How we measure your visibility in AI
No black box, no magic number. This page explains, in plain language, how we calculate every figure in your report — and what we still can't guarantee.
01 · How we measure
Real questions, against the real AIs
We run questions — the ones your customers would actually ask — against ChatGPT, Gemini, Perplexity and Claude, and read exactly what they answer.
Questions with real intent
The questions look like what someone searching for a business like yours would type: recommendations, comparisons, «best [your sector] in [your city]». No jargon nobody actually asks.
Real consumer apps, not just the API
For ChatGPT, Gemini and Perplexity we use Bright Data to sign in to the app the way your customer would, with a session from your business's country. The official API and the app people actually use don't always answer the same, so measuring through the API alone gives an incomplete picture. If Bright Data fails, we fall back to the official API.
Claude, via API
We query Claude through Anthropic's official API.
02 · The sample, by plan
How many questions and how many AIs, by plan
The more questions and AIs you cover, the sturdier the sample. Here's how it breaks down by plan:
| Plan | Questions | AIs |
|---|---|---|
| Free | 5 questions | 1 AI — ChatGPT |
| Starter | 40 questions | 3 AIs |
| Growth | 75 questions | 4 AIs |
Today we run each question once. With a single run, the result for a specific question can change from one day to the next without your business changing at all — that's normal, it's how these models behave.
Starter and Growth repeat your same questions every week; Free runs a one-off audit when you sign up. That's why the real value is in how your figure moves over several weeks, not in any single day's number.
03 · What each metric means
What each metric in your report means
Mention vs. citation — not the same thing
A mention means the AI names your business in its answer. A citation means it also links to your website as a source. Both count, but citations are worth more: they mean the AI used your content, not just your name.
Coverage
The share of your questions where you show up, at least once, in any of the AIs we audit.
Share of voice
Out of every time an AI mentions a brand in your sector across your questions, what percentage is you versus your competitors.
Competitors in the quick report
The competitor scores you see in the free report are an estimate, not a full audit like yours. That's why you'll always see them marked with the ≈ symbol.
04 · Limits we own up to
What we still can't guarantee
No tool that measures AI visibility is perfect — anyone who tells you otherwise is selling you smoke. Here's what we can't guarantee today:
AIs answer differently from one run to the next
Ask ChatGPT the same thing twice in a row and it can answer differently without anyone touching anything — that's just how these models work. If your figure moves up or down slightly from one week to the next, it's likely normal model noise, not a real change in your business. Watch the trend across several weeks, not the single jump.
We measure the answer without personalization
We audit with a neutral session, with no user history or memory. The answer one of your actual customers gets — who does have their own history — can differ from what we see.
We don't yet measure Google's AI-generated search summaries
Today we audit the Gemini app, but not the AI summaries that appear inside Google Search itself (AI Overviews). It's a limitation we've identified and have on our radar.
What we can't verify, we say so
If we don't have enough data to verify something — for example, whether you have reviews on Google —, we mark it as «unverified». We never dock your score for something we couldn't verify.
05 · What we do to not lie to you
How we avoid making data up
Deterministic technical checks
Everything that can be checked objectively — your sitemap, your robots.txt, your structured data, whether Google has you indexed — we check with fixed, reproducible rules. The same site always gives the same result.
We verify before we claim
Before showing you a figure or a recommendation, we cross-check it against real data. If we can't verify it, we don't present it to you as fact.
A filter that strikes out unproven claims
We review every claim in your report and remove the ones not backed by real data. If something can't be proven, it doesn't make it into your report.
Got questions about your report?
Write to us and we'll walk you through it with your own data in front of us.