Methodology

How does Visiby measure AI visibility?

Updated

Short answer: Visiby measures AI visibility by asking AI engines a fixed set of buyer prompts on a schedule, saving every answer with the sources it cites, and counting how often your brand is named (a mention) and how often a page on your own site is linked (a citation). It covers ChatGPT, Perplexity and Google AI Overviews on every plan, and Claude on Claude Pro.

By default, each run asks every tracked prompt once on each engine in your plan, and runs repeat weekly on Lite and Starter, daily on Pro, Claude Pro and Agency. Only completed answers count, prompts that contain your brand name are left out of the rates, and the results are separate metrics, not one blended score. A run is a dated sample: it shows what appeared in those answers, not why.

See what ChatGPT says about your brand today (ChatGPT only, three buyer questions).

What does Visiby count?

The unit behind every number is one answer: one prompt, on one engine, in one run. From each answer Visiby records two things, and keeps them apart, because an answer can name you without linking to you, or link to you without naming you. For the wider idea, see what AI visibility means.

The metrics Visiby reports and how each is counted

MetricWhat is countedRead it as
Mention rateAnswers that name your brand ÷ completed answersHow often you come up
Owned citation rateAnswers that link a page on your domain ÷ completed answersHow often your site is a source
Share of voiceYour mentions ÷ all brand mentions in the runYour share; all brands add up to 100%
Mention positionThe order brands are first named in an answer. On Pro, Claude Pro and Agency.#2 of 4 means one brand came first
Matched-cohort changeThe same prompts on the same engines, in both runsChange in percentage points

The dashboard does not show a separate recommendation measure today; mention position is the closest measure. Full definitions are in how Visiby calculates AI visibility and share of voice and citation rate, defined.

Which prompts does Visiby ask?

At setup you give at least three priority keywords or topics, and Visiby writes a prompt set up to your plan's limit: 25 on Lite, 50 on Starter, 150 on Pro, 150 on Claude Pro and 500 on Agency. You can upload your own prompts as a CSV file instead. See how setup works.

Comparisons between two runs use only prompts whose wording is the same in both, so a reworded prompt does not show up as a change. Prompts that contain your brand name are useful for checking how AI engines describe you, but they are left out of the rates.

How are the answers collected?

  • ChatGPT: asked with web search turned on, so the answer comes with the sources it used.
  • Perplexity: the answer and the sources Perplexity cites. Perplexity answers are not location-targeted.
  • Google AI Overviews and Google AI Mode: the prompt is run as a Google search, shortened to keyword form, and Visiby reads the AI answer and the pages it references.
  • Claude: asked with web search turned on, and asked to answer as a user in your country.

Which engines a workspace has depends on the plan: ChatGPT, Perplexity and Google AI Overviews on every plan, and Claude on Claude Pro. Claude and Google AI Mode can be added to a plan that does not include them, as a paid add-on priced on request. The country you chose at setup is applied to ChatGPT, Claude and the Google searches. Step-by-step detail is in how Visiby tracks citations in AI answers, and the full list is in which AI engines Visiby monitors.

How often are the prompts asked?

By default, each run asks every tracked prompt once on each engine in your plan. Runs repeat weekly on Lite and Starter, daily on Pro, Claude Pro and Agency, and each run is kept as a dated snapshot you can select and compare with the run before it.

When an engine returns an error or an empty answer, Visiby retries where it can. An answer that still fails is left out of both sides of every rate, so an outage shrinks the sample instead of counting as a drop. See how often Visiby updates data.

How are mentions and citations decided?

  • Mention: your brand name, a short form of it, or a product name you gave at setup appears as a whole word in the answer text. It is a fixed text match, not an AI judgement, and every tracked brand is checked the same way.
  • Citation: a cited source is on your domain or a real subdomain of it. A different domain that merely contains your brand name does not count.
  • Other sources: every other cited page is labelled as a tracked competitor’s or as another source, such as a review site, forum or publisher.

What does a Visiby number mean, and what does it not?

A 40% mention rate means your brand was named in 40% of the completed answers in that run, after prompts with your brand name were left out. It is a measure of those prompts, on those engines, on that date.

  • It is not your share of all AI traffic or of every question buyers ask.
  • It is not what every user sees: someone signed in, with their own history or location, can get a different answer from the one Visiby saved.
  • It is not a cause: a number shows what appeared in the answers, not why an engine chose a source.
  • “—” means there was nothing to measure, such as no eligible answers. 0% means you were measured and did not appear.

How big a change is a real change?

AI engines write each answer fresh, so the same prompt can name different brands from one run to the next. The size of the sample decides how much one changed answer moves a rate.

A made-up example: how sample size changes what one answer is worth

Named you (mention rate)One answer moves it by95% range (Wilson)
4 of 10 answers (40%)10 pointsabout 17% to 69%
30 of 75 answers (40%)about 1.3 pointsabout 30% to 51%
40 of 100 answers (40%)1 pointabout 31% to 50%

The ranges assume the answers are alike: same wording, same engine, same setup. Read a change across several runs and per engine, not from one run. Why answers move, and how to test any tracker, is covered in are AI visibility tools accurate?

In the dashboard, hover over a change to see how many answers were matched. Trend charts start once two runs have data. See also why answers change from run to run.

Where does the method stop?

  • It measures the prompts you track, not every question buyers ask, so the prompt set should match how your buyers ask.
  • It does not reproduce signed-in or personalised answers.
  • Google AI Overviews and AI Mode are checked with the prompt in keyword form, which is closer to a Google search than to a chat question.
  • Perplexity answers are not location-targeted.
  • Visiby does not track Gemini or Microsoft Copilot.

To check our numbers yourself, open the saved answers and their sources in the Sources view, or export a view as CSV. The other data Visiby reads is listed in what data sources Visiby uses, and what it connects to in Visiby integrations. Plan limits are on pricing. To see the method on your own brand first, run the free check (ChatGPT only). It is a one-off check, not tracking over time; see how the free report works.

Frequently asked questions

How many times does Visiby ask each prompt?

By default, once on each AI engine in your plan, in every run. Runs repeat weekly on Lite and Starter, daily on Pro, Claude Pro and Agency, so a trend is built from many runs rather than from one answer.

Does Visiby give one AI visibility score?

No. It reports separate metrics (mention rate, owned citation rate, share of voice, mention position and matched-cohort change), each with its own numerator and denominator, so every number can be traced back to the answers behind it. Mention position is on Pro, Claude Pro and Agency.

Why are prompts that contain my brand name left out?

A question that names your brand almost always gets an answer that names you, so counting it would raise your rates without showing whether AI engines bring you up on their own. Those prompts are left out of the rates.

Which AI engines does the method cover?

Visiby covers ChatGPT, Perplexity and Google AI Overviews on every plan, and Claude on Claude Pro. Claude and Google AI Mode can be added to a plan that does not include them, as a paid add-on priced on request. Visiby does not track Gemini or Microsoft Copilot.

Can I see the answers behind a number?

Yes. Every run saves the full answer and the sources it cites. The Sources view lists every cited page and domain, and most dashboard views have an Export CSV button.

Related help