Methodology

How we measure AI visibility

The category's biggest trust gap is opaque sampling — numbers with no stated depth, run count, or cadence. Here is exactly how ApexGEO measures, end to end. Our first rule: if we didn't measure it, we don't show it.

The 7 engines we track

We monitor your brand across the answer engines people actually use. Every headline metric is computed only from engines we sample directly — we never extrapolate to engines we don't query.

  • ChatGPT
  • Claude
  • Gemini
  • Perplexity
  • Grok
  • DeepSeek
  • Google AI Overviews

How we sample

You define tracked prompts — the questions your customers ask an AI engine where you want to appear. On each run we send every tracked prompt to the conversational engines and record what came back: whether your brand was mentioned, whether it was cited, the sentiment, and which sources the engine drew on. One prompt sent to one engine on one run is one sample. Included monthly sampling scales with your plan:

PlanIncluded samples / monthRefresh cadence
Free60Weekly
Starter180Weekly by default · daily on request
Growth800Weekly by default · daily on request
Professional2,000Weekly by default · up to 4×/day on request
Agency3,000Weekly by default · up to 4×/day on request
Enterprise6,000Weekly by default · up to 4×/day on request

The cadence above is for the conversational engines (ChatGPT, Claude, Gemini, Perplexity, Grok, DeepSeek). Google AI Overviewsis sampled separately — weekly, via a search API — measuring per-prompt inclusion, mention and citation in Google's AI answer surface.

Confidence, not false precision

AI answers vary run to run, so a single observation isn't a fact — it's a sample. We report every headline rate with a 95% Wilson confidence intervaland the run count behind it, so you can see how solid the estimate is. A metric reads like:

35% ± 8, n=120 runs / 30 days

— the point estimate, the ± margin (tighter as the run count grows), the number of runs, and the window they span. We use the Wilson interval specifically so small samples never produce nonsense like "100% ± 0".

What "Not measured" means

When a metric has no runs behind it — a brand-new brand, an engine we haven't sampled yet, or a segment with too little data — we show:

Not measured — 0 runs

…not a fabricated zero, a placeholder midpoint, or an average of other brands. An empty state is the honest answer, and it's the one we give. This is the same rule everywhere in the product: a value you see is a value we measured.