How the score is made

The Hide and Seek score is what a machine can check about your site, from the checks, 0 to 100 with a letter. It is the same number for a free scan and a watched site. Your North Star, whether the assistants name you, is counted beside it and is not a score: it is a results panel, question by question.

The score, from the checks up

Every one of the 40 checks gives a finding a score from 0 to 1: the fraction of what it asks for that your site does. A check that looks at several things (the home page and five sampled pages, say) scores the share that pass. The ✓, !, and ✗ marks are bands on that score, for reading: ✓ from 0.9, ! from 0.4, ✗ below.

A check that could not run (a timeout, a service that didn’t answer, a page that didn’t load) is skipped, with the reason, and left out of both the numerator and the denominator. A check with a weight of 0 is information: reported, never counted.

Each category is the weighted mean of its counted checks, 0 to 100. The score is the weighted mean of the categories:

Within a category, a check’s weight is relative to the others’: the page for each check says what it is out of. A category with nothing counted has no score and is left out of the total. The grade is a banding of the score: A from 90, B from 80, C from 70, D from 60, F below. Only a true 100 shows as 100.

Versions

Every report and every weekly count carries the version of the rubric it was made under; this is version 5. The number goes up by hand whenever a check is added or dropped, a weight changes, or an assistant’s model changes. Scores made under different versions are not compared: your account shows no change across the boundary, the weekly digest counts none, no drop alert fires, and a thin tick on the trend line marks where it happened. An old report keeps its number and its version; nothing is scored again.

You and others

Each check says who controls it. Most are yours to do, and the report lists those under “Fix first”, worth the most first. A few have to be earned from others (In Common Crawl, Mentioned on Wikipedia, and Talked about elsewhere): they count the same, because they matter to an assistant, but the report marks them with a hollow chip and never calls them a fix.

What we deliberately don’t score

• llms.txt. Reported, not scored: no assistant is known to read it, and a file that repeats what the pages already say is not a signal.
• Speed, beyond a floor. A home page that answers in under two seconds is fast enough for anything that crawls; what is faster than that is for people, and the score is not a performance audit.
• “Answerability”. Whether your prose is phrased the way an assistant likes to quote it is writing advice, not a check anyone can verify, and the outcome it promises is what your North Star measures directly.

What we don’t check

Whether Bing or Google has the site. ChatGPT search answers from Bing’s index; AI Overviews, AI Mode, and Gemini from Google’s. We wanted to ask each whether it has a site, since a site missing from either is invisible to the largest answer surfaces whatever else it does right. But Microsoft retired the Bing Web Search API in August 2025, and Bing’s results page ignores a site: search from anything it takes for a bot (DuckDuckGo, which reads Bing’s index, answers the same fetch with a captcha). Google closed whole-web search to new Programmable Search Engines in January 2026, and an engine restricted to a list of sites returns nothing for site: searches on other hosts. So neither index can be asked whether it has a site, not without pretending to be a person, and we won’t. Google will still answer the site’s owner directly: search for site:yoursite.com and see what comes back.