What is AI search visibility, and how do you measure it?

When a buyer asks an AI assistant for a shortlist, the answer decides which vendors they look at next. If you're not on it, you never get the visit.

• Shortlists form inside the answer. If a competitor is named and you're not, you can lose a deal you never knew existed.
• The description sticks. If AI describes an old product, the wrong category or outdated pricing, buyers take it at face value.
• Investors and candidates ask too. People researching your company may get their first summary from an assistant.
• Your analytics can't see it. Most AI answers don't produce a click, so referral traffic undercounts how often you're shown, or left out.

Each metric is calculated over a fixed set of prompts, counting only the answers that actually came back.

Mention rate

The share of answers that name your brand. If 20 prompts produced 18 answers and 6 named you, your mention rate is 33%. This is the headline number.

Rank in answers

When an answer is a list, your position in it. Being named fifth out of seven is very different from being the first pick. Track your average position across the answers where you appear.

Citation rate

The share of answers that cite or link to your own domain. A mention without a citation means the AI knows you but isn't using your site as evidence. A citation without a mention usually means your content was useful, but someone else was the recommendation.

Share of voice

Your mentions as a share of all brand mentions in the answers, yours plus your competitors'. If competitors are named 30 times and you're named 10 times, your share of voice is 25%. It shows whether you're gaining ground on competitors, which raw mention counts can hide.

A fifth signal sits beside these: brand perception, meaning what AI says when someone asks about you by name. Does it know you? Is the description accurate? It isn't part of the score, but it's often the quickest thing to fix.

Ask ChatGPT the same question twice and you can get two different lists. That's normal, and it's why single checks mislead. The main causes:

• Generation is probabilistic. Models sample their wording, so order and inclusion shift between runs.
• Live search results change. Assistants that search the web read different pages from one day to the next.
• Models get updated. A new version can change answers across a whole category overnight.
• Personalization. Logged-in apps can use memory, past chats and location, so your own checks reflect your history.
• Wording. "Best CRM for startups" and "which CRM should a five-person startup use?" can produce different shortlists.
• API versus app. A model's API can answer differently from the consumer app, even for the same question.
• Google's AI answers are selective. Many searches don't show an AI Overview at all.

What helps is method: a fixed prompt set, consistent settings, enough prompts and engines to smooth out the noise, and trends over time instead of reactions to one answer.

You can do a rough version by hand:

1. Write 10 to 20 prompts your buyers would ask: category, use case, alternatives to a named competitor, and one about your brand.
2. Run each in a fresh, logged-out or temporary session on every engine you care about.
3. For each answer, record whether you're mentioned, your position, whether your domain is cited, and which competitors appear.
4. Calculate mention rate, average position, citation rate and share of voice.
5. Repeat with the identical prompts next month.

It works, but it's slow across six engines, and logged-in checks drift with your own history. Our free checker runs 3 buyer-intent prompts and 1 brand prompt across 6 AI surfaces in about 1–2 minutes, with consistent settings, and keeps the full answers and sources so you can see why you scored what you did.

Our score turns those metrics into one number from 0 to 100. It's built only from the buyer-intent prompts, the ones that don't mention your brand, because those show whether AI recommends you to people who don't know you yet. Each answer that came back is scored like this:

Each answer is capped at 100. Your AI Visibility Score is the rounded average across every buyer-intent check that came back, where a check is one prompt on one engine. Checks that produced no AI answer (for example, a search where Google showed no AI Overview) or failed to run are left out rather than counted as zero.

A worked example. Brookhaze, a fictional customer-feedback tool, gets 15 answers back from 18 buyer-intent checks, because 3 searches showed no AI Overview. It's named in 6 answers, ranked #1, #2 (with its domain cited), #2, #4, #5 and #7. One more answer cites its blog without naming it. Those answers score 80, 90, 70, 60, 60, 50 and 20, which adds up to 430. Spread across 15 answers, that's a score of 29.

The score falls into one of four bands:

• Leading (70–100): named often and near the top.
• Visible (45–69): in the conversation, but not consistently.
• Emerging (20–44): occasional mentions, usually low in lists.
• Invisible (0–19): rarely or never named.

Brookhaze is emerging, with a 40% mention rate, a 13% citation rate and an average position of 3.5. Share of voice is calculated over the same answers, counting each competitor once per answer. The brand prompt ("What is Brookhaze?") is reported separately as perception and doesn't change the score. Every step, including how each engine is queried, is on How it works, and you can see a full example in the sample report.

• Now: take a baseline before you change anything. Without one, you can't tell what worked.
• Monthly: the right default for most startups. It's frequent enough to catch shifts, and slow enough for changes to be crawled and picked up.
• Weekly: during active pushes, such as a launch, a PR campaign or a batch of new comparison pages, and after a major model update.
• Not daily. Day-to-day variation looks like signal when it isn't.

Keep the prompts and settings identical between checks, change a few things at a time, and judge progress on the trend across several checks. We're building a weekly AI Visibility Tracker for exactly this. It's in early access.