How To Measure Share Of Voice In AI Answers Against Your Top "3" Competitors

Report one number a month. It is your brand's mentions divided by the combined mentions of your brand and your three competitors, counted across the same fixed set of buyer prompts in ChatGPT, Perplexity and Google AI. Hold the prompt set, counting rule and competitor list fixed, or month-over-month movement tells you nothing.

This applies when you run a fixed set of at least 30 buyer prompts and name the same three competitors every cycle. A smaller or shifting set moves too much to report.

The result is a sample of what the engines said on the days you ran the prompts, so its margin of error can be larger than the month-to-month change you are trying to read.

What AI Share Of Voice Measures (And What Classic Share Of Voice Does Not)

AI share of voice is your brand's share of all the mentions that your brand and three named competitors receive across a fixed set of buyer prompts run in ChatGPT, Perplexity and Google AI. The denominator is the combined mention count of those four brands, and never the number of answers.

Share of voice is an old marketing metric. In its classic form it measures a brand's slice of media coverage, social chatter, PR or ad spend. This guide covers none of that. It covers one question, which is how often AI answers name your brand next to the competitors a buyer weighs you against.

Classic share of voice centres on rankings, impressions and clicks. AI share of voice asks a blunter question, which is whether your brand appears in the answer at all. A brand can hold the top Google position for a keyword and still be missing from the paragraph a buyer reads in an AI answer. That is why the two numbers sit side by side later in this guide instead of replacing each other.

Share of answer is the close cousin people confuse with it. Share of answer is the share of tracked prompts where your brand appears. Share of voice takes that presence and compares it against a named competitor set. One tells you whether you show up. The other tells you how you stack up.

Share of voice is also not market share. Market share counts revenue or customers. Share of voice counts how often the engines talk about you, which is attention, and attention can run ahead of or behind sales.

For the wider set of numbers that sit around this one, see AI search visibility metrics and KPIs.

Step 1: Choose Your Three Competitors And Lock The List

The rule is short. Track three direct competitors, keep the same three every month, and review the list once a quarter. A changed list breaks month-over-month comparison, because the denominator moves with it. Swap one rival for a bigger name and your share can fall even though nothing about your own visibility changed.

Here is how to set the list.

1. Collect the real comparison set. Pull the brands that come up on sales calls and in lost-deal notes. Then run five or six category prompts in each engine and note who shows up next to you. The brands a buyer actually compares you with are the only ones worth counting.
2. Pick three direct competitors. Choose rivals that sell to the same buyer at a similar price and size. Do not pad the list with the biggest name in the category just because it appears in every answer. A giant in the denominator buries every other share.
3. Write the list down with exact names. Record each brand with its product names and common abbreviations, so two people counting the same answer reach the same total.
4. Freeze the list for the cycle. Review it once a quarter. When you change a name, start a new baseline instead of comparing across the change.

Sources vary on how many brands to track. Texta advises 3 to 5 named direct competitors with the same list each cycle, and Eclincher's guide gives a competitor set of 4 to 8 brands including your own, with 3 to 5 direct rivals. This guide uses three. The report stays readable, and the denominator stays comparable month to month.

Step 2: Build A Prompt Set You Can Rerun Every Month

The prompt set is the instrument. Change it and you have changed the ruler, so every later month is measured against a different thing. Build it once, approve it, and leave it alone.

1. Start with 30 buyer prompts. Split them into ten branded, ten category and ten problem-or-solution prompts. Averi recommends a starter set of 30 prompts for a B2B SaaS brand in exactly that split, and gives 25 to 50 as its recommended starting range.
2. Write each prompt the way a buyer would type it. Use full questions with a real situation in them. A prompt stuffed with your own marketing language returns answers no buyer would ever see.
3. Source the wording from real conversations. Sales calls, search queries and support tickets all carry the phrases buyers use. Copy their wording and resist the urge to tidy it.
4. Freeze the set once it is approved. Within a reporting cycle, no prompt is added, edited or dropped. If a prompt turns out to be poorly worded, note it and fix it at the next quarterly review.

Here are three examples, one per group, for a generic B2B SaaS category. They show the shape and are not a set to copy.

• Branded: "Is Brand X a good fit for a 50-person agency?"
• Category: "What are the best project management tools for a growing software company?"
• Problem or solution: "How do I stop client deadlines slipping across five projects at once?"

Sizing advice varies by source. Eclincher recommends a prompt library of 30 to 50 prompts and a pilot of 20 prompts over three platforms for one month. Texta recommends 30 to 100 prompts in the set. Those are vendor guidance, and none of them is a study.

Start at 30. Fewer than that and the number swings too much to report. More prompts narrow the noise, which Step 5 puts numbers on. If you already have a keyword list, mapping keywords to prompts shows how to turn it into the monthly set without starting from a blank page.

