Skip to content

Measurement methodology

How we measure, step by step.

A number means nothing until you know what it is made of. Here is exactly what we ask, where, how often and how we turn answers into AI visibility.

1 measurement

one question asked to one AI platform in one market

2× – 10×

a week per question and platform, by plan and how stable the result is

4:00 – 22:00

UTC (an hour later in Slovakia, two in summer), spread across the whole week

5+

below 5 measurements we show no number

Platforms and models

  • ChatGPTAPIgpt-5.6-luna, forced web search, market locationall plans
  • ChatGPT – appThe real appanswers from chatgpt.com as a user sees them (collected by DataForSEO)paid plans and agencies
  • GeminiAPIgemini-2.5-flash with forced Google Search groundingall plans
  • PerplexityAPIsonarall plans
  • Google AI OverviewsGoogle result (SERP)real Google result via DataForSEOall plans
  • Google AI ModeGoogle result (SERP)the real Google AI Mode result via DataForSEOall plans
  • ClaudeAPIclaude-haiku-4-5 with web search, at most 2× a week per questionagency accounts

Answers are read by the gpt-5.6-luna parser: whether the brand came up, in which position, in what tone and which competitors. Cited URLs come straight from the platform's answer. The full text of every answer stays in the archive.

Metrics

What the numbers mean

AI visibility
Share of answers in which the AI mentioned the brand. Always with “X of Y” and a 95% confidence interval (Wilson).
Question coverage
For how many tracked questions the brand came up at least once.
Consistency
Questions with at least 5 measurements fall into “almost always” (80% of answers or more), “sometimes” and “never”.
Citation share
How many links in the answers point to the brand's website versus tracked competitors.
Position and tone
Where the AI names the brand and whether it speaks of it positively, neutrally or negatively.

Measurements

Why not “daily”

Measuring every question on every AI daily would be expensive, and pointless when the result doesn't change. So we measure several times a week and shift measurements to where something is happening. Every number shows how many measurements are behind it.

Google AI Overviews shows an overview only for some searches. A search without an overview does not count towards AI visibility — we show the share of searches with an overview separately. In the free audit such a search counts as an answer without a mention. Check against live Google (August 2026, logged-out browser with a Slovak location): the overview appeared or did not appear the same way in 27 of 30 searches. It is a small sample, not general proof of agreement.

Reliability

How sure a percentage is

AI answers a little differently every time, so we ask every question repeatedly and show next to every percentage how many answers it comes from. Plus the 95% Wilson confidence interval: the range the true share lies in with 95% confidence. With few measurements it is wide; every further one narrows it.

Example: the brand came up in 6 of 20 answers, so 30%. The interval is 14.5% – 51.9%, so a 5-point gap to a competitor means nothing yet at that number of measurements.

With fewer than 5 measurements we don't show a percentage at all, only “—” and the measurement count. Changes are shown in percentage points (pp): when a brand goes from 30 % to 35 %, that is +5 points, not +17 %.

How often we measure

How often we ask

Measurements per question and platform per week, by plan.

PlanStandardWhen results swing
Monitor2×2×
Optimize5×5×
Dominate5×up to 10×
Agencies5×up to 10×

In the week you add a brand we measure up to 7× per platform so first results come right away. When the result stays the same we measure less often: on Optimize and higher 3×, after 8 identical results in a row 2× a week. Agencies split their credits among active clients, so a large portfolio may get fewer measurements than the table shows.

Reputation

AI reputation score

The Reputation screen combines five signals from the last 30 days into one number from 0 to 100. The weights are fixed and you see every signal on its own too. If fewer than 50% of tracked questions are measured, the score is marked as preliminary.

  • 35%Visibilitylower bound of the 95% interval across tracked questions
  • 20%Toneshare of positive minus negative mentions
  • 15%Positionthe brand's average position in answers
  • 15%Presence in AI Overviewsshare of searches where Google showed an overview
  • 15%Citation qualityshare of links to media, institutions and encyclopaedias

What we don't show

Limits of measurement

  • · ChatGPT via the API is not the same as the app — we measure both and show them separately.
  • · We don't estimate traffic or revenue from AI. We take those only from your Google Analytics 4.
  • · We don't estimate market share or question volume in ChatGPT.
  • · We don't promise growth. We measure what AI answers and suggest what to try.