AI Share of Voice: How to Measure It
AI share of voice is the percentage of a fixed prompt set where you're named or linked instead of a competitor, tracked weekly across four assistants.
AI share of voice is the percentage of a fixed set of buyer prompts, run weekly across ChatGPT, Claude, Gemini and Perplexity, in which your brand is named or linked instead of a competitor. You calculate it per engine by dividing your appearances by the total prompts run, then track the trend, not one run.
That is a narrower measurement than the full discipline of AI brand visibility, which also covers sentiment and citation quality. Share of voice is the competitive slice of it: not "do they mention me" but "do they mention me instead of the other five companies in my category."
On this page
How do you calculate AI share of voice?
Run the same set of buyer questions through each engine on a schedule, and count two things per prompt: whether your brand was named, and whether it was linked.
This is a borrowed idea, not a new one. Marketers have measured share of voice against paid media and organic rankings for years — the share of impressions, the share of first-page slots. The AI version keeps the logic and swaps the surface. Instead of a search results page with ten slots, you have a chat answer with a handful of sentences and, on some engines, a handful of citations underneath it. The denominator changes; the question does not.
What makes the AI version harder to compute cleanly is that there is no fixed page of ten results to count against. One answer might name three companies, another might name none. So the honest metric is not "our rank" but "our appearance rate": across everything we asked, how often did we show up at all, and how often did that showing-up come with a link.
Build the prompt list
Pull twelve to forty questions from sales calls and support tickets, not a keyword tool — category questions, comparison questions naming competitors, and questions about you by name.
Run it on a schedule
Weekly, on the same day, across every engine you care about. Do not add or remove prompts mid-run; a changed list resets your baseline.
Log two states per prompt
Named (a mention) and linked (a citation) — for you and for each competitor that appears instead.
Divide and report per engine
Your mentions ÷ total prompts = mention share of voice. Your citations ÷ total prompts = citation share of voice. Keep the two engines-worth of numbers separate; never blend them into one score.
For the mechanics of picking and logging the prompt list itself, including how to keep the list stable enough to trust week over week, see AI prompt tracking. The two exercises share a prompt set but answer different questions: prompt tracking asks what the engine said, share of voice asks how you compared to whoever else it said it about.
Do this by hand for the first month before you buy a tool. A spreadsheet with a date column, an engine column, a named column and a cited column is enough to prove the exercise is worth doing before you pay a vendor to automate it. Our own AI visibility audit starts from exactly this spreadsheet, run manually, before we recommend anything more expensive.
Why is the mention number always higher than the citation number?
Because naming a brand and sourcing the claim are two different decisions an assistant makes, and it makes them at very different rates depending on the engine. A model can recall a company's name from its training data and say it with no source at all, which is why "mentioned" is always the larger, looser number and "cited" is the stricter one that actually sends a reader to your page.
| Engine | Mentioned | Cited | Gap |
|---|---|---|---|
| ChatGPT | 56% | 47% | 8 points |
| Perplexity | 67% | 44% | 23 points |
| Claude | 70% | 40% | 30 points |
| Gemini | 72% | 41% | 31 points |
8.1 sources per Gemini answer
ChatGPT is the outlier worth planning around: it has the smallest gap between naming a brand and linking it, so a citation is a realistic near-term target there, and moving your citation share of voice on ChatGPT specifically is a reasonable first goal for a new tracking programme. Claude and Gemini will discuss your category freely without sourcing the answer, which means a strong mention share of voice on those two engines can still hide a weak citation share — you are being talked about, but nobody reading the answer gets a link to check the claim.
Perplexity sits in between, and it cites more sources per answer than any of the other three, which mechanically gives you more chances to appear even when it is not naming you as confidently in the prose. A high citation share of voice with a lower mention share, on Perplexity specifically, usually means your page is being pulled into the source list without the model committing to naming you as the recommendation — worth knowing, because it changes what you fix. A citation problem is a content and access problem; a mention problem, without a citation problem, is closer to a reputation problem.
Does a famous brand's share of voice mean the same thing as a small brand's?
No, and this trips up a lot of first-time tracking programmes.
Well-known brands tend to show a wide gap between mentions and citations: the assistant recalls them from training data and talks about them without needing to fetch a source. A brand nobody has heard of only gets mentioned when the model actually retrieves and reads its site, so its mention and citation numbers move almost in lockstep — when a small brand's citation share drops, its mention share usually drops with it, because there was no memorised fallback to fall back on.
