How do I measure share of voice for our brand in AI answers versus competitors?

Quick Answer: To measure share of voice in AI answers, run a fixed set of buyer questions on each assistant, count how often your brand and each competitor is named, and divide your mentions by all brand mentions and multiply by 100. A spreadsheet covers a few dozen prompts on one or two engines. Daily reruns across several assistants need a tool.

Key takeaways

• Write down your competitor list and your denominator rule before the first run, and start a new series if either one changes.
• Give each assistant its own line in the report before you combine any figures.
• Keep sentiment, position and cited sources in their own columns beside the headline percentage.
• Wait for a second run on identical prompts before you call any movement a trend.

What does share of voice mean in AI answers?

AI share of voice is the percentage of brand mentions in AI assistant answers that belong to your brand. This applies to any brand owner who wants to know how often AI assistants name them compared with competitors when buyers ask about their category. Classic share of voice compares your slice of the conversation in ads, search or social media. Shopify's guide gives the general formula as "Share of voice (%) = Your brand metrics ÷ Total market metrics × 100" [1]. The same guide says tracking the version for large language models (LLMs, the systems behind assistants such as ChatGPT) "is still an emerging tactic" [1].

Answers also change from one run to the next. Profound's product page puts it plainly: "AI platforms don't return the same answer twice." [2] So fix the prompts and the competitors and run them more than once.

How do you measure share of voice step by step?

You run the same buyer questions on each assistant, tally every brand the answers name, and divide your tally by the total.

First, pick a formula. Shopify's guide writes the LLM version as "Mentions of your brand in LLM answers ÷ Total competitor brand mentions in LLM answers × 100" [1]. Mention's calculator uses a different denominator: your mentions plus competitor mentions [3]. This article uses the second, so all brand shares add up to 100%.

1. Write down the questions buyers ask when choosing in your category.
2. List your brand and each competitor you want to compare, including common spelling variants.
3. Run every question on each assistant you track, and save the answer text with the date.
4. In a sheet, give each brand a column. Count one mention for each answer that names that brand.
5. For each brand, divide its count by the sum of all brand counts and multiply by 100.
6. Recompute the percentage from your tally sheet.

Expected result: Your mentions divided by the sum of all brand mentions, times 100, equals the number you report, and the brand shares add up to 100%.

Say your brand is named 18 times across the answers and three competitors are named 42 times between them. That gives 18 / 60 = 30%.

Or hand the runs to a tool: Welt Analytics, for example, reports a brand's visibility, share of voice, position, sentiment, sources and competitors in AI answers.

Which counting choices change your share-of-voice number?

What you count in each answer decides what the percentage tells you.

Sentiment needs its own column because Mention's calculator page notes that "online mentions can be positive, negative, or neutral." [3] A high share built on warnings works against you.

The sources column is where you find the next action. Profound's documentation gives two cases. If your content is cited but your brand is not named, it says: "Review how the cited pages present your brand." If you are named but other pages are cited, it says: "Build content that answer engines can cite." [6] The same columns also show you which competitors AI assistants recommend over you.

Freshness differs by engine. Welt Analytics checks ChatGPT, Gemini, Google AI Overviews and Google AI Mode daily, and Claude and Perplexity weekly.

What mistakes make AI share of voice misleading?

These mistakes make the figure misleading.

Friendly prompts skew the set: a question that names you up front almost always gets an answer that mentions you. Build the headline set from neutral category questions.

Drift is next. If you add a rival halfway through the series, your share falls even though no answer changed. Fix the denominator rule (your mentions plus competitor mentions) and the brand list before the first run. Change them only by starting a new series.

Paid-search share is a different measure. Google Ads impression share is "the number of impressions your ad received divided by the total number of impressions your ad was eligible to receive" [4]. Auction insights compares your ads with other advertisers for bidding decisions [5]. Keep impression share and Auction Insights as separate paid-search measures. The AI figure counts brand mentions in answers.

The last mistake is mixing cadences. A single trend line that blends daily and weekly engines jumps each time the weekly engines refresh. Report each engine on its own line.

How do you read the percentage and know it is working?

Compare the percentage only with your own earlier runs on the same prompt set. A competitor's published figure uses different questions.

1. Keep the prompt file, the competitor list and the engines exactly as they were in the first run.
2. A week later, rerun every prompt on the same engines.
3. Tally the answers with the same rule, and compute each brand's percentage for each engine.
4. Put the two runs side by side, one row per engine.

Expected result: You get a second percentage on the same basis, so a change reflects the answers and not the method.

A large swing with nothing changed usually means the prompt set is too small, so add questions before you trust the trend. A move in the same direction across several reruns is a real signal: check the sources column to find out why.

Treat the figure as a measure of presence. It does not promise traffic, sales or a position in any answer.

Do you need a tool to measure share of voice in AI answers?

You need a tool once the prompt set or the list of engines grows past what you can rerun by hand. If you are weighing options, read about which GEO tool is easiest to set up.

Welt Analytics runs the checks on the engine cadence described in the counting section. It offers a 7-day trial, writes the first prompts and needs no card [7]. It also publishes new articles on your blog, written to answer those questions, and every article belongs to you [7]. Other tools track AI answers too. Before you commit to one, check its engine list and how often it reruns.

If rerunning the prompt set by hand is the part you would rather hand over, see how Welt Analytics tracks it.