AI Share of Voice: How to Compare Brands Fairly

AI share of voice tells you what proportion of counted brand mentions in selected AI answers belongs to your brand. The percentage becomes useful when you define a stable comparison set and a clear counting rule. This guide explains the formula, works through an example and connects changes in your share to practical content decisions.

Begin your brand comparison

What does AI share of voice mean?

Picture a set of real buying decisions: which solution fits a small agency, which vendor meets a particular requirement and which product suits a specific use case? Share of voice describes the relationship between counted brand mentions in answers to those questions. Its value depends on the brands you include and the answers that enter the calculation.

That makes the metric useful for competitor analysis. You can see whether your brand receives a large or small share within that particular sample. For a marketing decision, a relevant sample is more helpful than an apparent overall score across unrelated topics. An agency and a large retailer may need different comparison sets even when both describe their requirement as project management.

FameScan measures how a brand appears in AI answers to selected buyer questions and connects competitor comparisons with answer and citation evidence. Share of voice is one perspective within that work. Mention rate looks at answers containing your brand; competitive share looks at counted brand mentions. Our guide to measuring AI visibility explains the broader measurement framework.

Choose a useful comparison set

Start with the vendors your customers actually consider. Give each brand a reason for inclusion, such as serving the same use case or company size. Adding famous names without a connection to the decision makes the comparison larger without necessarily making it more useful. A smaller specialist may be the more relevant competitor for a particular segment.

Your own brand belongs in the list. Distinguish direct alternatives from adjacent solutions, too. A question asking for a simple spreadsheet instead of specialist software explores a different decision from a comparison between specialist vendors. Both may be worth observing, but interpret them separately first. Otherwise, an expanded brand list appears more comprehensive while quietly mixing different markets.

Record names, relevant spelling variations and the reason for each change in your own measurement log. Inspect ambiguous terms in the answer text before building a strategy around them. A word can be both a product name and an ordinary industry term. A fair comparison depends on identifying the intended brand. The competitor view supports comparisons centered on selected brands.

Agree on the list with the people who will use the report before comparing periods. This prevents a strong competitor from disappearing simply because removing it makes your own percentage look better. If you later discover a relevant alternative, add it for a documented reason and mark the start of the expanded comparison.

How do I calculate my share of voice?

The basic formula is your brand’s counted mentions divided by counted mentions across all comparison brands, multiplied by 100. The important word is counted. A tool might count each brand once per answer, count individual occurrences or limit repetitions. Results calculated with different methods are therefore not directly comparable. When you build a share of voice AI report, put the counting rule next to the result.

The competitor view and corresponding score category in FameScan count a maximum of three mentions per brand per answer. Those values are added across the usable answers in the relevant run and scope. The denominator is the total for all tracked comparison brands, including your own. The calculated value is rounded to two decimal places. This limit prevents unlimited repetition in a single answer from dominating the share.

Record the view used for every comparison. The separate Insights page calculates its share of voice from the number of answers mentioning each brand within its limited brand list. That list may also include newly discovered brands. Its result is therefore not directly comparable with the capped mention share explained here.

Branded questions are excluded from visibility scoring and retained for sentiment analysis. A question that already names your brand explores a different issue from an open buying decision. When there are no counted mentions across the comparison set, the share cannot be calculated. That is a missing value, which should remain distinct from a measured share of zero percent.

A worked example with four answers

This example is entirely hypothetical. It describes neither a customer measurement nor an actual provider response. We examine four valid answers to questions that do not name your own brand in advance. The comparison set contains your brand A and brands B and C. A appears five times in the first answer, which contributes three counted mentions.

Every detected count for B and C is already within the cap. The denominator is 4 + 4 + 3 = 11. Brand A therefore receives 4 ÷ 11 × 100 = 36.36 percent share of voice. The answer containing none of A, B or C adds nothing to this denominator, but it remains part of the four valid answers in the sample.

A’s mention rate is different: two of four answers, or 50 percent. Both results are correct because they answer different questions. A appears in half the answers and receives just over a third of the mentions counted under this rule. Neither 50 nor 36.36 percent describes the proportion of all AI users who have seen A.

Keep both underlying counts in the finished report. Four counted mentions out of eleven is something a reader can check; a percentage alone conceals the calculation. Add the answer count so nobody reads eleven mentions as eleven answers. With a small sample, the actual wording deserves more attention than the second decimal place.

A failed request is also different from a valid answer without a brand mention. Record errors separately so a gap in collection does not appear as missing brand presence. All four answers in this example are valid; only the fourth contains none of the three comparison brands. Put that distinction beside the formula in your own measurement template.

