The denominator behind every number we hand you

As of 2026-08-13

How many questions one audit puts into the market

An audit builds a question list, then puts the same questions to several places several times. Engines answer the same question differently from day to day, so asking once only tells you what that day looked like.

Panorama puts 50 questions 7 times each, and reads every one of the 17 places listed below.

One question becomes 15 places × 7 repeats = 105 cells. Each square below is one cell.

Every place we ask, named

Buyers in different countries use different places, so the rosters differ. Below are all 15 places we ask in this market.

Every screen a name can appear on

We ask in 15 places, but answers surface on 17 screens. One question can split into two screens, so the two counts differ. Here are all 17.

We also count what arrives alongside

One search returns more than a list of links. Videos, maps, shopping, people-also-ask — they arrive on the same screen. In this market those add up to 57 spots. We read them from the answer we already paid for, without asking again.

The same question, asked 7 times

The same question does not get the same answer twice. One ask gives you the value of that one ask. 7 of them show how far the answer moves.

So every rate ships with a confidence interval: the band the rate could actually sit in. We use the 95% interval.

Why 7, and what changes when you ask far more times than that, is written out on the methodology page. Methodology: why it is measured this way

What counts as a completed round

These three decide whether a round can close. Miss any one of them and the round does not ship.

A citation is the AI linking to you as a source. A mention is your name in the text with no link. That split, and the rest of the words a judgement turns on, is fixed one sentence at a time in the glossary. Glossary: the words a judgement turns on

The same ruler goes over our own pages

When we find something that makes pages easier for search and AI answers to pick up, it goes on our own site first and gets one line in a ledger. That ledger holds 190 entries today: 57 apply to every single page, 111 to the site as a whole, and the remaining 22 to particular page types.

Every deployment re-checks that all 190 are still in place. Improvements live on the page, not in the codebase, so building a new page quietly rolls them back.

The three market sites carry their structure in a separate registry that gets compared page by page: 114 in Korea, 105 in the United States, 110 in Japan, 329 in all.

Running exactly this over your pages is the first step of the improvement work.

Read next: Methodology: why it is measured this way · Audit tiers and pricing · Glossary: the words a judgement turns on · Public claims registry

Who writes this spec and signs off on each measurement is on the author page.