Get Optimised The GEO industry’s memory

Claim File First published 27 July 2026 Last reviewed 27 July 2026

Is share of voice a reliable metric for AI search visibility?

The answer

Share of voice is a popular but contested metric for AI search; the evidence shows it is widely promoted but increasingly questioned, with no consensus that it reliably measures brand visibility in LLM outputs.

Verdict: contested, not settled Confidence: moderate

The claim

Since mid-2025, a wave of articles has promoted share of voice as the essential KPI for brand visibility in AI search. The idea: measure how often your brand appears in LLM responses relative to competitors, then optimise to increase that share. Tools and agencies rushed to offer measurement and improvement services. But a counter-narrative emerged in early 2026, arguing that share of voice is the wrong metric and that alternatives like citation share or share of model better capture how AI engines actually select sources.

The receipts

Our corpus contains 78 articles from 38 publishers that mention share of voice in the headline. The earliest is from June 2025; the most recent is July 2026. Of those, 39 take a promotional stance, 10 express doubt, and 29 are neutral. The first doubt article appeared on 18 January 2026, titled "Share of Model: The New Metric That Replaces Share of Voice" from Aether Agency. After that date, 31 more promotional articles were published by 21 distinct publishers, showing that the hype continued despite the challenge.

31

Number of promotional articles about share of voice published after the first doubt article appeared in January 2026.

Despite the early challenge, promotional content continued at a high rate. 31 articles from 21 different publishers promoted share of voice after the doubt was raised. This suggests that commercial incentives to sell SOV tools and services outweighed the emerging skepticism.

Meanwhile, only 10 articles in total took a doubt stance. The ratio of promotion to doubt is roughly 4:1, but the doubt articles come from respected sources and propose specific alternative metrics, making them harder to dismiss.

What this means for you

Founders should treat share of voice as a directional signal, not a definitive KPI. The metric is easy to measure but hard to validate: no study in the corpus proves that a higher share of voice correlates with actual business outcomes like traffic or conversions. Meanwhile, alternatives like citation share (the proportion of AI responses that cite your brand) and share of model (how often your content is used as a source in model training) are gaining traction. The safe move is to track multiple metrics and watch for convergence, not to bet your strategy on a single contested number.

  1. Track citation share instead of share of voice. Citation share measures the proportion of AI responses that actually name your brand or link to your site. It is more directly tied to the LLM's output than a vague 'voice' metric. Use tools like Omnia or Presenc to get this data.
  2. Run your own controlled experiments. Do not rely on vendor dashboards alone. Pick 10 high-value queries, measure your citation rate weekly, and correlate changes with your content updates. If your citation share moves but business metrics do not, the metric is noise.
  3. Watch for convergence across metrics. If share of voice, citation share, and share of model all point in the same direction, you have a signal. If they diverge, the metric you are optimising for is probably the wrong one. Prioritise the one that best predicts actual referral traffic from AI sources.

Confidence, and what would change our mind

Confidence is moderate. The volume of promotional content suggests strong commercial interest, but the emergence of doubt articles from credible sources (AuthorityTech, Machine Relations, Topify) indicates genuine debate. The lack of a controlled study linking share of voice to revenue or ranking changes keeps the question open. If a major LLM provider (OpenAI, Google, Perplexity) officially endorsed share of voice as a metric, or if a peer-reviewed study showed a clear causal link, the verdict would shift. Conversely, if the doubt articles continue to accumulate and the promotional pieces fail to provide evidence, confidence in the metric would drop further.

Sources