Methodology
How we measure AI visibility
AI answers are non-deterministic, so a credible measurement has to be built for that, not around it. Here is exactly what we do, and what we deliberately don't claim.
1. We ask the real buyer questions
We don't ask “tell me about [your business].” We ask the questions a customer asks when they're choosing, “best plumber in [city]”, “emergency dentist in [city]”, “where should we stay in [town]”. Each business and city gets a set of these buyer-intent prompts, auto-discovered for your category and editable by you. Every prompt is a separate battle with its own shortlist, so we track them separately.
2. Across four assistants, unprimed
We put each question to four assistants through their APIs with live web access, the way a customer would ask, with no hint that you exist. We also read Google's AI Overviews and AI Mode, the SERP-AI answers where many “near me” searches now resolve.
We report the assistant families rather than pin exact version strings, because providers update models continuously, that churn is the reason monitoring exists.
3. Rates, not single samples
Because the same prompt can return different names on different runs, one answer proves nothing. We report rates across every prompt and assistant:
- → Mention rate: the share of answers that name you at all.
- → Citation rate: the share that link to you as a source, not just say your name.
- → Sentiment: when you are named, whether the description is positive, neutral, or negative.
- → Share of voice: how often you're named versus the competitors AI recommends instead.
Your 0-100 score is a weighted blend of these, so a single lucky or unlucky answer can't swing it.
4. How we know it's really you
Matching a business name in free-form text is where naive tools get it wrong (counting “Ace” inside “grace”, or “Inn” inside “dinner”). We match on word boundaries, fold accents so “Café Río” and “Cafe Rio” are the same business, normalize punctuation so “Joe's” equals “Joes”, treat “&” and “and” as interchangeable, and tolerate the legal suffixes assistants routinely drop (“Ace Plumbing” for “Ace Plumbing LLC”). We'd rather miss a borderline mention than credit you with one that isn't yours.
5. We show you the receipts
Every scorecard includes at least one real, timestamped answer: the exact prompt we asked, an excerpt of what the assistant actually said, whether you were named, and any source it cited. You never have to take a number on faith, you can read the underlying answer yourself.
What we deliberately don't claim
- • We can't guarantee an assistant will recommend you. We don't control ChatGPT, Claude, Gemini, or Perplexity. Anyone who promises rankings there is guessing.
- • A single scan is a snapshot in time. The answers move, which is the point of monitoring, not a flaw in it.
- • We measure what the models say, not why. We surface the sources AI leans on so you can act, but the models don't publish their reasoning.
Our guarantee is scoped to what we actually control: showing you where you stand and giving you specific actions.
See your own measurement
Run a free scan and read the receipts yourself. About 30 seconds, no signup.