When the denominator changes

Extend the hypothetical example: answer four also mentioned brand D four times. D was previously outside the comparison set. Including it in the calculation adds three counted mentions under the same cap. The new denominator is 14, while A remains at four. A’s share falls to 28.57 percent even though no answer and no mention of your own brand has changed.

This is a change in the comparison set. It may be a sensible change, but the report should make it visible. Label the original and expanded comparisons separately and include their brand lists. If you recalculate historical data using a new rule, document that step as well. This is a recommendation for your measurement log, not a promise of automatic historical recalculation in the product.

A similar effect occurs within a stable set when competitors receive more mentions. Your own mentions can remain constant while your share falls. Conversely, your share can rise because other brands appear less often. Inspect numerator and denominator first. Then decide whether you are investigating progress for your brand, a stronger competitor or a change in what the comparison includes.

If all brands together have zero counted mentions, substituting a number does not help. Record that no share can be calculated and check whether the answers name any specific vendors. That preserves a potentially useful content finding without turning missing data into a last-place position in a ranking.

What is a good share of voice percentage?

A universal percentage target would be misleading. A share among three closely related vendors differs from a share among many brands with different offerings. The chosen questions also change the starting point. A specialist may appear frequently for a precise requirement and rarely for broad category questions. Both observations may reflect the company’s positioning.

Define success around a relevant gap: appearing as a suitable option for a specific buying decision or correcting a recurring mischaracterization. Relative share then helps you observe that part of the picture. Establish your baseline across repeated, comparable runs and retain the individual results. An unusual run should remain visible rather than disappearing behind a single average.

Also inspect how the answer describes the brand. Repeated criticism can increase mention share without helping your position. A brief, accurate mention may be more useful than a long paragraph ruling out your solution for the use case. Our explanation of the Visibility Score introduces the additional perspectives. Read them alongside share and inspect the underlying answers.

Read segments before the average

An overall value can hide differences between customer groups. First examine a coherent use case under comparable conditions: the same engine, language, target market and stable core of questions. That makes a comparison over time understandable. If you add more broad questions in the next period, the overall share may move even when little has changed for the original buying decisions.

When comparing engines, account for their different retrieval methods. An API request is not an identical replay of a personal chat session. Document the access method and known conditions; treat observations from a consumer interface as a separate series. The prompt and answer view supports inspection of selected questions and their sources. Interpretation begins with the actual answer text.

Citation analysis adds a separate view of the pages used as sources. Keep your samples separate from reports about citations shown on provider surfaces. Bing describes AI Performance as reporting citations across supported experiences and explicitly distinguishes those counts from placement or importance within an answer. Documentation checked October 8, 2026. Use that evidence as another perspective on your website, without adding its citations to the denominator of your own brand comparison.

Turn findings into decisions

A low share is not yet an assignment. First read relevant answers that present another brand as a suitable choice. Which requirement supports that choice? Which source is cited? Then check whether your own public information answers the question precisely. The gap might be a missing feature explanation, an unclear requirement or an inaccurate description.

Consider a hypothetical case: buyers ask about approvals involving external clients. The answer names an alternative and explains its process, while your page only discusses collaboration in general. If your product really supports approvals, explain the actual workflow, participants and relevant boundaries on the appropriate page. If it does not support that need, the question is no reason to make an unsupported capability claim.

Next, write a testable working hypothesis: “We will add verified details about the approval process and observe the same questions again.” Record the change, date and expected observation in your working document. If share later rises, inspect answer texts and the denominator again. A change happening before an increase does not by itself establish that it caused the increase.

The FameScan Action Center provides recommendations derived from rules with supporting evidence. You decide which recommendations fit the facts and implement changes to your content. The one-time Discovery provides a starting point; the dashboard and further recommendations require an active subscription. An active trial of that subscription also qualifies. The next action remains connected to an observation you can inspect.

For Google, the usual technical requirements and helpful, reliable content still apply. Google’s documentation about AI features, checked October 8, 2026, does not guarantee inclusion even when requirements are met. Our GEO checklist supports implementation. Choose a justified change and review it repeatedly rather than rebuilding everything in response to each movement.

Examine your comparison questions

Diagnose your next report

Finish the report with a decision someone can own. The matrix below connects common observations to specific checks. It is an editorial working aid rather than an automatic product diagnosis. Add an answer link, a responsible person and the next review date for your case. A monthly percentage then becomes a task with a traceable reason.

Your next report starts with the same comparison basis and the documented change. When you deliberately expand the basis, show the transition. Each movement can then be interpreted as a change in your own presence, a development among competitors or a new question being investigated. That distinction makes the share useful for decisions.