How to measure your brand's AI Share of Voice

AI Share of Voice is the percentage of AI answers to buyer questions that name your brand. To measure it, build a fixed basket of 20–50 real buyer prompts, run each through every AI engine, then score three things: how often you appear (share of voice), how prominently (position-adjusted prominence), and how often you're linked (citation rate). Re-measure on the same basket to track the delta.

Traditional rank tracking is going blind. In the first four months of 2026, 68% of Google searches ended without a click, up from 60.5% in 2024 (SparkToro / Similarweb, 2026). AI Overviews now appear on more than 20% of searches and cut click-through rates by nearly 60% when present (Search Engine Land, 2026). If your only metric is blue-link position, you're measuring a shrinking surface. AI Share of Voice measures the surface that's growing: the synthesized answer itself.

This is the exact loop we run inside Ravenopus with our internal geo_audit.py tool. The tool automates it across engines, but the method is what matters, and you can run it by hand.

What is AI Share of Voice, and why is it different from citation tracking?

Citation tracking asks a binary question: did an engine link to my page? Share of Voice asks a richer one: of all the buyer questions AI answers, in what share is my brand named, described, and recommended, versus my competitors? A brand can be cited in a sources list yet never appear in the paragraph the user reads, and another brand can be recommended in the prose without any link at all. You need both signals. This method scores three complementary metrics:

• Share of Voice — the % of prompts where your brand is named in the generated answer.
• Position-adjusted prominence — a weighted score that rewards earlier, longer, more-detailed mentions over a buried name-drop.
• Citation rate — the % of answers that link to your domain (the AEO overlap).

Step 1: Build a prompt basket of 20–50 real buyer questions

Your basket is the measurement instrument. Pull the actual questions that decide deals — from sales-call transcripts, JTBD research, support tickets — not keywords. Cover four intent types: category-defining ("What is [category]?"), comparison ("[you] vs [competitor]"), recommendation ("best [category] for [ICP] to do [job]"), and objection/trust ("Is [brand] legit?"). Freeze the list — the delta only means something if you ask the same questions every cycle.

Step 2: Choose your engines and query each prompt

Run every prompt through each engine your buyers use: Perplexity, ChatGPT (with AND without web search — the answers differ), Google AI Overviews and AI Mode, Gemini, Copilot, Claude. Fresh session, no leading context, one prompt one answer.

Step 3: Capture the raw answer as dated evidence

Engines drift week to week, so the raw answer is the proof. Save the full answer text with date + engine labeled; screenshot the consumer apps for surfaces an API can't reach. This dated evidence set is what lets you defend a "we moved from X% to Y%" claim later.

Step 4: Score each answer on three metrics

For every captured answer record: Named? (yes/no), Prominence (first mention / dedicated paragraph / buried), Framing (hero / one-of-several / cautionary / absent), Cited? (yes/no). Do identical scoring for your top 2–3 competitors on the same answers — 30% Share of Voice means nothing until you know the leader sits at 70%.

Step 5: Calculate your three headline numbers

Share of Voice = prompts where you're named ÷ total prompts. Position-adjusted prominence = average weighted prominence. Citation rate = answers linking your domain ÷ total answers. Report each next to competitors, broken out per engine — you'll usually find you're strong on one engine and invisible on another. That gap is the roadmap.

Step 6: Do gap analysis on the sources

For prompts where a competitor gets synthesized in and you don't, open the cited URLs. Those sources are the corpus the engine trusts for that question — Reddit, review platforms, a competitor's guide, an industry journal. Now you know what to get into or out-author.

Step 7: Re-measure on the same basket

Ship fixes, wait for re-crawl (weeks, not days), re-run the identical basket. The deliverable is never "we added FAQs" — it's "Share of Voice moved from X% to Y%, prominence rose N points, against a competitor baseline of Z%."

How often should you re-measure?

Monthly for an active program; quarterly minimum to catch drift. Keep the basket frozen; only add prompts, never swap them.

Ravenopus runs this exact loop for clients as part of full-cycle growth on one flat monthly retainer, using our internal geo_audit.py engine across Perplexity, OpenAI, Gemini, and Anthropic. Want your baseline scored against your top competitors? Start with a diagnostic.

Keep reading

• Next: How to audit your AEO and Copilot visibility.
• See how measurement fits the whole loop at our answer engine optimization page.