# A percentage is only as good as the prompts behind it.

> AI share of voice, calculated three ways: by mentions, by citations and weighted by position. What the figure needs to be valid, and what it cannot say.

- Page: https://mohammedteto.com/ai-share-of-voice/
- Language: en — French version: https://mohammedteto.com/fr/visibilite-ia/
- Author: Mohammed Teto (https://mohammedteto.com/#person)
- Published: 2026-10-07
- Updated: 2026-10-04

AI share of voice compares how often a brand appears in AI answers with how often its competitors do. This page sets out the three ways it is calculated, what each requires, and where the figure stops meaning anything.

AI share of voice is the proportion of brand mentions, in AI-generated answers to a defined set of category questions, that go to one brand as opposed to the others tracked. It is a relative measure: it needs a list of prompts, a list of competitors and repeated runs before it means anything. It says how often a brand is named. It does not say whether the brand is described accurately, or favorably.

<a id="definition"></a>

## The definition, and its three terms

LLM Pulse defines AI share of voice as "the percentage of AI-generated responses that mention, cite, or recommend your brand across a defined set of category prompts, relative to all brand mentions in those same responses" ([LLM Pulse, 2026](https://llmpulse.ai/blog/measure-ai-share-of-voice/)). Three things in that sentence have to be fixed before any number is produced.

- **The prompt set** Which questions are asked. The figure describes these questions and no others.
- **The tracked brands** Whose mentions are counted in the denominator. Add a competitor that was present in the answers all along and every other brand's share falls, without anything in the answers having changed.
- **What counts as a mention** A name in the text, a link among the sources, a recommendation. Each gives a different number.

<a id="methods"></a>

## Three ways to calculate it

The same source describes three calculations.

| Method | What is counted | What it shows | Its weak point |
|---|---|---|---|
| Mention-based | The brand's mentions, divided by all tracked brands' mentions, times 100 | How often the name appears in the text of answers | Counts a passing mention like a recommendation |
| Citation-based | The brand's domains among the sources cited | Whether the brand's own pages are used as sources | Depends on how much an engine cites: LLM Pulse reports the measure as much more meaningful on Perplexity than on ChatGPT |
| Position-weighted | Mentions weighted by rank in the answer: 1 for the first, 1/2 for the second, 1/3 for the third | Whether the brand is named first or last | Unstable on a single run; LLM Pulse calls single-run position weighting "statistically reckless" |

The three are not interchangeable. A report that gives one share-of-voice figure without saying which method produced it cannot be compared with another.

<a id="requirements"></a>

## What a valid reading requires

**Enough prompts.** LLM Pulse states that "a minimum of 50 prompts gives you a defensible measurement; 100 to 200 is the sweet spot", balanced across category, comparison and use-case questions in roughly equal thirds.

**One engine at a time.** The engines it lists are ChatGPT, Perplexity, Gemini, Google AI Mode and Google AI Overviews. A figure that blends them hides the differences between them.

**Repetition.** Answers vary from one run to the next, so each prompt is run more than once. For tracking over time, the source recommends a weekly cadence for most brands. These are one provider's recommendations, not a standard.

**A written protocol.** The prompts, the tracked brands, the counting rule, the engine and the date are recorded with each figure. How a run is recorded is set out in [what is recorded](https://mohammedteto.com/ai-search-intelligence/#record).

**A category that exists.** Share of voice supposes competitors and questions that buyers ask about a category. For a person, or for an organization without a defined category, the measure has little to measure; a description is assessed for accuracy instead.

<a id="limits"></a>

## What the figure cannot say

The general limits of these measures are set out in [what the numbers cannot say](https://mohammedteto.com/ai-search-intelligence/#limits). Two are specific to share of voice.

It counts a mention, whatever the mention says. Being named as the option to avoid, or with a former positioning, counts like a recommendation.

It is not market share. The prompts model what buyers may ask; they are not a record of what was asked.

And it explains nothing. A change in share points to a question, which is answered by reading the answers and their sources: see [source and citation authority](https://mohammedteto.com/source-citation-authority/).

<a id="place"></a>

## Where it sits in the work

Share of voice is one of the metrics defined in [AI search intelligence](https://mohammedteto.com/ai-search-intelligence/#metrics), and it belongs to the last layer of [The Authority Architecture System](https://mohammedteto.com/method/), [Monitoring and Governance](https://mohammedteto.com/method/#monitoring-and-governance). It follows the earlier work; it does not replace it. An entity that systems cannot identify is fixed through [entity optimization](https://mohammedteto.com/entity-optimization/) and, for Google, through the steps described in [knowledge graph strategy](https://mohammedteto.com/knowledge-graph-strategy/). Whether its pages are extracted and used is the subject of [answer engine optimization](https://mohammedteto.com/answer-engine-optimization/) and [generative engine optimization](https://mohammedteto.com/generative-engine-optimization/); the three acronyms are set side by side in [SEO, AEO and GEO compared](https://mohammedteto.com/compare/seo-vs-aeo-vs-geo/). The wider subject is [AI visibility](https://mohammedteto.com/ai-visibility/), and related analyses are collected in [Insights](https://mohammedteto.com/insights/).

<a id="faq"></a>

## Frequently asked questions

### What is AI share of voice?

It is the proportion of brand mentions in AI-generated answers, across a defined set of category prompts, that go to one brand relative to the others tracked. It depends on the prompt set, the list of tracked brands and the counting rule.

### How is AI share of voice calculated?

In its simplest form: the brand's mentions divided by the mentions of all tracked brands, multiplied by 100. Two variants exist, one that counts cited domains and one that weights each mention by its position in the answer.

### How many prompts are needed?

LLM Pulse gives 50 prompts as a minimum for a defensible measurement and 100 to 200 as the range it recommends, spread across category, comparison and use-case questions.

### Is a high share of voice a good result?

Only if the mentions are accurate and favorable, which the figure does not show. It has to be read together with what the answers say.

### Can share of voice be measured for a person?

Rarely in a meaningful way. The measure supposes a category and competitors. For a person, what is assessed is whether the description is present, accurate and consistent.

---

Understand how search and AI currently interpret your authority. Request a confidential assessment of your visibility, entity signals, narrative consistency and source authority.

[Request a Private Assessment](https://mohammedteto.com/private-assessment/)