How do you measure share of voice in AI-generated answers?

To measure share of voice in AI-generated answers, run a fixed set of buyer prompts across the AI engines you care about, record every brand each answer mentions, and divide your brand’s mentions by the total mentions of all tracked brands. Repeat runs to smooth out variation, and report the result per engine and per buying intent.

The basic formula

AI share of voice = your brand mentions / total mentions of all tracked brands

If, across 100 answers, your brand is mentioned 30 times and your four tracked competitors are mentioned 120 times combined, your share of voice is 30 / 150 = 20%.

Some teams also report mention rate (answers mentioning you / total answers) alongside it. Mention rate shows presence; share of voice shows presence relative to competitors. Both belong in a report, as covered in the most important KPIs for GEO.

Step-by-step method

flowchart TD
    A[Define competitor set] --> B[Build fixed prompt set by intent]
    B --> C[Choose engines and sampling frequency]
    C --> D[Run each prompt several times per engine]
    D --> E[Extract brand mentions, order and citations]
    E --> F[Calculate share of voice per engine and intent]
    F --> G[Review trends and cited sources monthly]

1. Define the competitor set

List 3 to 8 brands buyers compare you with. Include the category leader even if you rarely win against them, and any brand that engines keep naming on your prompts. Freeze the list for the quarter.

2. Build the prompt set

Write prompts the way buyers ask, grouped by intent:

Intent	Example prompt
Category	“What are the best API security tools for a fintech startup?”
Comparison	“Compare [competitor A] and [competitor B] for a 300-person company”
Alternatives	“What are alternatives to [competitor] with better pricing?”
Use case	“How can a DevOps team detect secrets in CI pipelines?”

Exclude prompts containing your own brand from share of voice. Aim for enough prompts per intent that one answer does not swing the total.

3. Choose engines and frequency

Pick the engines your buyers actually use, commonly ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Gemini and Microsoft Copilot. Sample at least weekly. Run each prompt more than once per sampling period, because the same prompt can return different brands on different runs.

4. Extract mentions consistently

For each answer, record:

• Every tracked brand mentioned (including name variants and product names).
• The order of mentions.
• Whether each brand was recommended, listed or only mentioned in passing.
• Which URLs were cited.

Manual extraction works for a small pilot. At scale, use a monitoring tool or a script with a consistent brand-matching dictionary.

5. Calculate and segment

Calculate share of voice:

• Overall.
• Per engine.
• Per intent.
• Per product line or persona, if relevant.

Segmenting matters. A healthy overall number can hide that you are absent from comparison prompts on ChatGPT, which may be where your buyers make shortlists.

6. Optionally weight by position

A first recommendation influences buyers more than a seventh listed option. A simple weighting scheme:

Position in answer	Weight
Primary recommendation or first named	1.0
Second or third named	0.7
Fourth or later	0.4
Passing mention	0.2

Weighted share of voice = your weighted mentions / all brands’ weighted mentions. Keep the weights fixed once you choose them.

Reading the results

• Low share of voice, competitors cited from third-party sites: your gap is off-site. Look at which review sites, listicles or forums engines cite and work on presence there.
• Low share of voice, competitors cited from their own sites: your gap is on-site content. They have clearer pages answering those prompts.
• High mentions, poor sentiment or wrong facts: fix accuracy on your pages and the cited sources.
• Big differences between engines: engines use different indexes and source preferences. Prioritize the engines your buyers use most.

Manual vs tool-based measurement

Approach	Good for	Limits
Manual checks in each engine	A first baseline, under 30 prompts	Slow, single runs, personalization bias
Spreadsheets plus API calls	Teams with engineering help	API answers can differ from consumer apps; maintenance
Dedicated monitoring tool	Ongoing tracking across many prompts and engines	Cost; methodology varies by vendor

Whichever you choose, document the method so the numbers stay comparable over time.

Common pitfalls

1. Too few prompts. Small sets produce noisy percentages.
2. Changing prompts every month. You end up measuring the prompt set, not your progress.
3. Counting only links. Many engines mention brands without linking. Count text mentions.
4. Ignoring name variants. Match product names, abbreviations and former brand names.
5. Personalized results. Logged-in sessions with memory can skew answers. Use clean sessions or APIs where possible.

Reporting share of voice to leadership

Show a trend line of share of voice against your top two competitors on high-intent prompts, plus the three sources most responsible for the gap. That is usually enough for a CMO to understand the problem and approve the fix. The guide to measuring AI share of voice has more reporting examples, and what an AI visibility score is explains how share of voice rolls up into a single score.

Tools such as GrackerAI automate prompt runs, brand extraction and share of voice across up to 10 AI engines, which removes most of the manual work once the method is defined.