How to Measure AI Share of Voice: The Three Decisions That Change the Number
The same prompt plays a different game on every platform -- Perplexity routinely stacks several times more citations into an answer than ChatGPT. Yet every guide hands you one formula: mentions divided by total, times 100. Share of voice is three decisions -- denominator, position weighting, aggregation -- before it's a number. This is the measurement playbook.
AI share of voice is the percentage of AI-platform answers that name your brand, relative to every brand named across the same prompt set. The formula every guide hands you -- your mentions divided by total brand mentions, times 100 -- is correct and nearly useless on its own, because the number it produces depends on three decisions most teams never make explicitly: which brands count in the denominator, whether a first mention is worth more than a fifth, and whether you blend platforms into a single figure.
The stakes are real. With ChatGPT at 900 million weekly active users, "who does the AI recommend" has become a board-slide question, and share of voice is the number that answers it. The articles ranking for this topic all give the same three steps -- pick 4-8 competitors, run some prompts, divide -- and then pivot to a tool pitch. Each of those steps buries a decision that can double your score or halve it without anything changing in the real world.
The formula is the easy part. The denominator, the weighting, and the aggregation are where the number gets made.
This post walks the three decisions in order, with the honest default for each: how to define the competitive field, when position weighting clarifies and when it misleads, and why a blended cross-platform score is the one number you should refuse to put on a slide.
What AI share of voice actually is
Share of voice in AI answers comes in two distinct flavors, and most guides conflate them. Mention-based share of voice counts how often your brand is named in the answer text -- the AI says "consider Brand X" and that counts, regardless of which sources informed it. Citation-based share of voice counts how often your domain appears among the URLs the platform cites -- your site is the source, whether or not your name leads the answer.
The two move independently. A brand with strong third-party coverage can be named constantly while its own domain is never cited; a publisher can be cited in every answer while no one recommends its product. They also reward different work -- the distinction maps onto the mention graph versus the citation graph we walked in the DR-to-citation-share mental model. Decide which one you are measuring before you compare any two numbers, because a 20% mention-based score and a 20% citation-based score describe different realities. Everything below applies to both; the decisions are the same.
Decision 1: Who counts in the denominator
Share of voice is a fraction, and the denominator is a choice. Measure against the five rivals your marketing team tracks and you might score 18%. Measure against every brand the platforms actually surfaced on the same prompts -- often fifteen to twenty-five names once adjacent categories and open-source alternatives show up -- and the same brand scores 6%. Same prompts, same answers, three times the score, depending on a spreadsheet decision nobody documented.
The three decisions hiding inside one share-of-voice number
The formula is the same everywhere. The number it produces depends on three choices most measurement setups never make explicitly.
| Decision | The default mistake | The honest choice |
|---|---|---|
| The denominator | Hand-pick 4-8 named rivals and divide by that closed pool. | Count every brand the platforms actually name across your prompt set -- let the answers define the field, then segment. |
| Position weighting | Treat the first and ninth brand in an answer as the same mention. | Weight earlier mentions higher -- but only across N runs per prompt, never from a single answer. |
| Platform aggregation | Average ChatGPT, Perplexity, and Gemini into one blended score. | Report per-platform. If forced to blend, weight by where your buyers actually ask -- and say so on the slide. |
The hand-picked set fails in both directions. It inflates your score by excluding brands the AI actually names, and it blinds you to challengers you did not think to list -- the fast-rising tool that is quietly winning your category's recommendation answers never appears in your dashboard because you never typed its name in. The honest choice is to let the answers define the field: log every brand the platforms name across your prompt set, compute share of voice against that full field, and segment afterwards if you want a direct-rivals view. This is also how Ranqo computes it -- competitors are auto-discovered from the platform responses themselves, because the AI's answer, not the marketing team's shortlist, defines who you are losing to. Here is what that looks like in practice:
Share of Voice
CompetitorsBrand mention share across AI platforms
The Share of Voice card from the Ranqo dashboard, showing sample data for Notion. Every competitor in the legend was auto-discovered from platform answers, and Notion's 16.5% share is its 681 mentions divided by the full 4,120-mention pool -- where the Others slice alone holds 1,322 mentions, the long tail a hand-picked rival list never sees.
Decision 2: Whether position counts
AI recommendation answers are ranked lists in disguise. Being the first brand named in "what is the best CRM for early-stage SaaS" is a different commercial outcome from being the ninth, the same way position one and position nine on a search results page were never equivalent. Raw share of voice treats them identically. The research version of this intuition is formalized in Princeton's GEO paper, whose Position-Adjusted Word Count metric weights citations by where they appear in the answer -- earlier positions weighted higher, decaying exponentially toward the tail.
To make the difference concrete with round numbers: Brand A is named in 30 of 50 answers but typically in fourth position; Brand B is named in 22 but usually first or second. Raw share of voice crowns Brand A at 33%. Weight the same mentions by position and Brand B leads at 42% while Brand A drops to 21%.
