AI Share of Voice: How to Measure Brand Visibility

AI share of voice measures how often your brand appears in AI answers compared with competitors. It is the AI-search version of market presence: not where you rank, but whether the model names, cites, or recommends you.

The metric matters because AI answers collapse the SERP into one response. If your competitor appears and you do not, your classic Google rank may not matter for that buyer.

This guide covers mention share, citation share, weighted share of voice, and the reporting mistakes that hide platform-level gaps. Use it alongside LLM visibility tools and the ChatGPT visibility tracker. Share of voice is one metric inside broader AI visibility tracking.

The death of traditional SoV

Traditional Share of Voice was a game of probabilities.

• If you ranked #1, you got ~30% of clicks.
• If you ranked #3, you got ~10%.
• If you bought ads, you bought impressions.

You could buy your way into the conversation. AI Share of Voice is a game of binaries. AI engines are deterministic in their output but probabilistic in their reasoning. When ChatGPT generates a response, it typically recommends 3-5 options at most.

If there are 50 competitors in your niche, 45 of them effectively cease to exist in that conversation. The long tail is cut off. You cannot buy a sponsored mention in a ChatGPT organic response (yet). You have to earn the model’s trust.

What is Share of Model (SoM)?

Share of Model (SoM) is the frequency with which a specific Large Language Model (LLM) cites, recommends, or references your brand in response to relevant non-branded queries.

It’s a measure of brand salience within the neural network: whether the model “knows” you at all, and whether it associates you with the solution.

Despite the different name, SoM is not a separate metric from the AI share of voice defined below: both are the same mentions-divided-by-total-tracked-mentions calculation, and cloro’s own AI Visibility Leaderboard methodology uses “share of voice” for exactly this number. This guide uses “Share of Model” and “AI share of voice” interchangeably from here on. The real contrast is with traditional, pre-AI Share of Voice, covered above, which used a different formula entirely.

Example:

• Query: “Best email marketing tools for creators”
• Response: “ConvertKit and Mailchimp are popular options…”
• SoM Analysis: ConvertKit and Mailchimp have high SoM. ActiveCampaign, effectively absent, has low SoM.

How to calculate share of voice in AI search answers

This is the number every share-of-voice dashboard boils down to: what fraction of the conversation about your category your brand is winning. The formula: share of voice = your brand’s mentions ÷ total mentions across every tracked brand, counted over the same set of AI search answers. Step by step:

1. Fix a prompt set and a brand panel. The prompts are your category’s buying questions; the panel is your brand plus the competitors you count.
2. Run the prompts on a schedule and collect every answer.
3. Count mentions per brand across the same answers, your brand and every panel brand alike.
4. Divide your brand’s mentions by the panel’s total mentions. If your brand is named in 40 answers and the panel’s brands collect 200 mentions between them, your share of voice is 20%.

Two properties of the denominator matter: it sums only across brands you chose to track, so the metric rescales whenever the panel changes (a jump after adding a competitor is an artifact), and filtering to your own brand collapses it to 100%. Keep the panel fixed inside any window you trend.

The three dimensions of AI visibility

Measuring SoM isn’t just counting mentions. It requires a closer look at how you are mentioned. The AI Visibility Leaderboard reports share of voice this way across twelve software categories each week, if you want to see the dimensions separated on real answers before building your own.

1. Mention frequency (volume)

The raw number of times your brand appears across a set of relevant queries.

• Metric: “We appeared in 40 out of 100 ‘Best CRM’ queries.”

2. Share of recommendation (rank)

Being mentioned is good. Being recommended is better.

• Metric: “We were the #1 recommended tool in 15% of queries, and in the top 3 for 60%.”

3. Sentiment and context (quality)

Is the AI praising you or warning users about you?

• Metric: “80% of mentions highlighted our ‘ease of use’, but 20% flagged ‘high pricing’.”

A worked example: calculating your SoM

Imagine you sell project-management software and you want to know your Share of Model against three main competitors.

Step 1: Build a query set. Pick 50 non-branded prompts a prospect might realistically type into ChatGPT. Examples:

• “Best project management tool for remote teams”
• “Asana vs alternatives for small businesses”
• “Agile project management software with time tracking”

Step 2: Run each prompt 5 times across 4 models (ChatGPT, Claude, Gemini, Perplexity). That gives 50 × 5 × 4 = 1,000 responses. Run each prompt multiple times because LLM outputs are stochastic; sampling reduces noise.

Step 3: Tally mentions. From the 1,000 responses, an illustrative tally:

Brand	Total mentions	Top-3 recommendations	First mention
Asana	612	441	198
Monday.com	548	389	172
ClickUp	401	267	89
Your brand	218	118	34
Others (long tail)	821	285	7

Step 4: Compute Share of Model.

SoM (mention-weighted) = (your mentions) / (sum of mentions across tracked brands)
                       = 218 / (612 + 548 + 401 + 218)
                       = 218 / 1779
                       ≈ 12.3%

Step 5: Compute Share of Recommendation (more meaningful for purchase intent):

SoR = (your top-3 appearances) / (sum of top-3 appearances)
    = 118 / (441 + 389 + 267 + 118)
    = 118 / 1215
    ≈ 9.7%

Step 6: Compute Share of First Mention (the “the” answer, almost always the buyer’s default):

SoFM = 34 / (198 + 172 + 89 + 34)
     = 34 / 493
     ≈ 6.9%

Reading the result. This brand has decent presence (12.3%) but loses on the conversion-weighted slots. It’s rarely the first name an LLM reaches for (6.9%). The strategic implication is more digital PR aimed at “Best X for Y” listicles, because those are the inputs that move the first-mention metric.

