AI visibility score explained

How the score is calculated

The calculation is deliberately simple so it can be compared over time:

1. Pick a set of prompts that represent how buyers ask about your category.
2. Ask each prompt to each AI engine you track. Every prompt and engine pair produces one answer.
3. For each answer, detect whether your brand is mentioned.
4. Visibility score = answers that mention your brand divided by total answers, times 100.

The same calculation applied to one engine gives you a per-engine score. It is common to see large gaps, for example 60% on Perplexity and 10% on Gemini, because each engine retrieves different sources.

The metrics that sit next to the score

A single percentage hides useful detail. Read it together with these:

Average position

When an answer names several brands, the order matters. Readers skim, and the first recommendation gets the most attention. Average position is the mean rank of your brand among all brands named, counted only in answers where you appear. An average position of 1.5 is excellent; 4.0 means you are usually an afterthought.

Share of voice

Share of voice counts how many times each brand, yours and your competitors', is mentioned across all answers, as a share of all brand mentions. It tells you who the AI considers the default choice in your category. If one competitor holds 40% share of voice and you hold 8%, you know who you are really competing against in AI answers.

Cited sources

Engines with web retrieval list the pages they used. Counting citations by domain shows which websites shape AI answers for your category. This is the most actionable metric: it tells you where to get featured.

What is a good AI visibility score?

It depends on your prompt set and your market, so compare yourself to your competitors and to your own past scores rather than to a universal benchmark. As a rough guide for a balanced prompt set that includes unbranded category questions:

• Below 10%: AI engines rarely consider you. Focus on getting listed on the sources they cite.
• 10% to 30%: you are known but not a default recommendation.
• 30% to 60%: you are a regular recommendation in your category.
• Above 60%: you are a category leader in AI answers.

Branded vs unbranded prompts

Prompts that contain your brand name, such as "{brand} alternatives" or "Is {brand} worth it?", almost always produce an answer that mentions you. They are useful for checking how AI describes you, but they inflate the score. Unbranded prompts such as "best {category} for startups" are the true test of discovery. When comparing scores, keep the mix of branded and unbranded prompts constant.

Why your score moves between runs

AI answers are not fixed. The same prompt can produce different brand lists on different runs because models sample their output, search results change, and engines update their models. Some practical consequences:

• A change of a few points between two runs is usually noise.
• More prompts give a more stable score. 50 prompts across three engines (150 answers) is far steadier than 10.
• Track a moving trend over several weekly or daily runs before drawing conclusions.
• Keep your prompt set fixed. Adding or removing prompts changes the baseline.

How to improve your score

Start with the cited sources list: get your brand accurately represented on the domains that appear most often. Then make your own pages easy to quote with plain definitions, real pricing and fair comparison pages, allow AI crawlers in robots.txt, and re-measure. Our guide on getting recommended by ChatGPT walks through each step.

Summary

• Visibility score is the percentage of AI answers that mention your brand.
• Read it with average position, share of voice and cited sources.
• Separate branded from unbranded prompts, and judge trends, not single runs.