How to Measure Share of Voice in AI Search Answers

By UpGeo · 2026-07-18

Share of voice (SOV) in AI-generated search answers tells you how often your brand appears in the text of generative AI responses for a set of queries. You calculate it as: (your brand’s total mentions ÷ sum of all tracked brand mentions) × 100. It’s the key visibility metric for Generative Engine Optimization (GEO), and it shows exactly how often AI engines like ChatGPT, Perplexity, Gemini, Copilot, and Google AI Overviews mention you instead of a competitor.

Why AI Share of Voice Replaces Traditional Rank for Generative Search

Traditional rankings and click-through rates no longer give you the full picture. A mid‑2024 BrightEdge analysis found Google’s AI Overviews appearing on 15% of U.S. search queries. ChatGPT has already crossed 200 million weekly active users. Answers now appear directly inside the LLM’s response—no blue links needed. Being present in these answers matters as much as organic rankings once did, and share of voice in AI answers is the KPI for measuring it.

Step 1: Define Your Query Universe and Competitor Set

Start by listing the high-intent queries your business cares about. Group them by intent—informational (“what is …”), commercial investigation (“best …”, “compare …”), and transactional (“buy …”, “pricing”). For example, a CRM company might track “best CRM software,” “CRM vs ERP,” “free CRM tools,” and “CRM pricing for small business.” Next, pick 3–5 direct competitors that AI answers consistently pull in for these terms. A focused list of 20–50 queries keeps things manageable without blowing up your costs.

Step 2: Select the AI Platforms You’ll Monitor

Focus on the generative engines your audience actually uses. The main platforms right now:

• ChatGPT (via OpenAI API or the chat interface)
• Google AI Overviews (appearing in Google’s SERPs for informational queries)
• Perplexity (API available)
• Gemini (Google’s API for Gemini models)
• Copilot (Microsoft, accessible through Bing Chat)

APIs are ideal for programmatic measurement—you can run the same prompt repeatedly and parse structured JSON. Where an API isn’t available (Google AI Overviews), script browser queries using the &udm=14 search parameter to surface the overview block and extract the text.

Before you start monitoring, make sure AI bots can actually reach your site. Whitelist the relevant user agents from our list of AI crawlers. Also, give LLMs a curated map of your most important pages with an llms.txt file—read the llms.txt guide and create one quickly using our free llms.txt generator.

Step 3: Collect Response Data Reliably

Run each query 3–5 times during a measurement window (say, once a week or month) to account for the non‑deterministic nature of LLMs. Even with identical prompts, the engines can phrase answers differently and cite different brands each time. Averaging across runs gives you a stable SOV figure.

• Use a locked‑down prompt template—like “What is the best CRM software?”—with no extra instructions so comparisons stay fair.
• Log every full response, timestamp, and platform for auditing later.
• When querying via API, set temperature as low as possible (often 0) to reduce variability, but still expect some randomness.
• For Google AI Overviews, capture the entire summary block—not just the links—because brand mentions often hide in the synthetic text.

Step 4: Parse Brand Mentions and Context

Pull every occurrence of your brand name and competitor names from each saved response. A simple regex or Python script with a list of brand strings (include common misspellings and product names) does the job. Count every mention—if a brand shows up twice in one answer, count it twice.

If you want more nuance, you can tag each mention by its prominence:

• First mention – the brand the engine introduces first tends to carry the most recommendation weight.
• Bold/hyperlinked – in Perplexity or ChatGPT, bolded or linked citations signal higher authority.
• Direct recommendation – phrases like “we recommend X” or “the top choice is X.”

But starting with raw mention counts gives you a solid baseline SOV.

Step 5: Calculate Your AI Share of Voice

At the end of each measurement period, add up all brand mentions across your query set. Then calculate SOV for each brand:

(Brand mentions ÷ Total brand mentions) × 100

Say you tracked 10 CRM queries and found these total mentions across all responses:

Brand	Total Mentions	Share of Voice
Your Brand (AcmeCRM)	24	26.7%
Competitor A	45	50.0%
Competitor B	21	23.3%
Total	90	100%

That 26.7% is your single number to watch month over month. If you later add a weighted prominence score, you can create a secondary “weighted SOV” that values first‑mentioned or directly‑recommended brands more heavily.

Step 6: Track Trends and Act on Insights

Plot your SOV monthly. Drops often mean a competitor published new, authoritative content that LLMs are now citing. Gains usually follow after you’ve tightened up the technical foundation (llms.txt, structured data, explicit AI crawler access) and refreshed core pages. Use these trends to:

• Spot which topic clusters a competitor owns and close the content gap with better resources.
• Test small wording changes on key pages, then see if SOV rises in the next cycle.
• Connect SOV movements to downstream metrics (demo requests, free‑trial sign‑ups) so you can tie AI visibility to revenue.

Overcoming Common Measurement Pitfalls

Measuring SOV in LLM responses isn’t as predictable as scraping a SERP. Watch out for:

• Non‑determinism – Always query multiple times and average. Report a confidence interval if you run enough trials.
• Query drift – AI engines can interpret vague queries differently. Use concrete, consistently worded prompts and note when a platform switches its underlying model.
• False positives – A brand name might appear in a negative sentence (“unlike AcmeCRM, …”). If that matters, add a basic sentiment check or manually review a sample.
• API costs & rate limits – Plan your measurement cadence accordingly. Even 20 queries across 3 platforms weekly can work with free tiers or low‑cost API calls.
• Google AI Overview volatility – Overviews don’t always show up. Treat a missing overview as a zero‑mention data point, not a measurement error.

Treating AI share of voice as a structured, repeatable KPI lets you stop guessing about generative visibility and start optimising it with the same discipline you once applied to organic search.