How to Measure AI Share of Voice: Methods, Tools and Benchmarks (2026)

Rather not build the spreadsheet? Get a free visibility check and we'll run your buyer questions through seven AI engines and show you who gets named, competitors included.

What is AI share of voice?

AI share of voice is your slice of all the brand mentions AI engines make when they answer the questions your buyers ask. If ChatGPT, Gemini and Perplexity name ten brands in total across your questions and three of those mentions are you, your share is 30%.

Semrush's guide, published 17 July 2026, defines it as your brand's mentions divided by all brand mentions in the category, times 100 (Semrush on measuring AI share of voice). LLM Pulse, in a guide published 5 July 2026 and updated 28 September 2026, uses the same mention share and adds a citation rate and a position-weighted score (LLM Pulse's three share of voice formulas). We use all three below.

For the one-off yes-or-no test, see our guide to checking if ChatGPT recommends your business.

How do you measure AI share of voice, step by step?

1. List the questions your buyers ask

Write 20 to 50 questions a buyer types before they pick a supplier, with no brand names in them. "Best accounting firm for a Dubai startup" counts. "Is [your firm] any good" does not, because a prompt with your name in it almost always returns your name.

Group them by intent (comparisons, price, "alternative to", how-to), then freeze the list so each month is comparable.

2. Choose the engines and how often to run each question

Pick the engines your buyers actually use. For most businesses that means ChatGPT, Google AI Overviews and Google AI Mode at minimum, plus Gemini, Perplexity, Claude, Grok or Microsoft Copilot depending on the audience.

Then decide how many runs you need. SparkToro and Gumshoe had 600 volunteers run 12 prompts 2,961 times through ChatGPT, Claude and Google AI in late 2025. The same list of brands came back twice less than once in 100 runs, and the same list in the same order about once in 1,000 (SparkToro's AI consistency study, 28 January 2026). SparkToro suggests about 60 to 100 runs of a prompt for a statistically meaningful reading. Daily tracking over a month gets you close.

3. Record every brand named and every source cited

For each answer, write down every brand in the order it appears, and every website the answer cites as a source. Count a brand once per answer, however many times the answer repeats it.

Write your counting rules down. In our scan we counted only tools in the category, merged name variants such as "Otterly" and "Otterly.AI", and left out review sites and search engines. Apply the same rules every month.

4. Calculate mention share

Mention share = your mentions / all brand mentions across the answers x 100.

Leave out answers the engine failed to return, but keep answers that named no brand at all in any rate that divides by answers. Those empty answers are real: a buyer who asked that question saw nobody.

5. Add presence rate, citation share and a position weight

Mention share alone can mislead, because it moves when competitors come and go. Add three more numbers:

• Presence rate: answers that name you / answers returned x 100. SparkToro calls this visibility % and concluded it is a reasonable metric.
• Citation share: answers that cite your domain as a source / answers returned x 100. LLM Pulse calls this citation rate.
• Position-weighted score: the sum of (100 / your position) across answers, divided by answers returned. First place scores 100, second 50, third 33.3. Semrush's Brand Performance report also weights by how high the brand appears.

Treat the position score with care. SparkToro found that rank position is not a reliable metric, because the order changes from run to run far more than presence does.

6. Split the numbers by engine and by question

A single blended number hides the most useful finding: which engine is missing you. Engines draw on different sources, so one can name you daily while another never does. In our 28 September 2026 scan, Perplexity answers cited 19.3 websites on average against 4.0 for ChatGPT, and our wider research shows the same spread (where AI answers come from).

Split by question group too. You may lead "best tool" questions and be absent from "cheapest" or "alternative to" questions, which point to different fixes.

7. Track the trend against your own baseline

Take your first full month as the baseline, and report each month against it and against two to five named competitors on the same questions. A rise on a fixed question set is real progress; a comparison with someone else's category is not.

What does the calculation look like with real numbers?

Here is the whole method applied to real data. Our Cituna scan of 22 buyer questions across seven engines on 28 September 2026 asked questions such as "best AI visibility tools", "best Profound alternative" and "how much do AI visibility tools cost". The engines were ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode.

That gives 22 x 7 = 154 answer slots. Google AI Overviews did not return one answer, so 153 answers were returned. We read every answer and counted each AI visibility tool it named, once per answer: 844 tool mentions in total, covering 164 different tools. One run per question per engine, so treat this as a snapshot.

Mention share and presence rate

Otterly.AI was named in 84 answers. Mention share = 84 / 844 x 100 = 9.95%, so about 10%. Presence rate = 84 / 153 x 100 = 54.9%: a buyer asking any of these questions in any of these engines saw Otterly.AI in just over half the answers.

