How do I measure share of voice in AI-generated responses?

To measure share of voice (SOV) in AI-generated responses, you want to know:

“How often does my brand, product, or topic appear compared with competitors when AI systems answer relevant prompts?”

Because AI answers are dynamic and model-dependent, SOV is usually measured by repeatedly querying the model with a defined prompt set and then counting/weighting mentions.

1) Define what “voice” means

Decide what you’re measuring:

• Brand SOV: your brand vs competitors
• Category SOV: how often your brand appears in a category like “best project management tools”
• Sentiment-weighted SOV: mentions plus whether they’re positive/neutral/negative
• Citation SOV: whether the AI cites your site or sources when answering

2) Build a prompt set

Create a representative set of prompts users might ask, for example:

• “What are the best CRM tools for small businesses?”
• “Which skincare brands are best for sensitive skin?”
• “Compare [your brand] vs [competitor].”

Use variations:

• informational
• comparison
• recommendation
• troubleshooting
• purchase-intent

3) Run the prompts across the AI systems you care about

Test across:

• ChatGPT / GPT-based models
• Claude
• Gemini
• Perplexity
• Copilot
• custom assistants

For each prompt, collect:

• full response text
• model name/version
• date/time
• location/language, if relevant
• whether browsing/search is enabled

4) Extract mentions

Count:

• direct mentions of your brand
• mentions of competitors
• mentions in first position or top recommendations
• citations/links to your domain
• sentiment associated with each mention

You can do this manually for small sets or with NLP/LLM-based extraction for scale.

5) Calculate SOV

A simple version:

[ SOV = \frac{\text{Your mentions}}{\text{Your mentions + competitor mentions}} \times 100 ]

Examples:

• Your brand mentioned 40 times
• Competitors mentioned 60 times

[ SOV = \frac{40}{40+60} = 40% ]

You can also calculate:

• top-3 inclusion rate: % of prompts where you appear in the top 3 recommendations
• share of citations: your citations / total citations
• share of sentiment-weighted mentions: positive mentions weighted more than neutral/negative

6) Weight the results if needed

Not all mentions are equal. Consider weighting by:

• position in the answer
• recency of the response
• confidence or strength of recommendation
• sentiment
• whether the model explicitly recommends vs merely lists

Example:

• first mention = 3 points
• second/third = 2 points
• later mentions = 1 point

Then compute weighted SOV.

7) Track over time

AI outputs change frequently. Measure SOV:

• weekly or monthly
• after major content/product launches
• after SEO/PR campaigns
• by model version

This gives you a trend, not just a snapshot.

8) Watch for key limitations

AI SOV is tricky because:

• responses vary across runs
• models may hallucinate or omit brands
• personalization and location affect answers
• browsing-enabled models may depend on search rankings
• prompt wording can strongly change results

So always use:

• a consistent prompt set
• multiple runs per prompt
• averages, not single outputs

9) Useful metrics dashboard

A practical dashboard might include:

• Brand mention rate
• Competitor mention rate
• Top-3 recommendation rate
• Citation rate
• Positive/neutral/negative sentiment
• Share of voice by prompt category
• Share of voice by model

10) Example workflow

1. Choose 50 prompts in your category
2. Query 3 models, 3 times each
3. Extract brand/competitor mentions
4. Compute:
   • overall SOV
   • top-3 SOV
   • citation SOV
   • sentiment-weighted SOV
5. Compare month over month

If you want, I can also give you:

• a spreadsheet template
• a Python script to calculate AI SOV
• a sample KPI framework for monitoring it regularly