AI Share of Voice

AI share of voice measures how often your brand appears in AI-generated answers relative to competitors. It adapts the classic marketing metric (your share of all advertising, search results or conversation in a category) to assistants such as ChatGPT, Perplexity, Gemini and Google’s AI Overviews and AI Mode. Because these systems give one synthesised answer instead of a ranked list, presence in the answer is what you measure.

How it works

There is no single standard formula, but most approaches follow the same steps:

• Build a prompt set that reflects real questions in your category, across intents, personas and funnel stages.
• Run each prompt in each AI system you care about, several times and on a schedule, because answers vary between runs.
• Record whether each brand is mentioned, whether it is cited with a link (see AI Citation (LLM Citation)), its position in the answer and the sentiment.
• Divide your mentions or citations by the total for all tracked brands to get a share, overall and by topic, platform and market.

MLforSEO separates mention rate (named in the answer) from citation rate (linked as a source), since one can rise without the other.

How it’s used in SEO and marketing

• Reporting AI visibility alongside rankings, which cannot show selection in a synthesised answer.
• Finding topics where competitors are consistently recommended and you are not.
• Spotting entity problems: misattributed facts or confusion with another brand.
• Tracking the effect of digital PR, reviews and third-party mentions that AI systems draw on.
• Combining with classic share of search from Google Trends or SERP data for a fuller picture.

Example

A CRM vendor tracks 200 buying-intent prompts weekly in three assistants, five runs each. Across 3,000 answers, it is mentioned 450 times, while all tracked brands together are mentioned 3,600 times. Its AI share of voice is 12.5%. Broken down by topic, it is strong on “integrations” and nearly absent on “pricing”, which points to a content and data gap.

Limitations and things to watch

• Results depend entirely on the prompt set. A biased or narrow set gives a flattering or misleading number.
• Answers are personalised and non-deterministic, so single checks are anecdotes; look at trends.
• Tools measure different things under the same name. Ask whether a figure counts mentions, citations or weighted positions.
• A high share does not guarantee traffic: much AI influence arrives later as direct or branded visits.

Learn it on MLforSEO

• Your rank tracker is lying to you about AI search
• 10 Ways to Automate SERP Analysis with AI
• LLM Citation Tracking: 4 Step Implementation Guide
• Dark AI traffic: why 70% of your AI influence hides as direct
• AI Search Optimisation & Agentic SEO