula. Visibility = answers that name your brand / all answers.

Demo example. 58 of the 84 answers name Fernhill Garden Rooms.

58 / 84 = 0.690, shown as 69%.

Cited as a source

Definition. The share of answers that link to a page on your website. An answer can link to your site without naming you, so this is counted separately from Visibility.

Formula. Cited as a source = answers that link to a page on your website / all answers.

Demo example. 9 of the 84 answers link to a page on fernhill.example.

9 / 84 = 0.107, shown as 11%. One of those 9 answers links to the site without naming the brand, so the tile also notes "1 cited you without naming you".

Share of voice

Definition. Your mentions as a share of all mentions of the brands you track (you and your competitors). A brand you do not track does not count.

Formula. Share of voice = mentions of your brand / mentions of all tracked brands.

Demo example. Across the 84 answers, the tracked brands are mentioned 184 times:

58 / 184 = 0.315, shown as 32%.

Position

Definition. When you are named, your average place among the tracked brands in the answer. The first tracked brand named is 1, so lower is better. Answers that do not name you are left out.

Formula. Position = sum of your place in each answer that names you / answers that name you.

Demo example. Fernhill is named in 58 answers, and its places in those answers add up to 105.

105 / 58 = 1.81, shown as #1.8.

Named first

Definition. The share of all answers where you are the first tracked brand named. It is out of all answers, so it can never be higher than Visibility. The Position tile's hover shows it as a count; reports and the Competitors page show it as a percentage.

Formula. Named first = answers that name you first / all answers.

Demo example. Fernhill is named first in 28 of the 84 answers.

28 / 84 = 0.333, shown as 33%. The Position tile's hover shows "Named first in 28 of 84 answers."

Sentiment

Definition. How answers describe you when they name you. An AI model scores each mention of you from 0 (very negative) to 100 (very positive), with 50 as neutral. 60 and above counts as positive and 40 and below as negative.

Formula. Sentiment = mentions of you scored 60 or more / mentions of you that were scored. The Sentiment score on the hover is the average of all the scores.

Demo example. All 58 mentions were scored: 44 positive, 14 neutral and 0 negative.

44 / 58 = 0.759, shown as 76% positive. The 58 scores add up to 3,964, and 3,964 / 58 = 68.3, shown as Sentiment score 68/100.

Visibility score

Definition. A score from 0 to 100 for how often and how high up you are named. 100 means every answer names you first.

Formula. Each answer earns points: 0.9 to the power of (your position minus 1) when it names you, and 0 when it does not. Named first is 1.00, second 0.90, third 0.81, fourth 0.73. Then:

1. Average the points across the repeats of each prompt on each engine.
2. Average those across the prompts on each engine, every prompt counting equally.
3. Average those across engines, every engine counting equally.
4. Multiply by 100 and round.

Demo example, step 1. On ChatGPT, the prompt "Best garden office for working from home all year round" was asked 3 times:

Step 2. The average over all 28 prompts on ChatGPT is 0.637, so the score on ChatGPT alone is 64.

Step 3. With every engine in the demo selected, each engine counts once:

0.427 times 100 = 42.7, shown as 43. More on what the score does and does not tell you: the visibility score.

More than one engine

With several engines selected, Visibility, Cited as a source, Share of voice, Position, Named first and the sentiment score pool the answers from all of them. Each prompt counts once on each engine, however many answers it has, so a prompt asked more often does not count for more. The Sentiment tile's % positive is not weighted: it is a plain count of the scored mentions of you, as above.

Demo example. In the demo's latest run, one Claude answer timed out, so one prompt has 2 answers on Claude instead of 3. Each of those 2 answers counts as half of that prompt, and every other answer as a third. Across all 7 engines, 269 of 587 answers name Fernhill. A plain count gives 269 / 587 = 45.8%; with each prompt counting once on each engine it is 45.7%. Both are shown as 46%. The hover shows the plain count, "269 of 587 answers".

When every prompt has the same number of answers on every engine, as in the ChatGPT example above, the two are exactly the same.

Ranges

Visibility, Cited as a source, Share of voice and Position are also worked out for each repeat on its own (repeat 1, repeat 2, repeat 3), and the lowest and highest of those are each figure's range, shown on its tile's hover. Daily checks have one repeat each, so their range runs from check to check instead.

Demo example. On ChatGPT, the 3 repeats name Fernhill in 18, 21 and 19 of their 28 answers: 64%, 75% and 68%. Visibility is 69%, with a range of 64% to 75%.

Who decides what counts

Whether your brand is named, and in what order, is decided by exact text matching on your brand name and aliases, never by an AI model. Only the sentiment scores come from an AI model.

See also what each figure means, how accurate the numbers are and the glossary.