Raw vs position-weighted share of voice: the rank order flips
Illustrative example: a five-brand field across 50 answers, weighted by reciprocal rank (first mention = 1.0, second = 0.5, third = 0.33). Editorial math, not measured data.
| Brand | Mentions | Typical position | Raw SoV | Weighted SoV |
|---|---|---|---|---|
| Brand A | 30 of 50 | 4th | 33% | 21% |
| Brand B | 22 of 50 | 1st-2nd | 24% | 42% |
| Brand C | 18 of 50 | 2nd-3rd | 20% | 21% |
| Brand D | 12 of 50 | 3rd | 13% | 11% |
| Brand E | 8 of 50 | 5th | 9% | 5% |
Brand A leads on raw mentions; Brand B leads once position is weighted. Both numbers are "share of voice" -- which one your dashboard shows is a decision, not a given.
One honest caveat before you re-rank your dashboard: position within a single run is noisy. The same prompt re-run minutes later can reorder the brand list, so a position read from one answer is a coin flip, not a fact. Weight positions only across a sample -- N runs per prompt, positions averaged -- and never conclude anything from a single screenshot. Used that way, position-weighted share of voice is the closer proxy for what a buyer actually experiences; used on single runs, it manufactures precision that is not there.
Decision 3: Refuse the blended number
The third decision is the one every tool dashboard quietly makes for you: whether 15% on ChatGPT and 8% on Perplexity becomes "11.5% overall AI share of voice." It should not, because the two numbers measure structurally different systems. Perplexity is a citation engine -- it composes every answer from a stack of cited sources, several times the citation density of a ChatGPT response, which changes how many mention slots exist per answer and how hard each one is to win; the architecture behind that is covered in our Perplexity playbook. The source mixes differ just as much: Tinuiti's Q1 2026 citations report puts Reddit alone at 24% of Perplexity's citations, while Gemini stays anchored on Google's Knowledge Graph and Google-owned properties.
Averaging across platforms is averaging across different games. Report share of voice per platform, always. If leadership insists on one number, weight the platforms by where your buyers actually ask -- usage in your category, not global market share -- and say so on the slide, because an unweighted blend silently assumes your audience is evenly split across surfaces, and it never is.
The measurement loop in practice
With the three decisions made, the mechanics are sampling discipline. One run is a coin flip: AI answers change between runs, sessions, and locations, so share of voice is only stable as an average over a defined prompt set, run in fresh sessions, at least five times per prompt. We walked the full manual method -- prompt-set design, logged-out sessions, the logging schema -- in how to check if your brand appears in AI; this post inherits all of it.
Tracked weekly, three readings tell you whether shipped work is moving anything: the per-platform share-of-voice trend, position deltas -- which competitors moved above or below you on which prompts -- and gap prompts, the prompts where competitors are named and you are not. Gap prompts are the work queue; they are the most direct translation of a measurement into a content decision.
The per-platform rule is not theoretical. When we ran the same prompt set across platforms for our CRM recommendations study, the share-of-voice leader differed by platform -- the brand winning ChatGPT's answers was not the brand winning Perplexity's. A blended score would have averaged away the entire finding.
The honest summary
AI share of voice is the most decision-useful number in AI visibility -- after three decisions. Count every brand the platforms actually name, not the rivals you listed. Weight positions, but only across a sample, never from a single answer. Report per platform, and refuse the blended score unless it is explicitly usage-weighted. Make those choices out loud and the metric tells you where you stand and what to fix; leave them implicit and it is a vanity number with a formula attached.
The same discipline applies to any tool's dashboard, including ours: when you see a share-of-voice number, ask who is in the denominator, whether position counts, and what got blended. The vendors who answer those three questions plainly are the ones measuring something.
See your share of voice, measured the honest way
Our free AI Visibility Checker runs a prompt set across ChatGPT, Perplexity, and Gemini and returns visibility and share of voice in minutes. For the ongoing loop -- per-platform share of voice with auto-discovered competitors, position deltas, and gap prompts -- that's what Ranqo tracks daily.
Check your AI share of voice freeWritten by
Nisha Kumari
Nisha Kumari is Co-Founder at Ranqo, where she leads growth strategy and client acquisition. With a background in digital marketing and financial management, she specializes in SEO, Generative Engine Optimization, and helping brands build visibility across AI platforms.
Share this article
Related articles
How to Optimize for the Agentic Browser Era
Adobe Analytics reported a 4,700% YoY surge in AI-driven traffic to US retail sites in July 2025 — and your client-side analytics doesn't see most of it. Comet launched July, Atlas in October, and the agent is now a third class of visitor with its own retrieval pattern. Here's the playbook for the agentic browser era.
How to Optimize for Google AI Mode
Google AI Mode decomposes every query into a multitude of sub-queries before it answers -- in Google's own words. Yet every ranking how-to recycles the same syndicated list that conflates AI Mode with AI Overviews. The optimization target isn't the head query; it's the fan-out set behind it. Here's the playbook, plus what Google itself says about llms.txt and AI Mode.
How to Check If Your Brand Appears in ChatGPT, Perplexity, and Gemini
AI answers change between runs, between sessions, and between platforms -- yet every guide tells you to ask ChatGPT once and count. A single check is a coin flip. This is the sampling method: a defined prompt set across ChatGPT, Perplexity, and Gemini, run in a fresh session N times, scored on mention rate, share of voice, and citation sources.