AI Share of Voice: How to Measure Your Brand's Visibility in AI Assistants

What AI Share of Voice means (and how it differs from traditional SOV)

Traditional Share of Voice measures how much of the conversation in a market your brand owns — ad impressions, social mentions, search visibility relative to competitors. The data comes from structured sources: Google Ads auction data, social listening platforms, SEO ranking tools.

AI Share of Voice is different in three ways.

First, there is no fixed index. Google ranks pages in a stable list. AI assistants generate a new response every time, and the brands they mention can change between sessions. Your AI Share of Voice on Monday may differ from Tuesday because the AI retrieved different sources.

Second, AI responses are not links — they are recommendations. When ChatGPT says “I'd recommend HubSpot for your CRM needs,” that carries different weight than a Google search result showing HubSpot's homepage at position 3. The user is more likely to act on a direct recommendation from an AI assistant than to click a ranked link.

Third, different AI platforms pull from different sources. ChatGPT uses Bing's index plus its training data. Perplexity runs its own web crawler with heavy Reddit weighting. Gemini uses Google's index. Measuring on one platform gives you an incomplete picture.

For a full overview of how AI visibility works across three layers, see: What Is AEO: Complete Guide.

Define the metrics that matter

AI Share of Voice is not a single number. Track these four metrics separately:

Mention vs recommendation matters. An AI response might say “Companies in this space include [Brand A], [Brand B], and [Brand C]” — that is a mention. An AI response that says “For your specific use case, I'd recommend [Brand A] because...” — that is a recommendation. Recommendations carry more conversion weight.

Position matters. The first brand named in an AI response gets disproportionate attention. Being mentioned third in a list of five is better than not appearing, but worse than being first.

Build your query set

Your measurement is only as good as the queries you test. Choose 10–20 queries that represent how your potential customers actually ask AI assistants for help.

Three types of queries to include:

1. Category queries (5–8): “What's the best [your category] for [use case]?” These test whether AI includes you when someone asks about your market.
2. Problem queries (3–5): “How do I [solve problem your product addresses]?” These test whether AI cites your content or recommends your solution.
3. Brand comparison queries (2–4): “[Your brand] vs [competitor]” or “Is [your brand] good for [use case]?” These test what AI knows about you specifically.

Avoid testing queries nobody asks. If you sell project management software, test “best project management tool for remote teams” — not “enterprise agile workflow orchestration platform.” Use the language your customers use.

For a step-by-step guide on choosing queries and running your first check: How to Check If Your Brand Is Recommended by ChatGPT.

Measure manually: the free approach

You do not need a paid tool to measure AI Share of Voice. The manual approach takes 2–3 hours for a thorough baseline.

For each of your 10–20 queries:

1. Open ChatGPT (with web search on). Ask the query. Record: Which brands are mentioned? Is yours among them? What position? Is the description accurate?
2. Repeat on Perplexity. Record the same data, plus which sources it cites (Perplexity shows inline citations).
3. Repeat on Google Gemini.
4. Repeat on Claude.
5. Repeat on Microsoft Copilot.

Calculate your metrics:

• Mention rate = (number of queries where you appeared ÷ total queries) × 100, per platform
• Recommendation rate = same calculation, but only counting direct recommendations
• Average position = sum of positions ÷ number of appearances
• Context quality = manual assessment per response

Time required: About 2–3 minutes per query per platform. 15 queries × 5 platforms = 75 tests = roughly 2.5 hours.

The manual approach gives you the most accurate picture because you can assess context quality, something no automated tool measures well. The tradeoff is time.

Measure by layer: parametric, contextual, fresh session

A mention on ChatGPT with web search on tells you that AI can find your brand. It does not tell you whether AI knows your brand from training data or discovered it through search. That distinction matters because the fixes are different.

We use a three-layer measurement approach:

Layer 1 — Parametric knowledge (training data)

Test with web search turned OFF. This shows what the AI model has internalized about your brand.

• Open ChatGPT. Disable web search.
• Ask your category queries.
• Record: Does AI mention you? Is it accurate?

If you appear here, your brand is in the model's training data. This is the most durable form of visibility — it persists across sessions and does not depend on your current web presence.

Layer 2 — Contextual web search

Test in a conversation where you have already established context about your category.

• Open ChatGPT with web search on. Start with: “I'm looking at [your category] options for [use case].”
• Ask follow-up queries.

If you appear here but not in Layer 1, AI finds you through search when it has context clues.

Layer 3 — Fresh session retrieval

Test in a fresh conversation with no prior context.

• Open a new chat. Ask your queries cold.

If you appear here but not in Layer 1, AI finds you through search even without context. This depends on your current web presence, reviews, and content freshness.

