Compare AI Visibility With Competitors

Choose a business question before collecting prompts. A B2B software team might want to know whether AI answers include it when a small marketing team builds a vendor shortlist. A local service business might care about recommendations for a particular service area.

Define the audience, market, language and buying task. Then use questions that express that task:

• “Which AI visibility tools suit a small B2B marketing team?” tests a category shortlist.
• “What should a two-person SEO team check before choosing an AI monitoring tool?” tests evaluation criteria.
• “Which tools retain the answers and citations behind an AI visibility score?” tests a specific requirement.

These are illustrative questions, not measured search-volume claims. Start with evidence from your own search queries, customer conversations and product use cases. Avoid generating hundreds of variations solely to make the sample count larger.

Keep unbranded discovery questions separate from branded comparisons such as “Brand A versus Brand B.” A question that already names a brand gives it a different opportunity to appear. Mixing both question types can make a brand look strong in discovery when you are mostly measuring recognition.

Use brands that a buyer could plausibly choose for the same task. Include relevant competitors that recur in the retained answers, even if they were absent from your initial shortlist.

Once you have a comparison set, preserve it for the period you are reviewing. If you add a brand later, either recalculate both periods using the same set or start a new comparison. Relative shares can move simply because the denominator changes.

Check identity as well as spelling. Similar company names, product aliases and parent brands can lead to incorrect matches. A name in an answer should be tied to the intended entity before it enters a competitor report; a guessed official domain can create a false citation match.

For each question, keep the language, market context, provider, requested model, prompt version and available search or grounding settings consistent. Record dates and configuration changes.

Report providers separately before creating a combined view. If one brand is strong in your Perplexity API samples and absent from your OpenAI API samples, the difference is useful information. An average can hide it.

BrandPresence currently measures supported OpenAI, Perplexity and Google Gemini APIs. These observations describe the API requests made; they are not a report of every personalized consumer chat experience.

Repeat comparable observations. The April 2026 research preprint Don't Measure Once: Measuring Visibility in AI Search (GEO) examines variability across runs, prompts and time and argues for repeated measurement. It does not establish a universal sample count for every business.

A brand mention rate asks how often a brand appears in the successful answers you measured. Each brand can appear in the same answer, so the rates across brands can add up to more than 100%.

For a worked example, suppose one fixed question set and provider configuration produces 15 requested samples: 12 successful answers and three failures. Count each brand at most once per successful answer. The 12 answers contain:

• your brand in three answers
• Rival A in six answers
• Rival B in four answers
• other brands in two answers

This is hypothetical data, not a BrandPresence customer result or industry benchmark.

Your mention rate is 3 divided by 12 = 25%. Rival A's is 6 divided by 12 = 50%, and Rival B's is 4 divided by 12 = 33.3%. Report the three failed requests separately; they are not confirmed brand omissions.

A mention-based share of voice needs a second, explicit definition. One usable convention is a brand's answer-level mention occurrences divided by the occurrences of all brands in a fixed comparison set.

For the three named brands above, that denominator is 3 + 6 + 4 = 13. Your share is 3 divided by 13 = 23.1%. If you include the two other-brand occurrences, the denominator becomes 15 and your share becomes 20%, even though your brand appeared in exactly the same three answers.

Label the included brands and counting rule. Do not compare this figure directly with another tool's score unless its prompt set, coverage and calculation match. This formula is an interpretation method for the example, not a claim that every BrandPresence screen or vendor uses that convention.

Return to the individual answers after calculating the summary. Look for commercially relevant questions where a competitor appears repeatedly while your brand is absent.

For each recurring gap, retain:

• the exact question and measurement conditions
• which successful samples include each brand
• the role in which each brand appears, such as a shortlist option or passing example
• links returned with the answer and the domains they belong to
• whether your own domain is cited, independently of whether your name appears

A citation to a competitor's pricing page and a citation to an independent comparison are different leads. The first might suggest reviewing whether your equivalent pricing information is clear and accessible. The second might identify a relevant third-party source worth examining for accuracy and inclusion criteria.

Neither observation proves that copying a page or obtaining a mention on that source will change the next answer. Keep the source trail as evidence for investigation rather than a guaranteed ranking recipe.

Suppose rivals repeatedly appear for a question about tools that retain answer evidence. Inspect your relevant page: does it clearly explain what is retained, show an example, and state the limits? Check whether the page is accessible and whether its claims are supported.

If that explanation is missing, improving it is a concrete response to the gap. If the page already answers the question well, more copy may add little. Review the recurring source ecosystem and whether the answer is confusing your brand with another entity before expanding the site.

Google's people-first content guidance emphasizes original information, clear sourcing and substantial value. Apply that standard to the page you change: give the buyer something useful to understand or verify.

Record the change date, then compare a later period using the same question set and measurement rules. Show model changes, failures and provider coverage alongside the result. A rise after an edit is an observation; the sequence alone does not establish that the edit caused it.

A useful weekly review should let you answer three questions: where do we recur, where do relevant competitors recur instead, and which gap has enough evidence to justify attention?

Use mentions to assess brand presence, citations to examine source visibility, and referral or conversion data to assess recorded visits and outcomes. Those are related views of discovery, not interchangeable proof of revenue.

Start with an AI visibility audit to establish your baseline. If you are selecting software to repeat the comparison, use the AI visibility tools evaluation guide to check coverage, retained evidence and workflow fit. For the broader measurement framework, read how to measure AI visibility.