AI share of voice describes a brand's portion of tracked brand appearances within a defined sample of AI answers. It is useful for comparing the brands that appear in the same monitored questions. It does not measure a platform's entire output, every buyer's experience, or a universal position for your brand.
The measure becomes actionable when the prompt set, brand set, collection conditions, and denominator remain visible. This guide establishes the convention used across this library. Start with AI search brand visibility monitoring for the wider system, then use the linked workflows below to calculate and interpret your results.
Use one explicit counting convention
Fix the tracked brands before collecting the comparison. Count each tracked brand at most once in each successful answer. Repeated mentions, headings, and repeated product names do not create extra appearances for that brand in the same answer. Count a valid appearance whether the answer is positive, negative, or neutral; record sentiment or context separately.
Use this formula:
AI share of voice for Brand A =
Brand A's brand-answer appearances
/
All tracked brand-answer appearances, including Brand A's
If there are no tracked brand appearances, share of voice is undefined; report the zero denominator rather than calling every brand's share zero. Successful answers with no tracked brands contribute no brand-answer appearances, but remain part of the sample and of mention-rate denominators. Failed collection attempts are reported separately.
In a purely hypothetical example, ten successful answers contain four appearances for Brand A, three for Brand B, and one for Brand C. Brand A's share is four divided by eight, or 50%. Its mention rate is four divided by ten, or 40%. Those numbers answer different questions. Adding an untracked fourth brand later changes the competitive denominator and requires a new series or a clearly labeled recalculation.
Separate mention, recommendation, and citation
Mention rate is the number of successful sampled answers mentioning a brand divided by all successful sampled answers. It answers how often the brand appeared in that sample. Share of voice answers what portion of tracked brand-answer appearances it received. A brand's share can rise while its mention rate falls if competitors fall faster.
Recommendation requires a separate definition. An answer can list a vendor as an option, advise against it, or recommend it for a specific requirement. Decide which statements qualify and retain the wording. Do not count every appearance as endorsement.
A linked citation is another separate observation. OpenAI's search guidance explains that responses may include citations. A citation can support background information while naming no tracked brand at all. Count what the captured answer actually displays, and inspect the source before assigning a commercial meaning. OpenAI search guidance
Build a fair competitive cohort
Start with buyer decisions, such as understanding a category, comparing approaches, choosing a shortlist, or checking a requirement. Use prompts based on genuine research and customer language. Keep branded comparison prompts separate from unbranded discovery prompts because explicitly naming a competitor changes the task.
Record platform, surface, date, language, locale assumptions, conversation state, and visible search settings. Do not label API outputs as the same experience as a consumer app without evidence. Distinguish Gemini from AI features in Google Search. Google documents AI Overviews and AI Mode as Search features; this is a specific surface boundary for your report. Google's Search AI guidance
Keep the brand set stable and document aliases, product names, subsidiaries, and ambiguous terms. An ordinary word that also names a company needs context-based review. Review uncertain matches manually and preserve the original answer so another analyst can challenge the decision.
Choose the next workflow
| Decision you need to make | Guide |
|---|---|
| Compute appearances and denominators consistently | Calculate AI share of voice |
| Select metrics that answer different business questions | AI visibility metrics |
| Identify buyer questions where competitors have an evidence advantage | Competitor visibility gaps |
| Inspect the pages linked beside competitor claims | Competitor citation analysis |
| Present a repeatable baseline and follow-up | AI visibility benchmark report |
Use LLM brand tracking to organize collection and AI citations and digital PR when the evidence points to a source or editorial problem.
Create a reporting contract
Agree on these fields before anyone presents a trend:
Report purpose and audience:
Prompt set version and buyer-question groups:
Tracked brand set and alias rules:
Platforms, surfaces, languages, and dates:
Collection repetitions and conversation conditions:
Successful answers, no-brand answers, and failed runs:
Brand-answer counting rule:
Mention, recommendation, and citation definitions:
Per-platform counts and SOV denominators:
Any combined-view weighting rule:
Changes since the previous report:
Evidence links, reviewer, and limitations:
Report per platform and question group before combining results. If one group has more prompts or more successful runs, it can dominate an unweighted total. Either show that composition plainly or apply a prespecified weighting scheme and disclose it. Avoid changing weights after seeing which view looks favorable.
Turn gaps into work you can justify
Read representative answers behind the numbers. A competitor's appearance may reflect a relevant capability, clearer documentation, an independent source, or a prompt that favors its market segment. Determine which explanation the evidence supports before creating a content task.
If your product does not fit the requirement, the useful decision may be to exclude that prompt from a future qualified-buyer cohort while preserving the current report's history. If the answer uses outdated facts, follow the correction workflow. If evidence is missing, use the AI visibility audit to prioritize a page or research project.
Recheck the same cohort after changes and describe movement within that sample. More prompts and repeated observations can improve your view of that cohort; they do not automatically make it representative of all platform users. Keep claims proportional, keep denominators visible, and explore BIU's AI visibility approach when planning a repeatable competitive review.