How to Measure Share of Voice in ChatGPT and Perplexity

What share of voice means in AI answers

Share of voice in ChatGPT and Perplexity measures how often your brand appears when people ask AI engines for recommendations, comparisons, or shortlists in your category. It is not the same as classic SEO rank tracking. AI answers can mention several brands, cite different sources, and change wording across runs.

A useful definition is simple. For a fixed set of buyer prompts, measure the percentage of answers where your brand is recommended, mentioned, or cited compared with competitors.

The goal is not to chase one perfect number. The goal is to see whether AI engines understand that your brand belongs in the answer, why they include or exclude it, and which sources shape that decision.

Why ChatGPT and Perplexity behave differently

ChatGPT and Perplexity both generate natural language answers, but they do not pick brands in the same way.

Perplexity is strongly retrieval based. It searches the web, selects sources, summarizes them, and usually shows citations. If your brand is absent from authoritative pages, recent reviews, comparison pages, analyst content, or high quality category articles, Perplexity may not include you even if your own site is strong.

ChatGPT depends on the mode being used. Without web browsing, it leans on model knowledge learned during training. With search or browsing features, it can retrieve fresh pages and incorporate current information. That means ChatGPT share of voice can come from two places, model memory and retrieved evidence.

This matters because you should measure three things separately:

• Answer presence, whether the brand appears in the AI response
• Recommendation presence, whether the brand is actively suggested as a good choice
• Citation presence, whether a source mentioning the brand is used to support the answer

A brand can be cited but not recommended. It can be mentioned as an alternative but not a leader. Those distinctions affect the score.

Build a prompt set that reflects real buying journeys

Start with prompts that match how customers ask AI engines for help. Do not only track head terms like best CRM software or best payroll provider. Include the decision context that changes the answer.

Create prompts across these groups:

• Category discovery, such as best tools for customer support automation
• Use case prompts, such as best AI chatbot for ecommerce returns
• Segment prompts, such as best accounting software for small agencies
• Comparison prompts, such as alternatives to a named competitor
• Constraint prompts, such as tools with SOC 2, EU hosting, or Shopify integration
• Local or regional prompts, if location affects the market

Use neutral phrasing. Avoid prompts that mention your brand unless you are measuring branded comparison visibility. If you only ask leading questions, you will inflate your share of voice and miss the real gap.

A good starter set has 30 to 100 prompts. Larger teams often track several hundred, split by market, product line, and funnel stage.

Decide how you will score each answer

Use a scoring model before collecting data. Otherwise, every answer becomes a judgment call.

A practical scoring system looks like this:

• 3 points, brand is recommended as a top option
• 2 points, brand is included in a shortlist but not clearly endorsed
• 1 point, brand is mentioned only in passing
• 0 points, brand is absent
• Negative flag, brand is mentioned with a clear warning or outdated limitation

Then calculate share of voice like this:

Brand score across all prompts divided by total possible score across all prompts.

If you use 50 prompts and the maximum score is 3 per prompt, the total possible score is 150. If your brand earns 45 points, your weighted share of voice is 30 percent.

You should also track unweighted presence. That is the percentage of prompts where the brand appears at all. Weighted share tells you quality of visibility. Presence tells you coverage.

Run tests repeatedly and control the variables

AI answers are not fully deterministic. One run can include your brand while another leaves it out. To reduce noise, run each prompt more than once.

For each prompt, record:

• Engine, such as ChatGPT or Perplexity
• Date and time
• Prompt text
• Location or language setting, if relevant
• Full answer text
• Brands mentioned
• Brand position in the answer
• Citations and source URLs
• Recommendation score
• Notes on outdated or incorrect claims

Run at least three trials per prompt for important categories. Weekly tracking is usually enough for most teams. Daily tracking can help during launches, PR campaigns, or major site changes.

Keep prompts stable over time. If you rewrite the prompt set every week, trend data becomes useless. Add new prompts when needed, but label them as new.

Measure competitors and cited sources, not only your brand

AI visibility is relative. If ChatGPT or Perplexity does not recommend you, it is recommending someone else. Track those competitors by prompt cluster.

Look for patterns:

• Which competitors appear in top recommendations most often
• Which competitors are cited by Perplexity sources
• Which review sites, blogs, forums, or publisher pages influence answers
• Which product claims appear repeatedly
• Which buyer segments trigger your exclusion

This is where the mechanics become actionable. If Perplexity repeatedly cites a third party list that omits your product, your issue may not be your website. It may be that the page shaping the answer has incomplete market coverage. If ChatGPT describes your product with old positioning, your issue may be stale public information across the web.

Seeno can help here by tracking prompts across AI answer engines and showing which brands and sources are winning when yours is missing.

Turn the measurement into a useful report

A good AI share of voice report should answer five questions:

1. Where do we appear?
2. Where are we recommended?
3. Where are we absent?
4. Who appears instead?
5. Which sources seem to shape the answer?

Report results by prompt cluster, not only as one blended metric. A single overall score can hide the fact that you win enterprise prompts but lose small business prompts, or that you appear in comparison prompts but not category discovery prompts.

Use a simple dashboard structure:

• Overall weighted share of voice
• Presence rate
• Top recommended competitors
• Share by prompt cluster
• Citation share by source domain
• Biggest gains and losses since last period
• Priority content or PR actions

Common measurement mistakes

Avoid these traps:

• Using only one prompt and treating it as the market
• Counting any mention as a recommendation
• Ignoring citations in Perplexity
• Mixing branded and non branded prompts in one score
• Running tests once and overreacting to random variation
• Looking only at your own site instead of the sources AI engines cite

The best measurement system is boring and repeatable. Same prompts, same scoring rules, same cadence, clear competitor tracking.

What to do after you measure

Once you know your gaps, improve the evidence AI engines can find. Update product pages with clear category language. Publish comparison and use case content that reflects real buyer questions. Make sure trusted third party sources describe your product accurately. Earn mentions in pages that Perplexity and other retrieval based systems already cite.

Share of voice in ChatGPT and Perplexity is not a vanity metric when measured correctly. It is a map of how AI engines understand your market, who they trust, and where your brand is missing from the answer.