The two numbers tell different stories. A 10% share sounds small, but the category is crowded: 164 tools were named, and 93 of them only once. The four most-named tools together took 84 + 68 + 66 + 63 = 281 mentions, exactly a third of 844.

Citation share

Being named and being cited are separate. Semrush was named in 63 answers, and semrush.com was cited as a source in 27, a citation share of 27 / 153 = 17.6%. Otterly.AI was named in more answers (84), but otterly.ai was cited in only 12, or 7.8%. LLM Pulse was named in just 19 answers, yet llmpulse.ai was the most-cited site in the scan, cited in 32 answers (20.9%). Its guides are sources the engines read, even when the answer recommends someone else.

Position-weighted score

Otterly.AI appeared first in 18 answers, second in 15 and third in 18, with the rest further down. Adding 100 / position for every one of its 84 appearances gives 3,784.8. Divided by 153 answers, the score is 24.7.

Position changes the order. Semrush had fewer mentions than Profound (63 against 68), but it was named first 16 times against Profound's 11, so its position score is 19.0 (2,902.8 / 153) against 18.5 (2,830.1 / 153). Ahrefs Brand Radar was named first more often than any tool, 19 times, yet ranks fifth on mentions. Given SparkToro's finding that order is unstable, read these gaps as a hint, not a result.

Split by engine

The per-engine split shows why a blended figure hides the story. Otterly.AI was named in 16 of 22 Grok answers (73%) but 7 of 22 Google AI Mode answers (32%). Semrush was named in 13 of 22 Perplexity answers (59%) and 5 of 22 Claude answers (23%). Same questions, same day, very different pictures by engine.

What is a good AI share of voice?

No one has published one. We looked for a benchmark in primary sources and found none. Semrush's guide says outright that it gives no benchmark for a good score. Figures in vendor worked examples, such as LLM Pulse's example of 15.2% mention rate, 8.8% citation rate and 31.7% share of voice over 250 runs, illustrate the maths. They are not targets.

What the research does give you is a sense of how much the numbers move:

• Lists vary run to run. SparkToro found the same brand list less than 1 time in 100, and answer lists ranged from 2 or 3 items to more than 10.
• Top brands still show up often. In tight categories, SparkToro's top brands appeared in 55% to 100% of runs. That is presence rate, not share of voice.
• Google's AI answers churn. Ahrefs tracked 43,000+ keywords for a month: an AI Overview had a 70% chance of changing between observations, and 45.5% of cited URLs changed, while the meaning stayed 0.95 similar (Ahrefs on how often AI Overviews change, 11 November 2025).
• Engines disagree on sources. Ahrefs found only 13.7% of cited URLs overlapped between AI Overviews and AI Mode for the same queries, though the answers were 86% similar (Ahrefs on AI Overviews vs AI Mode, 15 December 2025).

So set your own benchmarks: your first month, and the competitors you lose deals to.

Which tools measure AI share of voice?

You can do all of this by hand for a first reading, but repeat runs across seven engines soon outgrow a spreadsheet. These tools run the questions for you. Prices are from each vendor's own page on the date shown. For a wider comparison, see our guide to the best AI visibility tracking tools, and for how to pick one, the seven criteria for choosing an AI visibility tool.

Cituna

Cituna checks your buyer questions every day on seven engines: ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode. When it finds a gap, it writes the fix and can publish the article. Google Search Console and MCP are included. Plans are $39, $119 or $399 a month, with a 3-day trial that needs a card. It does not track Microsoft Copilot. It is the tool High Street uses for the scans on this page. Cituna's pricing page

Semrush AI Visibility Toolkit

The AI Visibility Base plan costs $99 a month per domain (pricing page, checked 28 September 2026) and tracks 25 custom prompts daily, with mentions from ChatGPT, Google AI, Gemini and Perplexity. Its Brand Performance report updates weekly and weights share of voice by position. Claude and Grok are not named on the pages we read, and extra domains or locations cost $99 a month each.

Peec AI

Peec AI lists source analytics and citation share, with daily tracking on its Starter, Pro and Advanced plans, and lets you pick three models on self-serve plans, including Microsoft Copilot. Plan prices were not published on the pricing page we read on 28 September 2026.

LLM Pulse

LLM Pulse starts at EUR 49 a month for weekly tracking or EUR 79 for daily, with 50 prompts and five engines including AI Mode and AI Overviews (checked 28 September 2026). It tracks mentions and citations, and MCP is on every plan. Claude, Copilot and Grok are paid add-ons.

Questions people ask