Buying guide

Check which answers count in an AI visibility percentage

Two percentages can describe different things even when both are labeled visibility. Before comparing tools or reporting a change, ask for the records, counting rule and denominator behind the number.

A marketing team sees its brand at 30% in one report and 20% in another. That difference could reflect different results. It could also reflect two calculations applied to the same answers: the share of answers that name the brand, and the brand's share of recorded mentions. Buying a tool on the larger percentage would miss the distinction.

Start a product evaluation with one reconstructable result. The vendor should be able to explain what was counted, which records were eligible and how the displayed percentage follows from those records. Apply that requirement to a Gist demonstration as well as to competing tools.

Answers and mentions produce different denominators

Peec's documentation defines visibility as the percentage of responses mentioning a brand. Its share-of-voice documentation instead divides the brand's mentions by total brand mentions and describes comparison with tracked competitors. Those are different units. The wording alone does not settle every counting detail, such as how repeated names within one answer are handled. Peec visibility, Peec share of voice

Profound's Reports v2 documentation also separates visibility, share of voice and average position. It describes visibility as the share of answers containing the asset, share of voice as the share of all mentions, and position as the average rank when mentioned. The documented grouping options include model, prompt and region. The navigation labels these v2 reports Beta; documentation is not proof of an entitlement in a particular subscription. Profound Visibility v2

These definitions establish what to investigate. They do not establish that two products use identical brand matching, sampling, repetition or exclusion rules.

Follow the arithmetic through one fictional ledger

Suppose a saved set contains 10 completed answers. Your brand appears in 3 of them. An analyst records 5 mentions of your brand and 25 mentions across the selected brands, counting repeated mentions in this particular example.

Scroll horizontally to read all columns. Use the arrow keys when the table is focused.

CalculationNumeratorDenominatorResult
Share of answers naming your brand3 answers10 completed answers30%
Share of recorded brand mentions5 mentions25 recorded mentions20%

Both calculations can be correct. Neither percentage states that the brand was recommended, that its website was cited or that anyone visited the site. Those require separate evidence from the answer, links or analytics.

A fictional set gives 30 percent answer presence and 20 percent mention share because the denominators differ.

Original arithmetic example, not a vendor result. Ten completed answers contain three answers naming the brand, five recorded mentions of that brand and 25 recorded mentions in total. Repeated mentions count in this illustration; a vendor's rule must be verified separately.

Now keep every answer unchanged but include another competitor's 10 recorded mentions in the comparison. The brand still appears in 3 of 10 answers, so answer presence stays 30%. Its mention share becomes 5 of 35, about 14.3%. The changed comparison population explains that decline; the example contains no change in the underlying answers.

This is a reason to preserve the competitor set alongside a chart. It is not a claim that adding a competitor triggers that exact behavior in either product.

Require enough detail to reproduce the displayed result

Use a saved result from the vendor's demonstration to work through these questions. A blank field should remain an unresolved evaluation item rather than become an assumed capability.

Scroll horizontally to read all columns. Use the arrow keys when the table is focused.

What to establishEvidence to request
Eligible populationExact questions, engine or product surface, collection dates, region and included answer IDs
Counting unitOnce per answer, each occurrence, named recommendation, citation or another documented event
Entity matchingThe names and aliases included, with ambiguous matches available for review
Failure handlingAttempted, completed, unavailable and valid-empty records shown separately
WeightingWhether each answer, question or engine contributes equally, and how missing records are handled
Comparison populationWhich competitors are included and whether changing that set recalculates historical figures

For example, if one question has many more completed answers than another, pooling every answer gives the heavily sampled question more influence. An average of question-level percentages can answer a different question. Ask the vendor to show the calculation it uses rather than select whichever method produces the preferred score.

A collection failure is also different from a completed answer that does not name the brand. If a provider was unavailable, counting that attempt as a negative mention result would change the meaning of the percentage. Record the missing coverage explicitly before drawing a comparison.

Choose a score you can explain to someone else

At the end of the demonstration, write one sentence using the underlying unit: for example, “The brand appeared in 3 of these 10 completed answers.” Then add the engine, dates and question set from the actual report. If the sentence cannot be completed from available evidence, retain that gap in the buying decision.

A tool's value can include convenience, workflow and coverage. A larger percentage by itself is not evidence of better measurement. Once the population and counting rules are clear, the team can use the existing weekly-report guide to report changes without silently changing what the number means.

Sources and scope

Gist publishes this guide and sells AI visibility software. The evaluation criteria apply equally to Gist and other vendors. Public documentation was checked November 18, 2026. We did not test vendor accounts or compare product accuracy, prices or commercial entitlements. The arithmetic and evaluation worksheet are original examples, not collected AI answers.