Methodology1.The Questions
Everything starts with the questions your customers really ask an AI before choosing a supplier. For the free audit, we read your website to work out your activity, your area and the language your buyers use, then generate 12 questions: 10 market questions that do not name any company (“which supplier…”, “best…”, “how much does…”) and 2 questions that name your business (reviews, reliability, price, comparison).
Subscribers track a panel of questions: questions common to their market, which we measure for every client in that market, plus questions of their own that they or their expert add: 25 to 50 questions depending on the plan, measured every week. The 12 questions above are the limit of the free audit only.
122.The Engines
We query ChatGPT (OpenAI), Perplexity and Gemini (Google) through their official interfaces, with web search turned on, so that the answers reflect what the assistant finds on the web today and not only what it learned during training.
3.The Free Audit
Each of the 12 questions is asked once to each engine. For every answer, we record whether your business is :
- Cited: one of your pages is given as a source of the answer;
- Mentioned: your name appears in the text of the answer, without a link.
What Counts as Visibility?
Cited
Source: acmecnc.com
Mentioned
for
Brand MentionedVisibility Score
Your visibility score is the number of questions on which at least one engine cites or mentions you, out of the questions tested
(for example 4/12). We also list the businesses named instead of you and how often each one appears.
One answer per question and per engine is a snapshot, not a statistical measurement: run the same question tomorrow and the answer may differ. That is why subscribers are measured repeatedly.
4.Weekly Tracking for Subscribers
Subscriber questions are measured in waves. In each wave, every question is asked several times to each engine included in the plan, because a single answer can be a fluke:
| Engine | How Often | Answers per Question |
|---|---|---|
 Perplexity | Every Week | 6 |
 ChatGPT (OpenAI) | Every Two Weeks | 4 |
 Gemini (Google) | Every Two Weeks | 4 |
These settings can change as engines evolve; when we replace a model or change the panel, the change is marked on your charts so that a jump caused by the switch is not mistaken for progress.
5.How We Read the Numbers
Rolling Windows
Scores are read over the last three waves, never on a single wave, to smooth out the natural variation of AI answers.
No Alert TriggeredAlerts Only When It Is Real
A drop or gain alert is sent only when the difference between the current window and the previous one passes a statistical test (a comparison of two proportions). A single bad wave does not trigger an alert.
Competitors on the Same Scale
The businesses named in the answers are scored exactly like you, so your share of voice is comparable question by question.
Nothing is Overwritten
The raw answers are stored unchanged and the way we extract citations and mentions is versioned. If we improve the extraction, scores are recalculated from the original answers.
6.From Measurement to Recommendations
To build recommendations, we crawl the public pages of the website (up to 150 pages, depending on the plan) and look for the reasons why it is missing from each answer.
Site-Wide Causes
Site-wide causes are checked once per website
- AI Accessibility
- Can AI crawlers access and read your pages?
We check robots.txt, JavaScript rendering and your sitemap. - Business Clarity
- Is it clear who you are, what you do, who you serve and where you operate?
- Informational Content
- Does your website contain the information AI assistants need to
answer buyer questions? - Answer Structure
- Are answers easy to find and extract?
- Content Freshness
- Is your content up to date?
Question-Level Causes
Diagnosed for every question you lose.
We compare the pages AI engines cite with your own pages to identify the gap:
- Missing Page
- No page directly answers the question.
- Indirect Answer
- The information exists, but the answer is not clear or direct enough.
- Missing Proof
- Prices, lead times, certifications or case studies are missing.
- Hidden Answer
- The page structure makes key information difficult to extract.
- Weak Third-Party Authority
- Other sources provide stronger external validation.
- Outdated Content
- Your information is old or no longer competitive.
Who Wins Each Question
We also look at who wins each question today.
The realistic goal depends on who is cited instead of you:
No Source Cited
A clear, direct answer has a real chance to earn the citation.
Official Source Dominates
Aim to earn a secondary citation alongside standards bodies or public authorities.
Competitors Get Cited
Identify what the winning pages do better and how to close the gap.
Expert Review
Every recommendation starts as a draft backed by evidence. An expert reviews each one and validates, rewrites or rejects it
before it reaches you.
7.Before and After
When you mark an action as done, we record the date. The following waves show how the related question evolves, for example “0 of 6 answers before, 4 of 6 over the last three waves”. With a few answers per wave, these are measured evolutions, not proof that the action alone caused the change: other sites, and the engines themselves, change too. We say so plainly.
Action Marked as Done
Oct 12, 2026 · Tracking continues in subsequent waves
8.Limits We Are Open About
Answers Are Not Always Consistent
AI answers vary between runs, between engines and over time. Repetition and rolling windows reduce this noise; they do not remove it.
Results May Differ by User
We query the engines through their official interfaces, without a user’s history or location.
The answer a specific person sees in the app can differ.AI Engines Keep Changing
Engines change their models and their search sources without notice. We monitor them daily and annotate changes, but a sudden shift can affect every business at once.
The Free Audit Is a Snapshot
The free audit is a snapshot built on one answer per question and per engine.
Questions about the method?
Write to us, we are happy to go into the details.