Step 3: Pick One Counting Rule And Write It Down

A mention is not the only thing you can count. Counting rules must be chosen and documented, because two teams using different rules will report different numbers for the same set of answers.

The table below lays out the four rules in use.

Pick one rule for the monthly number and keep the others as secondary figures. Use mention as the headline rule for your first baseline. It is the easiest to audit, because anyone on the team can open a response and check whether the name is there.

One scheme splits mention share from strong share, where strong counts recommended, cited or top-of-answer appearances. That is a good second number. It should not be the first.

Three details decide whether the count holds up.

• A mention is not a citation: A mention is your brand name in the text. A citation is your domain listed as a source. An answer can do one without the other, so they stay separate columns.
• Several brands in one response: A response that names several brands counts once for each brand it names. An answer that mentions you and two competitors adds three mentions to the total.
• Sentiment: Count the mention first. Then note negative or inaccurate mentions separately, so a brand being described wrongly does not quietly raise your score.

Before you trust the count, run the reliability test for false mentions. A brand with a common-word name or a near-identical rival can pick up mentions it never earned.

Step 4: Choose The Denominator And Run The Calculation

Here is the formula in words and symbols.

Share of voice (%) = Y ÷ (Y + C1 + C2 + C3) × 100

Every term is defined.

• Y: The total mentions of your brand across every prompt and platform in the month.
• C1, C2 and C3: The total mentions of each of your three competitors, counted with the same rule.
• Denominator: Y + C1 + C2 + C3, which is the combined mentions of all four brands.

The alternative is to divide by the number of responses. That gives the share of answers where your brand appears, which is share of answer, not share of voice. Both are legitimate. They answer different questions, and mixing them in one report is how a number gets challenged in the meeting.

The denominator also depends on the counting rule. AI share of voice can be read through mention share, citation share or recommendation share, each with its own denominator. With mention as your rule, the denominator is total mentions. With citation as your rule, it is total citations. Keep the pair matched.

When you combine platforms, add the raw mentions across platforms first, then divide. Never average the per-platform percentages. Say Perplexity returns far fewer mentions than ChatGPT in your set. Averaging the two percentages gives each engine equal weight even though one produced much more of the evidence. Raw counts weight each engine by what it actually said.

This guide uses total mentions of the four brands as the denominator. It keeps the four shares adding to 100% and makes the competitor comparison direct. The next section runs it on real arithmetic you can copy.

A Worked Example With Three Competitors Across Two Months

The counts below are constructed for this article. They are an illustration and not real data, built so you can check every figure and then rerun the same steps with your own counts.

The setup is 30 prompts run on 3 platforms, which gives 90 answers per month. Total mentions come to more than 90 because a single answer can name several brands.

Here is the arithmetic for your own brand, one line at a time.

• Month 1 total: 27 + 45 + 24 + 12 = 108 mentions
• Your brand, month 1: 27 ÷ 108 × 100 = 25.0%
• Month 2 total: 36 + 45 + 21 + 18 = 120 mentions
• Your brand, month 2: 36 ÷ 120 × 100 = 30.0%
• Change: 30.0% - 25.0% = 5.0 percentage points

Your brand gained 5.0 points. Competitor C gained 3.9 points. Competitor A lost 4.2 points and Competitor B lost 4.7 points. Competitor A still leads, but its lead over your brand shrank from 16.7 points to 7.5 points. The four shares add to 100% in both months, which is the quick check that the arithmetic is right.

Whether a 5-point gain is a real change is a separate question. The next section answers it.

Step 5: Handle Platforms, Weighting And Sampling Noise

Three decisions are left. Each one is short, and each one stops a good method from producing a misleading number.

Platforms

Report each engine separately, then build the combined figure from raw counts. ChatGPT, Perplexity and Google AI answer differently, and a combined number alone hides which engine moved. If your buyers also use Google AI Mode or Gemini, add them. Those are the other two engines named as most relevant for B2B SaaS.

Weighting

Weighting scores a prominent appearance above a passing one. One scheme scores each prompt's prominence on a 0 to 3 scale and multiplies it by a volume weight, dividing by the maximum possible total. Another gives a mention 1 point, a citation 2 points and a top recommendation 3 points, with share taken over total weighted points across all brands.

Start unweighted. It is easier to explain, and a plain mention count is easier to audit. The full weighted formula and the tool choice sit on our page for the weighted share formula and tool comparison, so this guide does not repeat them.

Sampling Noise

Your number is a sample, and samples wobble. The margin of error on a mention rate is 1.96 x sqrt(p(1-p)/n), where p is the mention rate and n is the number of answers, assuming the answers are independent. At a 30% mention rate it works out as follows.

Apply it to the worked example. Thirty prompts on three platforms gives 90 answers a month. In month 1 your brand was named in 27 of those 90 answers, which is a 30% mention rate. At 90 answers the margin is about 9.5 points.

