AI Share of Voice: Measure Your Brand vs Competitors

AI share of voice is the percentage of brand mentions you get versus your competitors across AI answer engines like ChatGPT, Perplexity, Gemini, and Google AI Overviews. You measure it by running a fixed set of category prompts, counting how often each brand shows up, then dividing your mentions by the total. The basic formula is: your brand mentions divided by total brand mentions across the prompt set, times 100.

That's it. That's the whole metric. But the way you set it up (and the way you read the result) is where most people get it wrong.

Let me walk you through it like a friend who's been staring at these dashboards for too long.

Quick answer

AI share of voice (AI SOV) = (your brand mentions / total brand mentions across a fixed prompt set, per engine) x 100.

• Run the same set of prompts on each AI engine. Don't blend them into one score.
• Below 15% usually means you've got a citation gap. The AI rarely brings you up.
• 25% to 40% is competitive. You're in the conversation.
• Above 40% is strong. You're a default answer.
• Even category leaders rarely clear 60%. If you see 80%, your prompt set is probably too narrow.

Track it per engine, watch it over time, and slap a confidence interval on it so you don't panic over noise. More on all of that below.

What is AI share of voice, really?

Old-school share of voice came from advertising and PR. It measured your slice of total ad spend or media coverage in a category. AI share of voice is the same idea, just moved to where people actually ask questions now: AI chatbots and answer engines.

Here's the shift. People used to Google "best CRM for small business" and scroll ten blue links. Now a big chunk of them ask ChatGPT or Perplexity and read one answer. If your brand isn't in that answer, you basically don't exist for that buyer. AI-driven referral traffic to US retail sites jumped 4,700% year over year, according to Adobe Analytics data cited by Nightwatch. And 58% of consumers say they've used AI tools to research products, per ChannelEngine's 2026 Marketplace Shopping Behavior Report.

So AI share of voice answers one blunt question: when a buyer asks an AI about your category, how often does the AI mention you instead of the other guys?

It's not the same as your website traffic. It's not your SEO ranking. A brand can rank #1 on Google and still get ignored by ChatGPT. The two systems pick winners differently.

AI SOV vs brand mention rate vs citation rate

People mix these up constantly, so let's keep them straight.

• Brand mention rate is the percentage of AI responses in your category that mention you at all. If you run 100 prompts and you show up in 17, your mention rate is 17%.
• AI citation rate is similar but stricter. It counts when the AI not only mentions you but links to or sources your content. Citations come with a clickable link. Mentions might just be your name in a sentence.
• AI share of voice is competitive. It's your mentions as a fraction of everybody's mentions. It tells you the split, not just the raw count.

Why does this matter? Because you can have a decent mention rate and still lose. If you show up in 17% of answers but your top competitor shows up in 60%, your share of voice is tiny. Mention rate is your absolute presence. Share of voice is your relative power.

The exact AI share of voice formula

Here's the clean version:

AI Share of Voice = (Your brand mentions / Total brand mentions across all tracked brands) x 100

Run it per engine. So you'll get a ChatGPT number, a Perplexity number, a Gemini number, and an AI Overviews number. Each one stands alone.

Quick example. You run 50 category prompts on ChatGPT. Across those 50 answers, all brands combined get mentioned 200 times. Your brand gets mentioned 50 of those times. Your ChatGPT AI share of voice is 50 / 200 x 100 = 25%.

LLM Pulse spells out the same math: if AI models mention brands 100 times total and your brand accounts for 25 of them, your SOV is 25%.

There's a fancier version some tools use, the citation-weighted SOV, which counts citations instead of plain mentions:

AI SOV = (Your brand citations across all models for the query set) / (Total citations for all tracked brands) x 100

AuthorityTech lays this one out. Use mentions when you care about awareness. Use citations when you care about who's actually getting the click and the source credit. I'd track both if you can. They tell different stories.

And there's a revenue version too: AI-attributed revenue divided by total category AI-driven revenue. That's the holy grail, but most teams can't attribute revenue to AI cleanly yet, so start with mentions.

How to build your prompt set (this is the whole game)

Your AI share of voice is only as good as the prompts you measure it with. Cherry-pick prompts where you already win and you'll fool yourself. Pick a fair, fixed set and you'll get a number you can actually trust.

Here's how to build one that doesn't lie to you.

