How we measure your AI visibility

What we measure

We ask the AI assistants your customers use the same questions real people in your area ask, for example "Who is a good divorce lawyer in your city?" or "Emergency plumber in your city, who can come now?" For each answer we record:

• whether your business is mentioned
• where it appears among the businesses named
• what the AI says about you, and whether it's accurate
• which websites the AI relied on

Platforms: ChatGPT, Google AI Overviews and AI Mode, Perplexity and Gemini. We also run periodic manual checks of Microsoft Copilot.

Why one screenshot proves nothing

AI answers change constantly. In a 2026 study, SparkToro found that ChatGPT and Google's AI returned the same list of brands less than 1% of the time when the same question was asked repeatedly. (SparkToro research)

So we don't report a single "ranking." We ask 50 questions, on 4 platforms, 3 times a month on different days, about 600 answers a month, and report how often you're mentioned, with a margin of error.

Our metrics

• Mention rate: the share of answers that mention you. This is the headline number.
• Top-3 rate: the share of answers listing businesses where you're among the first three named.
• Share of voice: your mentions compared with the competitors we track.
• Accuracy: whether facts about you (phone, address, services, languages) are correct.
• Sources: which websites AI assistants cite in your market, and which of them you can influence.

Questions that already contain your business name ("Is [your firm] any good?") are excluded from the mention rate. We use them only to check accuracy and tone.

Fair, repeatable conditions

• Every question is asked in a fresh session with no login, no chat history and no personalization.
• Local questions name your city, and location settings point to your area where the platform allows it.
• The question list is fixed for the whole engagement, so this month compares fairly with last month.
• Each answer is labeled with how it was collected: official API, interface data, or a manual check. Research shows these can differ substantially, so we never present one as the other.

Telling our work apart from noise

1. Baseline first. We measure twice before changing anything.
2. Control questions. We keep 15 questions we don't specifically optimize for.
3. Competitor tracking. We track up to 5 competitors to see how much the AI is shifting for everyone.

We only say our work likely made a difference when your improvement beats both the control questions and your competitors' changes by more than the margin of error. If it doesn't, your report says: "Not yet distinguishable from normal variation."

You can verify everything

• Every number in your report links to the original AI answer, with timestamp, platform and settings.
• You get the full monthly data file. Records are locked once collected, and each month's file includes a fingerprint (hash) so you can tell nothing was changed later.
• We give you 5 questions to ask the AIs yourself in a private browser window. Your results may differ from a single run of ours; that variability is exactly why we measure at scale.

What we don't promise

No one controls what AI assistants say. We don't guarantee rankings, mentions, calls or revenue. We promise the work, and honest, verifiable measurement of whether it's moving the needle.