Why separating layers matters

• Layer 1 gap → you need more mentions in training data sources (Wikipedia, press, Reddit, publications). Timeline: 3–12 months.
• Layer 2 gap → your content structure or topical authority needs work. Timeline: 2–8 weeks.
• Layer 3 gap → your external signals (reviews, Reddit, comparison articles) need strengthening. Timeline: 1–4 weeks.

For a detailed optimization playbook for each layer: How to Get Your Brand Recommended by AI Assistants.

Compare results across AI platforms

Each AI platform retrieves and recommends differently. Your AI Share of Voice on ChatGPT may look nothing like your Share of Voice on Perplexity.

What this means for measurement:

• If you score well on Perplexity but poorly on ChatGPT, you likely have good web presence but weak parametric knowledge.
• If you score well on Gemini but poorly elsewhere, your Google SEO is strong but your broader web presence needs work.
• If you score poorly everywhere, start with external signals and technical access.

Use your tracking template

Use this spreadsheet structure to track measurements over time.

Tab 1: Query Results

Tab 2: Monthly Summary

Tab 3: Layer Breakdown

Track these monthly. Changes in Layer 3 appear fastest (1–4 weeks after optimization). Layer 1 changes only when models retrain (3–12 months).

Choose a measurement tool (or don't)

Several tools automate parts of AI Share of Voice measurement. None of them do everything the manual approach does.

When the manual approach is better:

• First-time baseline (you need to understand what AI actually says about you, not just whether it mentions you)
• Fewer than 15 target queries
• You need Layer 1 vs Layer 2/3 separation
• You need context quality assessment

When a tool is better:

• Ongoing weekly monitoring after you have a baseline
• 20+ target queries
• Multiple brands or product lines to track
• You need historical trend data

For a comparison of specific tools: Best AEO Tools to Monitor AI Visibility.

Avoid these measurement mistakes

1. Measuring only mention rate and ignoring recommendation rate. Being mentioned in a list of 10 competitors is not the same as being recommended as the top choice. Track both separately.
2. Testing on one platform and assuming the result applies everywhere. Your ChatGPT visibility has almost no correlation with your Perplexity visibility. They use different indexes, different source weights, and different citation logic. Always test at least three platforms.
3. Using branded queries as your primary test. Asking “What is [Your Brand]?” tells you what AI knows about you. It does not tell you whether AI recommends you when someone asks a category question. Category and problem queries matter more.
4. Not controlling for web search settings. If you only test ChatGPT with web search on, you cannot tell whether your visibility comes from training data or real-time search. Test with web search off first, then on.
5. Measuring once and treating it as permanent. AI responses shift when models update, when competitors optimize, and when your content ages. A single measurement is a snapshot, not a verdict. Track monthly at minimum.
6. Ignoring what AI says about you and only tracking whether it mentions you. An AI response that describes your product inaccurately or positions you for the wrong audience can hurt your brand. Always assess context quality.
7. Comparing raw numbers across platforms without context. A 30% mention rate on Perplexity (which always searches the web) is not equivalent to a 30% mention rate on ChatGPT with web search off (which only uses training data). Compare within layers, not across them.

Set your measurement cadence

Weekly: If you are actively optimizing (publishing new content, building review profiles, restructuring pages), track Layer 3 weekly on your top 5 queries across 2–3 platforms. This catches fast-moving changes.

Monthly: Run the full measurement — all queries, all platforms, all three layers. Update your tracking template. Compare to the prior month.

After AI model updates: When a major model update is announced (GPT-5, Claude 4, Gemini 2.0), re-run Layer 1 tests immediately. Parametric knowledge changes only with retraining, and model updates are the trigger.

Quarterly: Run a full competitive analysis (Step 8 from our AEO audit guide). Check whether competitors have improved their AI visibility and adjust your priorities.

Ask your customers. Add a question to your sales process or onboarding survey: “Did you use an AI assistant while researching us?” This gives you qualitative data that no measurement tool captures.

When to use tools vs when to hire an expert

Use a tool when:

• You need ongoing automated tracking
• Your query set is stable (not changing month to month)
• You have internal capacity to interpret the data and act on it
• Budget: $100–500/month

Hire an expert when:

• You need a scored baseline with the three-layer breakdown
• You need to understand why AI recommends competitors instead of you
• You need a prioritized action plan, not just data
• You want before/after benchmarking to prove ROI

The progression that works for most brands: start with a manual baseline (free, 2–3 hours), then decide whether you need a tool for monitoring, an expert audit for strategy, or both.

Next steps

Run your first manual AI Share of Voice check using our guide: How to Check If Your Brand Is Recommended by ChatGPT.

If you find gaps, the next step depends on which layer is weakest: How to Get Your Brand Recommended by AI Assistants.