Refresh cadence. We re-run this monthly. Weekly is overkill (LLM weights move slowly); quarterly is too slow to catch the impact of a campaign.

Measurement approaches compared

Before you pick one, there is a normalization problem that invalidates most naive SoM numbers. Across roughly 13,000 AI answers from six engines, our LLM citations study found citation depth varies about twentyfold: Google AI Mode averages 17.7 sources per answer, ChatGPT, Gemini and Copilot land between 4 and 8, and Perplexity averages just 1.3, frequently zero. Depth is topic-gated too, with Perplexity at 2.3 sources for commerce and 0.0 for dining.

A share-of-model figure is a ratio whose denominator is that source count, so the same brand performance produces wildly different percentages by engine. An engine that cites 1.3 sources will hand a mentioned brand a huge share and an unmentioned one a zero, while a 17.7-source engine spreads share thin across many domains. Pooling engines into one headline SoM without weighting by depth mostly measures which engines you sampled. The takeaway: report per engine, and treat cross-engine comparison as a normalization exercise rather than a straight average.

There are four practical ways to measure SoM. Most teams need a blend.

Approach	Coverage	Cost / month	Effort	Best for
Manual prompting (spreadsheet)	20–50 queries × 1 model	$0	High (weekly)	Validating that automated tools are accurate
Custom script + OpenAI API	100–500 queries × 2–3 models	$50–$200	Medium (build once)	Engineering teams with capacity
Dedicated AI visibility tracker	500–5,000 queries × 4+ models	$100–$1,000	Low	Marketing teams without a data engineer
Brand-monitoring vendor add-on	Often shallow (mentions only)	$200+	Low	Enterprises that already pay for the suite

Our recommendation: start with 2 weeks of manual prompting on 30 high-intent queries to set a baseline and surface obvious gaps. Then move to an automated tracker so you measure the same query set the same way every month. Comparing across measurement methods is the fastest way to lose trust in the metric.

Measuring the unmeasurable

Traditional SEO tools are blind here: Google Search Console cannot see inside Claude’s context window. Measuring AI Share of Voice means simulating thousands of conversations to map the model’s latent knowledge.

You can do that by hand, typing 50 queries into ChatGPT and tallying the results in a spreadsheet. That is slow, biased, and unscalable. The alternative is a dedicated AI visibility tracker like cloro, which automates the “mystery shopper” process for AI: it runs thousands of queries across different models (GPT-4, Claude 3.5, Perplexity) and calculates your Share of Voice relative to competitors, so marketing and growth teams get a monthly benchmark instead of a manual spreadsheet exercise.

The winner-takes-all dynamic crushes the middle market

The most dangerous aspect of AI search is the consolidation of authority. In traditional search, users might click to the 2nd page of Google to find a niche provider. In AI search, users rarely ask “Give me 10 more options.” They accept the first answer and move on, which creates a winner-takes-all dynamic.

• The top 3 brands get 90% of the visibility.
• The middle market gets crushed.

If you’re a challenger brand, your priority in 2026 is breaking into that top tier of AI recollection.

How to increase your AI share of voice

You cannot edit the model’s weights directly. But you can influence the data it consumes.

1. Surround the sound

AI models read the entire web: Reddit threads, G2 reviews, TechCrunch articles, YouTube transcripts. To increase SoM, you must be present where the model learns.

• Action: a digital PR campaign that gets your brand cited in high-authority lists (“Best X for Y”).

2. Define yourself

Don’t let the AI guess what you do. Tell it.

• Action: Implement llms.txt on your site. Provide a clear, concise definition of your value proposition in a machine-readable format.

3. Data as a moat

Publish proprietary data. When you become the source of a statistic, you force the AI to cite you.

• Action: Publish an annual “State of the Industry” report. When users ask “What are the trends in X?”, the AI will cite your data.

4. Technical accessibility

Ensure AI crawlers can actually read your site. If GPTBot is blocked, your newest features do not exist to the model.

Common pitfalls in SoM measurement

We’ve audited dozens of SoM dashboards. The same handful of mistakes repeatedly produce numbers that look great but predict nothing.

1. Tracking branded queries instead of category queries. “What is [Brand X]?” will always return your brand. Of course you have 100% SoM. Track only non-branded queries a prospect would use before they know you exist.

2. One run per query. LLMs are stochastic. Run the same prompt twice and you get different brands cited. A single run is noisy. We use 5 runs per prompt at minimum; teams reporting weekly should use 10.

3. Sampling only one model. Brands often dominate ChatGPT but disappear in Perplexity (or vice versa). A SoM number from a single model overstates your real visibility.

4. Counting any mention as equal weight. Being mentioned in a 12-item bullet list at position 11 is not the same as being the first sentence of the response. Weight by position, or split the metric (mentions vs. recommendations vs. first-mentions).

5. Drifting query sets. If you add new queries every month “to keep things fresh,” you can’t compare months. Lock the query set quarterly. Add new queries to a separate cohort.

6. Ignoring the long tail in the denominator. If you only count the top 4 brands, your share looks higher than reality. Include an “Others” bucket so the denominator reflects all citations.

The new dashboard for CMOs

In 2024, CMOs stared at Google Analytics dashboards showing “Organic Traffic.” In 2026, they’ll stare at dashboards showing “AI Share of Voice.”

The metric tells you something deeper than traffic: whether your brand has cultural relevance to the models your buyers are asking. If an AI trained on the sum of human knowledge never names you, the conversation about your category is happening without you.

Start tracking your AI Share of Voice today. Use cloro to benchmark where you stand against competitors and watch how your GEO efforts translate into real visibility.