AI Share of Voice Is a Sample, Not a Share
- One brand, one dataset, three AI share of voice scores: 20% mention-based, 16.8% position-weighted, 31.4% citation-based.
- SparkToro and Gumshoe.ai ran 12 prompts 2,961 times and found fewer than 1 in 100 runs returned the same list of brands.
- The denominator is the vendor's prompt list. Change the list and your "share" moves while your site does not.
- Kymo samples nothing. It records what AI crawlers and live fetchers actually did on your site, per page.
- Kymo does not track mentions and does not tell you what an assistant said.
AI share of voice is a percentage of a sample you did not design. Understand that and every vendor number you see gets easier to read.
One brand, one dataset, three scores, three different places
Digital Applied ran one brand through a single dataset and scored it three ways. Mention-based share of voice came out at 20%. Position-weighted came out at 16.8%. Citation-based came out at 31.4%. Same brand, same data, and the competitive placement changed with the formula.
| Method | What gets counted | Score in the worked example |
|---|---|---|
| Mention-based | Any appearance of the brand in an answer | 20% |
| Position-weighted | Appearance, discounted by where it landed | 16.8% |
| Citation-based | Links or sources credited to the brand | 31.4% |
If two vendors handed you different llm share of voice numbers last quarter, this is usually why. They are counting different things and calling both "share of voice ai."
The denominator problem: the vendor's prompt list sets your score
Your share is your appearances divided by the total. The vendor picks the total.
Fifty prompts, five appearances, you sit at 10%. The vendor adds fifty prompts about industries you have nothing to do with. Your appearances stay at five. Your share drops to 5%. Your site did nothing. The list moved.
Measuring AI visibility means being able to name the denominator, not just the percentage. Ask any vendor what set of prompts produced your number and whether that set changes month to month. Many will describe it as "our proprietary prompt library," which is a polite way of saying "we decide."
Even a fixed prompt list does not give you a fixed answer
SparkToro and Gumshoe.ai had 600 volunteers run 12 prompts through ChatGPT, Claude and Google AI Overviews and AI Mode 2,961 times in January 2026. Fewer than 1 in 100 of those runs returned the same list of brands. Fewer than 1 in 1,000 returned the same list in the same order. Visibility percentage held up better across runs than position did, which is useful to know if you are tempted to report a rank. Rand Fishkin, whose company published the study, has called AI ranking-position tools "full of baloney." Gumshoe sells AI tracking, and the post discloses that.
The models wobble for their own reasons. Thinking Machines Lab ran the same prompt 1,000 times at temperature 0 in September 2025 and got 80 unique completions, traced to server load changing the batch size. Chen, Zaharia and Zou observed GPT-4 score 84% on a prime-number task in March 2023 and 51% in June 2023. OpenAI rolled back a GPT-4o update on April 29, 2025 because it was overly flattering. The answers moving around is not a bug on your end.
And the answers matter to the click. Pew Research Center surveyed 900 US adults in July 2025 and found users clicked a result in 8% of visits to a Google page with an AI summary, against 15% without one, and clicked a link inside the summary in 1% of visits. Ahrefs compared 540,000 query pairs in December 2025 and found Google AI Mode and AI Overviews cited the same URLs only 13.7% of the time.
What Kymo measures instead
Kymo does not type prompts into assistants and count brand names. It cannot give you a "share." What it does is sit server-side on your own site and record what AI engines actually did there.
Three categories, based on the crawler's own user agent. ai_answer covers a live fetch made for one person's question, like ChatGPT-User (OpenAI's live fetcher). indexing covers crawlers building a search index for later answers. training covers bulk crawls like ClaudeBot (Anthropic's training crawler) and Bytespider (ByteDance).
You get per-page records: which page got read, which category read it, and which confirmed click-throughs came back from AI referrals. Reading the AI report walks through the layout. AI crawler tracking does not count against your event limit.
Kymo does not track mentions. It does not know what an assistant said about you, and it never claims to. If you want a share-of-voice percentage, Kymo is the wrong tool and you should buy a sampler instead.
The free AI visibility check at kymo.in/tools/ai-visibility-checker reads a URL from public signals only. No account, no install. The report arrives by magic link over email.
What to do with a share-of-voice number
Treat it as a temperature reading, not a scoreboard. Track it over months, note the direction, and make sure the vendor's prompt list stays fixed while you do. If it changes, note that too.
Then go read your own server logs, or install something that can. Kymo is hosted and cookieless. Solo is $9 a month or $90 a year for up to 10,000 events a month. Studio is $29 a month or $290 a year for up to 100,000 events a month. Every feature is on both plans, and the plans differ only by event volume. Both start with 14 days free, no card required.
Two other things worth doing that are not share-of-voice:
Publish an llms.txt file if you want. Adoption is partial. No major operator has publicly committed to honouring it, and the only way to know whether a crawler fetched yours is your own server log.
Read LLM SEO: What It Means and How to Start and The Small Business Guide to AEO if the vocabulary is still new. Then compare tools against your actual budget in AI SEO Tools for Small Business: What's Worth Paying For.
The recommendation
Do not report a single share-of-voice number as if it were a fact. Skip the sample-based tools entirely and measure what happens on your own domain if you are a solo builder or a small team with no analyst. Which pages AI engines read, which live fetches came from real questions, which clicks came back. That data is yours, it is per page, and it does not move because a vendor changed their prompt list.
Stop guessing. Watch.
Most AI visibility tools ask a chatbot and report a guess. Kymo watches your own site instead: which pages AI engines read, how many people their answers send you, and which of those people buy. Every number comes from your server, not from a prompt someone else chose.
Start free → 14-day free trial. No card required.