AI Share of Voice: Measure It in ChatGPT and AI Search

In short: AI Share of Voice measures how often a brand appears in a fixed set of relevant AI answers. It is not the same as brand visibility. A useful measurement separates spontaneous mentions, prompted mentions, source links, answer quality, and competitor presence. Start with the questions buyers actually ask, keep the conditions stable, and compare like with like.

The formula, and a comparison example

Share of voice in an AI answer is calculated on a fixed set of queries, repeated on the same models over the same period:

Share of voice = (number of responses mentioning the brand / total responses tested) × 100

Example, on the query « best stock management software », asked 30 times across ChatGPT and Perplexity:

BrandMentionsShare of voiceCited as a source
Brand A18 of 3060 %5 responses
Brand B9 of 3030 %0 responses
Brand C3 of 3010 %0 responses

Brand A leads the share of voice but is only cited as a source in 5 responses: its presence depends on the model's habits, not on its own pages. The comparison is only valid at identical scope: same queries, same models, same dates.

What is AI Share of Voice?

AI Share of Voice is the share of answers in a defined test set that mention a brand. The key phrase is “defined test set.” Without a stable set of questions, a percentage is only a flattering snapshot.

If a brand appears in 18 answers out of 100 comparable answers, its basic AI Share of Voice is 18%. That figure becomes useful only when the same questions, language, market, and response conditions are used for the brand and its competitors.

For the query “AI share of voice,” the direct answer is simple: it is a presence metric. It tells you whether a brand enters the answer. It does not, on its own, prove that the answer is accurate, prominent, or commercially valuable.

AI Share of Voice vs brand visibility: the difference that changes the audit

These terms are often used as if they meant the same thing. They do not.

MeasureWhat it answersWhat it can miss
AI Share of VoiceHow often does the brand appear in the answer set?Whether the mention is correct, useful, or prominent.
Brand visibilityHow visible and recognisable is the brand across the measured conditions?Whether competitors take a larger share of the same answers.
Source citation rateHow often is a page or source linked or named?Brand mentions that appear without a source link.
Answer qualityDoes the answer describe the brand accurately and usefully?How often the brand is absent.

A brand can have a respectable share of mentions because its name is included in a prompt, yet have almost no spontaneous presence when the prompt asks for category recommendations. That is why an audit should keep prompted and unprompted questions apart.

Our position is blunt: a score that mixes those two conditions without a clear label is hard to act on. It can make a weak discovery presence look healthier than it is.

How to measure Share of Voice in ChatGPT and AI search

Use a question universe before you use a dashboard. A practical starting set has four groups:

  • Category questions: “What are the best…?”
  • Problem questions: “How do I solve…?”
  • Comparison questions: “X vs Y” and “alternatives to X.”
  • Brand questions: “What is X?” and “Is X suitable for…?”

Mark each question with its language, country or market, buyer intent, and whether the brand name appears in the wording. This sounds administrative. It prevents the most common error: comparing a branded answer from one run with a generic recommendation answer from another.

Record more than a yes or no

For every answer, capture the presence of the brand, its position when a list exists, any source URL, the competitors mentioned, and whether a material claim is wrong or unsupported. A plain mention count is useful; a count without the underlying answers is not auditable.

Keep ChatGPT, Google AI Overviews, Perplexity, and other answer systems in separate cuts. Their answer formats and retrieval conditions differ. A combined total is helpful for an executive view, but it should never hide an engine-specific drop.

Use the same denominator for every competitor

If you test five competitors, each one must face the same query set under the same conditions. Otherwise, a higher Share of Voice may simply reflect an easier denominator. Where a brand and a competitor are mentioned equally, AI Labs Audit’s current competitive approach deliberately ranks the competitor ahead: the brand being audited already benefits from more exhaustive detection. That rule is less flattering, but it makes the comparison more defensible.

A calculation template that does not hide the problem

IndicatorCalculationWhy it matters
Spontaneous SoVAnswers naming the brand on unbranded questions ÷ unbranded answersMeasures category discovery.
Prompted SoVAnswers naming the brand on branded questions ÷ branded answersMeasures how the system describes a known brand.
Source citation rateAnswers linking or naming a first-party source ÷ all answersShows whether the brand’s own material is used.
Competitive gapLeader’s SoV minus the brand’s SoVTurns a percentage into a concrete gap.

For example, 12 spontaneous mentions in 60 category answers is a 20% spontaneous SoV. If the leader appears in 24 of those same answers, the competitive gap is 20 percentage points. That is a more useful brief than “visibility needs work.”

