Methodology

How Cyted measures AI search visibility

One number — does AI recommend you or your competitors — measured honestly. Here is exactly how the audit works, so you can check every claim we make.

  1. 01
    Three AI engines, every audit
    Every audit samples ChatGPT, Gemini and Claude — three flagship answer engines — with live web grounding. We don't hold an engine back for a higher tier: a £20 one-off audit samples exactly what a Pro subscription does, on the same flagship models.
  2. 02
    Real buyer questions, per vertical
    We ask the questions a real customer would ask — comparison, specific-business and near-me questions drawn from a hand-authored pack for the trade, with the city filled in for local queries.
  3. 03
    k=3 samples → a rate, not a screenshot
    AI answers vary run to run, so each question is asked three times per engine. The result is a mention rate — how often you're named — not a single lucky (or unlucky) screenshot.
  4. 04
    A 95% confidence interval
    Every rate carries a Wilson score confidence interval, so you can tell a real signal from sampling noise. We never claim a change that sits inside the interval.
  5. 05
    An LLM judge extracts who was named
    A separate model reads each answer and extracts the businesses it recommends, matching your business against its name variants and citations so a mention under a slightly different name still counts as you.
  6. 06
    Fixes from what the engines cited
    Recommendations are derived from the domains and competitors the engines actually cited in answers that recommended someone else — concrete, evidence-backed fixes, not a generic checklist.