How to track share of voice in AI answers

Share of voice is an old marketing idea: of all the attention in your market, how much goes to you? In advertising it meant share of ad spend; in SEO it meant share of rankings or clicks. In AI search it means something simpler and more direct: when an assistant answers your buyers' questions, how much of the recommending is about you?

This post explains how to calculate it properly, the mistakes that make it meaningless, and how to turn it into decisions. For the definition in one paragraph, see our glossary entry on share of voice in AI.

Share of voice vs visibility score

The two metrics answer different questions, and you need both.

• Visibility score: in what share of answers are you mentioned at all? It is absolute. 30% means three in ten answers name you. See our guide to the AI visibility score.
• Share of voice: of all the brand mentions in those answers, what share is yours? It is relative. It tells you whether you or a competitor is the default recommendation.

They can move in opposite directions. If a new competitor starts appearing in answers, your visibility can stay flat while your share of voice drops, because the total pool of mentions grew. That is exactly the kind of change you want to catch early.

Why to count mentions, not rankings

It is tempting to weight share of voice by position, so being named first counts for more. Be careful. SparkToro and Gumshoe ran 12 prompts through ChatGPT, Claude and Google's AI 2,961 times and found that two answers listed brands in the same order roughly once in 1,000 runs. Order is mostly noise. Whether a brand appears at all, counted across many runs, was the measure the study found meaningful.

So build your core share of voice on mention counts. If you want a position signal, report average position separately and treat it as rough.

How to calculate AI share of voice, step by step

1. Fix your prompt set. Choose 20 to 50 unbranded buyer questions for your category, grouped by topic or buying stage. Do not change the wording between checks.
2. Choose your engines. For example ChatGPT, Gemini and Perplexity. Calculate each separately as well as combined.
3. Run each prompt several times per engine. Three to five runs per prompt is a practical minimum because answers vary.
4. Extract every brand named in every answer. Normalise names so "HubSpot CRM" and "HubSpot" count as one brand.
5. Count mentions per brand. Count a brand once per answer, even if the answer names it three times, so long answers do not distort the result.
6. Divide. Share of voice for a brand = its mentions / total mentions of all brands, times 100.

A worked example

Suppose 30 prompts, run 3 times on 3 engines, give 270 answers. Across them, brands are named 1,000 times in total. The table below shows how the shares come out.

These numbers are invented to show the arithmetic. They are not measurements of any real brand or category.

Reading it: Competitor A is the default answer. You are named in four answers out of ten, which is respectable, but you hold less than half of Competitor A's share. The large "all other brands" slice says the AI's consideration set in this category is wide, so there is room to grow.

Which competitors to include

Start with the competitors you know, then let the answers correct you. Count every brand the AI names, and review the list after the first run. You may find brands you never considered competitors, such as a horizontal tool that AI keeps suggesting for your niche, or a newer entrant that is well covered on review sites. If the AI names them, your buyers hear about them.

How many prompts and runs you need

More data gives a steadier number. The SparkToro study concluded that visibility measured across dozens to hundreds of prompts, run multiple times, is a reasonable metric. A follow-up found that when hundreds of people wrote prompts for the same intent, the average semantic similarity was only 0.081, so cover each topic with several phrasings rather than one "ideal" prompt.

• Fewer than about 50 answers: treat share of voice as a rough indication only.
• A few hundred answers: good enough to compare against competitors and spot big shifts.
• Repeated weekly or daily: good enough to see trends. Judge a change by whether it holds over several runs.

Break it down to make it useful

A single share of voice number is a headline. The decisions come from the breakdowns.

• By engine. Each engine retrieves different sources, so your share can be high in one and low in another. Fix the weakest engine's sources first.
• By prompt group. You may lead on "for agencies" prompts and be absent on "cheapest" prompts. That shows which positioning AI already associates with you.
• By competitor over time. A competitor whose share rises across several runs is doing something visible. Check which new sources cite them.

Citation share of voice

Apply the same calculation to the sources the answers cite. Perplexity says each answer includes numbered citations, and Gemini's grounding with Google Search returns citations linking answer text to source URLs. Count citations by domain and you get a ranking of the sites that shape AI answers in your category. In Google's AI summaries, Pew Research Center found that Wikipedia, YouTube and Reddit together accounted for 15% of cited sources; your own category's list will be more specific and more useful.

Pair the two lists. The domains that cite your top competitor but not you are your highest-value outreach targets.

Common mistakes

• Including branded prompts. Questions that contain your name inflate your share. Keep them separate.
• Changing the prompt set between checks. Then you are comparing two different markets.
• Counting a brand several times in one long answer. Count once per answer.
• Reading one run as a trend. Variation between runs is normal; look for changes that persist.
• Weighting heavily by position. Order is the least stable part of an AI answer.
• Mixing answers with and without web search. They draw on different information. Track them separately if your tool shows the difference.

Turning share of voice into actions

1. Find the prompt groups where a competitor holds a large share and you hold little.
2. List the domains cited in those answers. Check whether you are mentioned on each one, and how.
3. Get onto the most-cited pages with accurate information: comparison articles, directories, review profiles, community threads.
4. Publish or improve your own page for each weak prompt group, so there is a clear source that states what you offer for that use case.
5. Re-run the same prompts and track whether your share in that group rises over the following weeks.

Doing this by hand is possible for a one-off audit. Keeping it up weekly across hundreds of answers is what AI visibility trackers are for. Mentioned's paid plans track share of voice against your competitors, per-prompt results and cited sources, with CSV export; you can start with the free AI visibility check to see your first numbers.

Frequently asked questions

What is a good AI share of voice?

There is no universal benchmark; it depends on how many brands compete in your category and how you build your prompt set. Compare yourself with your named competitors and with your own past results. Being the brand with the largest share in your core prompt group is a strong position.

Should I track share of voice per engine or combined?

Both. The combined number is your headline. The per-engine numbers tell you where to act, because each engine relies on different sources.

How often should I measure it?

Weekly is enough for most brands; daily helps in fast-moving categories or during a campaign. Whatever the cadence, keep the prompts, competitors and engines fixed so results are comparable.