📊
Chapter 2
Layer 1: Visibility

AI Share of Voice

Share of Voice is the most widely used metric in AI visibility. It is also the most widely misreported. This chapter covers what it actually measures, why most tools get it structurally wrong, and how to make sure your number means something.

←
Previous chapter
Ch. 1 – AI Competitive Map

Next chapter
Ch. 3 – AI Topical Presence
→

⚡
TL;DR

The short version

Key points

Frequency beats ranking. Where you appear in a single AI response is unstable and changes every time. How often you appear across many responses is the number worth tracking.
The denominator must be open. If your tool asks you to pick your competitors upfront, your SOV is measured inside a pool you built, not the one the AI actually produces.
Presence is not the same as share. Appearing in 30% of prompts tells you about visibility. It does not tell you about share. Those are different things.
A low SOV score is a starting point, not a conclusion. Chapter 3 (AI Topical Presence) is where you find out why.

📌
Before you start

Two chapters worth reading first

Share of Voice measures how often your brand appears in commercial AI responses relative to everyone else. To get reliable data you need to understand which type of AI response you are measuring and how to design prompts that give you consistent results.

🔀

Ch. 7 – AI Visibility Channels

AI Share of Voice measures commercial grounded visibility, not training data recall. Understanding the difference helps you run the right kind of prompts and interpret what your score actually reflects.

📋

Ch. 8 – Prompt Tracking

SOV data is only as good as the prompts behind it. Chapter 8 covers how to design category-level prompts that produce unbiased, repeatable results.

You can start measuring without reading these first. But if your scores feel inconsistent or do not match what you expect, prompt design and channel confusion are the two most common causes.

📐
The definition

What AI Share of Voice actually is

A ratio, not a rate

AI Share of Voice is a ratio. It measures how much of the total brand mention landscape your brand accounts for across AI responses. The key word is total. Every brand the AI mentions goes into the denominator, not just the ones you chose to track.

The correct formula

AI SOV = Your brand mentions / All brand mentions across all responses

Computed across many runs. Every brand the AI names in any response contributes to the denominator, whether you expected it to be there or not.

This is straightforward in theory. The problem is that most tools do not implement it this way, and the difference between doing it right and doing it wrong produces numbers that are not just inaccurate but actively misleading.

⚠️
Common mistakes

Three ways most tools get this wrong

Problem 1: Presence rate reported as share of voice

The most common mistake. A presence rate tells you how often your brand appeared across the prompts you tracked. It does not tell you how much of the total conversation you owned.

What most tools actually calculate

Brand appearances / Number of prompts tracked

If your brand appears in 30 of 100 prompts, this reports 30%. But if four other brands appeared in every one of those same prompts, your real share is closer to 6%.

Problem 2: Position weighting is measuring noise

Some tools weight brands higher if they appear earlier in an AI response. This sounds logical but does not hold up. AI outputs are probabilistic. The same prompt produces different brand sets and different orderings every time it runs. In a 2026 study by Fishkin and O’Donnell, the probability of two responses producing the same ordered brand list was less than 1 in 1,000 across nearly 3,000 runs. Weighting by position is weighting by randomness.

What to track instead

Frequency across many runs. How often your brand appears in the total pool of responses is stable and meaningful. Where it appears within any single response is not.

Problem 3: The closed denominator

If a tool asks you to pick your competitors before you start tracking, it is computing SOV inside a pool you defined, not the one the AI actually produces. This creates three problems.

It is gameable

Remove a strong competitor from your list and your share improves immediately. The number goes up. Nothing in the real world changed.

It is not comparable

Two companies in the same market can report completely different SOV scores simply because they picked different competitor lists. The number is not describing the same thing.

It misses emerging rivals

If the AI has started consistently recommending a brand you did not put on your list, it will not appear in your data. You will never know it is there.

What closed-denominator tools actually compute

SOV = Your mentions / (Your mentions + Competitors you chose)

The competitive universe is defined by the user, not by what the AI actually says.

✅
The right way

How to measure it correctly

Open denominator, many runs

The correct approach has two parts. First, let the AI define the competitive pool rather than defining it yourself. Every brand that appears in any response goes into the denominator. Second, run enough times that frequency data is stable, not just one or two runs per prompt.

Open denominator formula

AI SOV = Your brand mentions / All brand mentions across all responses

The competitive landscape is defined by the data, not by assumptions. Run each prompt at least five to ten times. Aggregate across models for a fuller picture.

How Waikay calculates AI Share of Voice

Waikay extracts every brand mentioned across your tracked prompts and computes SOV against that full pool. No predefined competitor list required. New brands that appear in AI responses are automatically added to the pool, so your data reflects what the AI actually says rather than what you expected it to say.

➡️
Next steps

What to do with your score

The score tells you where you are. Not why.

Once you have a reliable SOV number, the next question is always what is driving it. Here is where to look depending on what your score shows.

Score is lower than expected

The most likely cause is missing topical associations. The AI is not connecting your brand to enough of the topics that buyers ask about in your category.

Go to Chapter 3 – AI Topical Presence

Unfamiliar brands are outranking you

You need to understand who the AI has built strong associations with in your space and why. The Competitive Map is the starting point for that investigation.

Go back to Chapter 1 – AI Competitive Map

Score looks healthy but results are not following

You may be appearing often but for the wrong topics. High frequency across irrelevant queries inflates SOV without generating the right kind of visibility.