What our own measurement work changes

Published AI Labs Audit product data currently cites 6,820 analysed AI responses across 104 audits. That is not a universal market benchmark, and we do not use it as one. It is evidence for a more practical point: answer-level review exposes differences that a single visibility score can conceal.

The platform’s public catalogue also covers 50+ AI models, while the Discovery audit scope is 3 AI systems across 30 analysis scenarios. Those numbers describe a measurement setup, not a promise that any brand will be cited. The useful output is the answer trail: which question produced a mention, what the system said, which competitor appeared, and what must be corrected.

For a broader explanation of how sources are retrieved and used in answers, see our definition of retrieval-augmented generation. The important operational lesson is simpler: measure the response itself rather than assuming that a good conventional ranking transfers unchanged into an answer engine.

How to increase AI Share of Voice without chasing a vanity metric

Do not begin by writing more pages. Begin by opening the missing answers.

  • Find recurring questions where competitors are cited and the brand is absent.
  • Check whether the brand has a clear, current page that answers the question without requiring a sales call.
  • Correct false claims and missing entity information before publishing another broad guide.
  • Make useful evidence easy to locate: definitions, named authors, dated research, methods, limitations, and first-party facts.
  • Re-run the same query set after a meaningful change; do not change the questions at the same time.

Structured data can help machines interpret the page, but it is not a citation switch. Use it to clarify what a page, organisation, product, or FAQ is; do not treat it as a shortcut around useful content. Our schema markup glossary entry explains the role in more detail.

From visibility score to conversion analysis

The query “analyze conversions with visibility score in Europe” points to a legitimate next question: does a visibility change affect commercial outcomes? It might, but the relationship has to be measured rather than assumed.

Keep a dated log of changes in Share of Voice, source citations, referral visits from answer engines where available, qualified enquiries, and sales conversations that mention an AI recommendation. Use the same market and period when comparing them. A rising SoV alongside flat qualified demand is a signal to inspect the query set and answer quality, not a reason to declare success.

Share of Voice is a diagnostic instrument. It earns its place in reporting when it leads to a specific decision: protect a strong category, repair a factual gap, create a missing answer, or investigate a competitor’s source advantage.

Questions people ask about AI Share of Voice

What is a good AI Share of Voice?

There is no universal “good” percentage. A 10% share can be strong in a fragmented category and weak in a category with only three serious providers. Compare the brand with the same competitors on the same question set, then watch the trend.

Can I measure Share of Voice in ChatGPT and Google AI Overviews together?

Yes, but report the engine-level figures first. A combined figure is an executive summary; it should not replace the separate evidence because the answer conditions and formats are different.

Does a mention equal a conversion?

No. A mention is an opportunity for discovery. Pair it with answer quality, source citations, referral data where available, and qualified demand before drawing a commercial conclusion.

What should an agency deliver to a client?

A defensible report includes the fixed question set, dates, model or engine conditions, raw answer evidence, competitor comparison, the measurement definitions, and actions tied to the observed gaps. An AEO audit can provide the broader technical and content context around that evidence.

Un exemple chiffré, et comment le lire

Sur une requête comme « meilleur logiciel de gestion de stock » interrogée 30 fois sur ChatGPT et Perplexity :

  • La marque A apparaît dans 18 réponses sur 30 (60 %), dont 5 comme source citée.
  • La marque B apparaît dans 9 réponses sur 30 (30 %), jamais citée en source.
  • La marque C apparaît dans 3 réponses (10 %), toujours en dernière position.

La part de voix se lit en trois temps : le taux de mention (combien de réponses parlent de vous), la position dans la réponse (vous êtes cité en premier ou en dernier), et la citation en source (l'IA renvoie-t-elle vers votre site). Une marque présente dans 60 % des réponses sans jamais être citée en source est visible mais fragile : elle disparaît si le modèle change de source.

La comparaison n'a de sens qu'à périmètre identique : mêmes requêtes, mêmes modèles, mêmes dates.

See all our resources on the subject in the AI visibility audit hub: guides, checklists and methodologies in one place.

About the author

Davy Abderrahman

Founder of

As a specialist in visibility in AI answers (AEO/GEO/LLMO), I help agencies and consultants measure and analyze their clients’ presence on ChatGPT, Claude, Gemini, Perplexity and other AI answer engines.

AEO GEO LLMO AI visibility AI audits
How do your brands appear in AI answers?

Agencies, consultants and in-house teams: compare the answers generated for your own scenarios, languages and selected models. Our infrastructure, your scenarios, your expertise.

Was this article helpful?

- (0 votes)