Methodology · Last updated October 27, 2026
How we measure AI visibility
Most AI visibility tools show a score without explaining it. We publish exactly how ours works: which models we ask, which questions we ask them, how each answer is read, the formula behind every number, and what the data can't tell you.
1. Which AI assistants we ask
Every analysis asks all four assistants through their official APIs. All four are included in every plan; none is a paid add-on.
| Assistant | Model we query | Live web search |
|---|---|---|
| ChatGPT | OpenAI gpt-4o | No (answers from the model) |
| Claude | Anthropic Claude Sonnet 5.5 | No (answers from the model) |
| Gemini | Google Gemini 3.8 Flash | No (answers from the model) |
| Perplexity | Perplexity Sonar | Yes (Perplexity always searches the web) |
We don't track Google AI Overviews, Google AI Mode, Microsoft Copilot, Grok, DeepSeek or Meta AI.
2. Which questions we ask
We read your website, then generate 22–25 questions a real buyer in your category would ask, such as “What are the best tools for X?” or “How does your brand compare to a competitor?”. You see every question before the analysis runs and can edit, remove or add your own.
Weekly checks reuse the same questions (up to 15) and the same competitors every week, so a change in your score reflects a change in the AI answers, not a change in what we asked.
3. How each answer is read
Each assistant answers each question once. We then read every answer in two ways: text matching that recognizes your brand and competitors (including common spelling variants), and a small AI model (OpenAI gpt-4o-mini) that extracts which brands are mentioned, the order they're ranked in, and whether each mention is positive, neutral or negative. You can open every original answer in the report and check it yourself.
4. The formulas behind every score
Visibility score
Answers that mention your brand ÷ all answers × 100
Out of every answer we collected, how many named you.
Share of voice
Your mentions ÷ mentions of all tracked brands × 100
Your slice of all brand mentions, compared with your competitors.
Sentiment score
Average of each mention: positive = 100, neutral = 50, negative = 0
How favorably the assistants describe you when they do mention you.
Average position
Mean rank where the answer lists brands in order
When an assistant ranks options, where you usually land.
AI Brand Strength
35% visibility + 25% sentiment + 20% share of voice + 20% ranking score
One summary number. Ranking score: position 1 = 100, 2–3 = 80, 4–5 = 60, 6–10 = 40, lower = 20, not ranked = 50.
5. Weekly checks
When you turn on weekly checks, the same analysis runs automatically once a week and you get an email with your visibility score and how it moved. AI answers vary from run to run, so a move of less than 3 points is reported as “about the same” rather than as a real change. A failed run costs nothing and is retried the next day.
6. What the data can't tell you
- API answers are close to, but not identical with, the chat apps. The apps can add web search, memory and personalization that an API call doesn't have.
- One answer per question per assistant. We don't sample each question several times, so a single analysis is a snapshot. Trends over several weeks are more reliable than any one score.
- No location or account simulation. Answers aren't localized to a city or tailored to a user profile.
- ChatGPT, Claude and Gemini answer from what the model knows. Only Perplexity searches the web live in our checks, so very recent changes to your site may show up there first.
7. What it costs
One full analysis uses 11 credits: 1 to read your website and 10 to ask all four assistants. Reading competitor websites uses 0 credits. A weekly check uses 10 credits and is only charged when it succeeds. Every price and plan limit is public on our pricing page, and we compare ourselves with other tools in AI visibility tool pricing.
Questions about our method? Email support@aibrandtrack.com.