Your share moved 5 points. That sits inside the margin, so it is not yet a finding. Read it as a hint and keep watching. The margin is a guide and not an exact bound, because share of voice divides by mentions and not by answers, but the practical reading is the same.

Reaching 400 answers on three platforms takes about 134 prompts. That is the price of a margin near 4.5 points, and most teams will run a smaller set and read the trend over several months instead.

How To Read The Monthly Trend Before You Report It

Sources disagree on cadence. Eclincher advises weekly snapshots on the same day and time with monthly reporting, Averi advises a monthly full run, a biweekly subset and at least 90 days of tracking, and Texta calls weekly the recommended cadence. They are all vendor guidance, and none of it is a study.

Pick a rhythm you can keep. Run the full prompt set once a month on the same day. Optionally rerun a subset mid-month for an early signal. Report monthly. Read the trend over a rolling four weeks, with a twelve-week view for direction. Averi uses a 4-week trend window and a 12-week window for direction, and that split is easy to explain to a leadership team.

Keep the prompts and competitors fixed within the cycle. A trend only means something when the ruler stayed still.

Then apply the reading rule. Act on a change only when one of two conditions holds.

• The change exceeds the margin of error: Treat it as a finding and decide what to change in your content or your coverage.
• The change persists across several weeks: Treat it as a trend, even if a single week sat inside the margin.

A single week is anecdotal. If a result surprises you, rerun it before you act on it. A surprising number that does not repeat was noise, and you will have saved yourself a bad conversation with leadership.

What To Put In The Monthly Report For Leadership

Leadership does not want your spreadsheet. It wants one number, what moved, and whether to believe it. Build the report in five parts.

1. The headline number. Mention share against the three competitors, all platforms combined from raw counts.
2. A per-platform table. One row per engine with your share and each competitor's share, so a movement can be traced to the engine that caused it.
3. The change against last month, with its margin. Show the points gained or lost and say plainly whether it sits inside the margin of error.
4. A method footnote. Publish the prompts, platforms, competitors, location and period. Without those, the number cannot be interpreted.
5. An appendix for the secondary numbers. Keep citation share, recommendation share, the weighted score and the negative-mention tally here, out of the headline.

Report the mention share and keep the rest in the appendix. One clear number beats five competing ones.

Be honest about sample size in the footnote. A manual run of 30 prompts on three platforms is 90 answers a month. Texta says a monthly reading becomes meaningful at around 1,000 or more AI evaluations per month. That is vendor guidance and not a study, but it points the same way as the margin table. A small set can still show direction over twelve weeks. It cannot settle a one-month argument.

Here is a two-line way to explain the number to a sceptical executive.

"Out of every 100 times an AI answer names us or one of our three closest competitors, we get 30 of them. Last month it was 25, and we will confirm the trend over the next few weeks before we change anything."

For the page this number goes into, use the monthly AI visibility report template. If you are moving the number onto a live view, see dashboards for reporting AI visibility to leadership. Teams that would rather not run the cycle themselves can look at our GEO + SEO growth programme reporting.

Pairing AI Share Of Voice With Classic Search Share Of Voice

The two numbers answer different questions, so report both for the same three competitors. Classic share of voice counts rankings, impressions and clicks on a keyword list. AI share of voice counts whether your brand appears in the answer an engine writes. A brand can win the first and lose the second, and the gap between them is where the interesting work sits.

Put the two side by side in one view. If your classic share is high and your AI share is low, the engines are not picking up pages that Google already rewards. If the reverse is true, your brand is being named in answers without owning the rankings behind them. Either pattern tells you where to spend the next quarter.

For the classic-search half of the comparison, Semrush can track the same three competitors across a keyword list, so the Google ranking picture sits beside the AI number in one report.

Traditional SEO versus AI SEO explains why the two numbers diverge and how strategy changes. The service that acts on what the AI number shows is Answer Engine Optimization.

This guide gives no benchmark for a good percentage. No named study gives a standard figure, and the ranges circulating online have nothing behind them. Your own three competitors and your own trend are the only fair comparison.

Common Mistakes That Make The Number Meaningless

Every one of these is avoidable.

1. Changing the prompt set mid-cycle. The month-over-month change then reflects a new ruler instead of a new result.
2. Changing the competitor list. A different set of rivals changes the denominator, so a share can rise or fall without any change in your visibility.
3. Averaging platform percentages. Each engine gets equal weight regardless of how many mentions it produced, which skews the combined figure.
4. Counting a response that names several brands once. Named brands drop out of the tally and every share is distorted.
5. Reading a single week as a trend. One week is anecdotal, and acting on it means reacting to noise.
6. Publishing a percentage without its prompts and period. A number with no prompts, platforms, competitors, location and period behind it cannot be interpreted or defended.