Start with 10 to 20 core prompts. Nightwatch recommends 10 to 20 category prompts as a baseline. I'd push toward 30 to 50 once you're serious, because more prompts means a tighter confidence interval (we'll get there).

Cover the buyer's whole journey. Mix the intent types:

• Informational: "what is generative engine optimization"
• Commercial: "best AI citation tracking tools"
• Navigational / comparison: "Brand X vs Brand Y"
• Problem-based: "how do I track if ChatGPT recommends my brand"

The best GEO tools let you segment by these intent buckets, and you should. Your share of voice on commercial prompts ("best X") matters way more for sales than your share on broad definitional prompts.

Freeze the set. Once you pick your prompts, don't keep swapping them. If you change the prompts every month, you can't tell if your number moved because you got better or because you asked different questions. Lock it. Measure the same thing over time.

Phrase prompts the way real humans do. Nobody types "enumerate the premier solutions in the category." They type "what's the best tool for tracking AI mentions." Write prompts that sound like your actual buyers, because the AI answers those differently than keyword-stuffed robot prompts.

2026 AI share of voice benchmarks

Okay, the part you scrolled down for. What's a good number?

Short version: there's no single magic threshold, because it depends on how crowded your category is. But here are the working benchmarks people are using in 2026.

By competitive position

AuthorityTech's 2026 ranges break down like this:

• Category leaders: 25% to 45% on their best engine. In tight, concentrated markets, leaders hit 35% to 50%.
• Challengers: 8% to 20%.
• New entrants: under 5% for the first two or three quarters. That's normal. Don't freak out.

By market shape

This is the nuance most people skip. The same percentage means totally different things depending on how many competitors you have.

• Duopoly (2 real players): 50% share of voice is just parity. You're tied. Below 30% in a two-horse race means you're losing badly.
• Fragmented market (10+ players): 15% share of voice can mean you're the category leader. There's just more pie to split.

LLM Pulse makes the same point: in fragmented markets, 15%+ signals strong positioning, while in concentrated ones you need 35% to 50% to lead.

The simple bands we use

For most brands, here's a clean way to read your per-engine number:

• Below 15%: citation gap. The AI mostly doesn't mention you. This is the alarm zone.
• 15% to 25%: present but weak. You show up, but you're not a default.
• 25% to 40%: competitive. You're a real option in the answer.
• Above 40%: strong. You're often a top recommendation.
• Above 60%: rare, even for leaders. If you're seeing this, double-check your prompt set isn't rigged in your favor.

That last point is real. When eight major AI models were tested, they agreed on the top recommendation only 43.9% of the time. The AIs disagree with each other constantly. So if one engine hands you 80% share of voice, either you genuinely dominate a narrow niche or your prompts are too soft.

A reality check for B2B: most B2B brands appear in under 30% of relevant category queries, no matter how good their SEO is. So if you're sitting at 20% across a fair B2B prompt set, you're doing fine.

Why you must measure each engine separately

Here's the thing nobody tells you upfront. A blended "average AI share of voice" score is mostly garbage. The engines behave so differently that averaging them hides the truth.

How different? Averi analyzed 680 million AI citations in early 2026 and found only 11% of domains cited by ChatGPT were also cited by Perplexity. Whitehat SEO ran an independent study of 118,000 responses and got the exact same 11% number. Two different studies, same answer. The platforms barely overlap on who they cite.

And the brands themselves swing wildly per engine. AuthorityTech reported one brand sitting at 28% to 38% on Perplexity but only 10% to 16% on ChatGPT. Same brand. Same category. Totally different visibility depending on which AI you ask.

The engines also pull from different kinds of sources:

• ChatGPT leans hard on earned media. Around 51.1% of its citations go to earned media.
• Perplexity loves Reddit. Roughly 46.5% of its citations point at Reddit.
• Google AI Overviews loves YouTube, with about 29.5% citation share going there.
• Claude prefers long-form editorial, the kind you'd find in publications like The Atlantic or The Economist.

So if you only chase one playbook, you'll win on one engine and tank on the others. Measure each one. Fix each one. The work to climb on Perplexity (get talked about on Reddit) is not the work to climb on AI Overviews (get on YouTube and structured pages).

One more useful number. The average brand mention rate across AI responses is just 17.2%. The bar is genuinely low. Most brands are basically invisible. That's bad news if you're one of them and great news if you decide to actually do something about it.