That asymmetry means the same raw percentage means different things depending on how established you already are. A household name sitting at a 40 percent citation share against an 80 percent mention share is not underperforming; that gap is close to normal for a brand the models already "know." A small or new brand sitting at the same 40/80 split should be far more concerned, because for a brand with no training-data presence, a citation is close to the only way it gets mentioned at all — so a wide gap there is a sign something is actively suppressing the citation half of the equation, not just an artifact of fame.
The practical fix is the same either way: track citations specifically, because citations are the number that moves when your GEO work moves. For newer or smaller brands, watch AI brand sentiment alongside share of voice — a rising mention count paired with negative or inaccurate framing is not a win.
Should you set a target number?
Set a trend, not a target. Share of voice numbers swing meaningfully between weekly runs because the underlying engines are not deterministic — the same prompt asked two days apart can return a different set of cited sources. A single low week is not a regression and a single high week is not a win.
What is worth tracking against a target is your position relative to named competitors in the same prompt set: are you closing the gap to whoever leads your category, run over run. That comparative number is more stable than your absolute score, because it is less sensitive to a single engine's off week — if every brand in the category drops together because an engine reshuffled its index, your relative position barely moves even though every absolute number just fell.
If you want one number that compresses this into a single score for reporting upward, AI visibility score covers how to build one without pretending it is more precise than the underlying data. Keep the per-engine, per-week detail somewhere too, because the single score is for a slide, not for deciding what to fix next week.
How do you report this without overselling it?
Plainly, with the caveats attached to the number, not buried in a footnote. Say what the sample size is, say how often the underlying answers reshuffle between runs, and say which engines you covered and which you did not. A stakeholder who is told "we appeared in 40 percent of tracked prompts this month, up from 28 percent, based on thirty questions run weekly on four engines" can act on that. A stakeholder who is handed a single unexplained score cannot tell a real three-week trend from a single unlucky Tuesday.
Resist the urge to smooth the number before it goes upward in a report. If a bad week is real, say so and say why you think it happened — a competitor got included in a listicle, an engine changed its retrieval behaviour, your own site had a deploy issue that blocked a crawler. The explanation is more useful to whoever reads the report than the number itself, and it is the only part of this exercise that still has value in six months, when nobody remembers what the score was.
Questions people also ask
What counts as a mention versus a citation?
A mention is when an assistant names your brand in the text of its answer. A citation is when it links one of your pages as a source. Tracking studies find mentions run well ahead of citations on most engines, so the two numbers need to be reported separately.
How many prompts do I need to track share of voice?
Twelve to forty is a directional sample most brands can run by hand. It will not survive a statistician's review, but it is enough to see a trend across weeks, which is the point.
Should I average my share of voice across all four engines?
No. Each engine cites a different number of sources and reaches for different domains, so a blended average hides exactly which engine you are losing. Report a share of voice per engine.
Why did my share of voice drop even though nothing changed on my site?
The engines are not deterministic. The same prompt run two days apart returns a meaningfully different set of sources, so a single week's dip is usually noise. Judge the number over three or more weekly runs before treating it as a real change.
Sources
- I tracked 13,184 AI citations across 4 LLMs: here's what I learned — Orbit Media Studios (checked 23 September 2026)
- AI features and your website — Google Search Central (checked 23 September 2026)
- Perplexity crawlers and bot access — Perplexity (checked 23 September 2026)
Get your AI share of voice run this week, not next quarter
Book a free 30-minute consultation with Ibrahem and Jeremy. We look at your site while we talk, and you leave with a prioritised plan.
Who wrote this
Ibrahem HelalCo-founder
Leads strategy and delivery.
Works on: SEO, GEO
Reviewed before publishing by Jeremy Zava, Co-founder.
- What Is an AI Visibility Score?An AI visibility score blends citation rate, mention rate and source diversity into one number. Here is what it usually measures, and where it falls short.
- What Is AI Brand Sentiment?AI brand sentiment is the tone assistants use when describing your company — favorable, neutral or hedged. Here is what shapes it and how to check it.
- AI Prompt Tracking: A Practical MethodHow to build the prompt list, decide how often to run it, and log what ChatGPT, Claude, Gemini and Perplexity say about your brand and competitors.
- AI brand visibility: how to measure and grow itHow often ChatGPT, Perplexity, Gemini and Google AI Mode name or cite your brand, how to measure it on a schedule, what moves the number and what it costs.