Confidence intervals: stop chasing noise

This is the part that separates real measurement from vibes.

When you run a prompt set, you're taking a sample. AI answers are non-deterministic, which is a fancy way of saying the same prompt can give you a slightly different answer each time. Ask ChatGPT "best CRM" three times and you might get three slightly different brand lists. So your measured share of voice has wiggle in it.

If you run 20 prompts and your share of voice comes back at 25%, that doesn't mean it's exactly 25%. The true value might be anywhere from, say, 18% to 32%. That range is your confidence interval.

Why you care: imagine your number drops from 25% to 22% one month. Panic? No. If your confidence interval is plus or minus 7 points, a 3-point move is noise. Nothing happened. But if you ran 200 prompts and your interval is plus or minus 2 points, then a drop from 25% to 22% is real and worth investigating.

Two rules of thumb:

1. More prompts = tighter interval. Twenty prompts gives you a wide, sloppy range. A hundred gives you something you can act on. This is why a serious prompt set beats a cute little one.
2. Repeat each prompt. Run every prompt a few times (3 to 5 runs) and average, because a single run of a non-deterministic model is one roll of the dice. Multiple runs smooth out the randomness.

When you compare yourself to a competitor, check whether the intervals overlap. If you're at 25% (plus or minus 5) and they're at 28% (plus or minus 5), those ranges overlap. You're statistically tied. Don't write a strategy memo about a gap that isn't real.

This is exactly why a good AI citation tracking tool reports a confidence interval next to every number instead of a single bold percentage that pretends to be precise. A number without a range is half a number.

How to actually track AI share of voice

Let's get practical. Here's the workflow, start to finish.

1. Pick your competitor set. List the 4 to 8 brands you actually compete with. These are the ones whose mentions count toward "total brand mentions" in the formula. Too few and your SOV looks inflated. Too many and it gets diluted into mush.

2. Build and freeze your prompt set. 30 to 50 prompts, mixed intent, phrased like real humans. Lock it.

3. Run it across all five engines. ChatGPT, Perplexity, Gemini, Google AI Overviews, and Microsoft Copilot at minimum. Add Claude and Grok if they matter to your audience. Run each prompt multiple times.

4. Count mentions per brand, per engine. Then apply the formula. You'll end up with a grid: brands down the side, engines across the top.

5. Add confidence intervals. Based on how many prompts and runs you did. Report the range, not just the point.

6. Re-run on a schedule. Monthly is fine for most. Weekly if you're in a fast-moving fight or running an active GEO campaign. Improvements from GEO work usually show up in 60 to 90 days, so don't expect overnight jumps.

You can do this by hand with a spreadsheet and a lot of patience. Or you can use a tool that runs the prompts, counts the mentions, computes the intervals, and tracks competitor share of voice for you. Doing it manually across five engines, multiple runs, and 50 prompts every month gets old fast. That's roughly 800 to 1,000 AI queries a month just to keep one chart honest.

How to raise your AI share of voice

Measuring is step one. Moving the number is the point. Quick hits that work in 2026:

• Get mentioned in earned media. ChatGPT pulls heavily from it. A few good press mentions and roundup inclusions do real work. Earned and news media made up 39.5% of all AI citations in the March to April 2026 window.
• Show up on Reddit and forums for Perplexity. Real, helpful threads where your brand comes up naturally. Not spam. The AI can smell spam.
• Publish answer-first content. Lead with the direct answer, then the detail. AI engines lift clean, quotable answers. Roughly 50% of cited content was published within the last 13 weeks, so freshness matters. Keep publishing.
• Add structured data and an llms.txt file so engines can parse and trust your pages.
• Get on YouTube for Google AI Overviews, which sends nearly 30% of its citations there.

Notice these are different moves for different engines. That's the whole reason you measure per engine in the first place.

Common mistakes that wreck your numbers

• Blending engines into one score. You already know why. They barely overlap.
• Tiny prompt sets. Ten prompts is a starting point, not a measurement system. Your interval will be too wide to trust.
• Swapping prompts every month. You lose your baseline. Freeze the set.
• Ignoring confidence intervals. You'll chase ghosts and celebrate noise.
• Picking the wrong competitor set. Include brands you don't really compete with and your SOV looks artificially small.
• Only running each prompt once. One roll of the dice. Run it several times.