Method g1.0

How we measure. All of it.

AI answers change every time you ask. So we never report one answer as the truth. We ask the same questions many times, count, and show the margin of error with every number.

What we ask

For a paying business we build 20 questions from real demand: your Search Console queries, your reviews and the questions people ask in your market. A free grade uses 5 questions from our public library for your trade.

We ask every question on seven engines: ChatGPT, Gemini, Perplexity, Copilot, Claude, Google AI Overviews and Google AI Mode. Each is asked three times a week, and four weeks are pooled into one measurement: 12 answers per question per engine, 1,680 answers in all.

A free grade is smaller (5 questions, 7 engines, 3 times: 105 answers), so it has a wider margin and says so.

What counts as an answer

Refused and failed answers don't count. When Google shows no AI Overview, that run is left out and reported as how often an overview appeared. An answer from any other engine that names nobody still counts: it is an answer that didn't name you.

The grade

One score out of 100 from eight parts. The weights are fixed for this method version.

budgiehq.com/method#grade

How often AI names youAnswers that name you, divided by valid answers, on your questions. Each engine counts equally. Scored on the curve below.60

Your share against rivalsYour mentions divided by yours plus your three closest rivals', on the same answers. Full marks at 40% or more.10

What AI says about youFacts AI states about you (hours, prices, phone, services) that match what's true. If it states none, half marks (50%).8

Present where AI looksThe sites AI cites for your trade and town, weighted by how often they are cited, and whether you are on them.8

AI crawlers allowedWhether your robots.txt lets the major AI crawlers read your site.5

Business details on your siteMachine-readable markup that says what kind of business you are, where, and how to call you.3

Your site answers buyer questionsThe share of your customers’ questions that your own site answers.3

Name, address, phone agreeWhether the sites AI cites list the same name, address and phone number as you do.3

The mention-rate curve

It is steep at the bottom, because that's where most businesses start, and small gains there matter most.

Named in	0%	5%	15%	30%	50%	70%+
Points (of 60)	0	12	24	36	51	60

Straight lines between the points: named in 10% scores 18 points.

Letters

Grade	Score	In plain words
A	80 to 100	Named in about half of answers or more, groundwork done
B	65 to 79	Named in roughly a third
C	50 to 64	Named now and then
D	35 to 49	Rarely named, or groundwork done but AI is not naming you yet
F	0 to 34	Named in almost none

The margin of error

Answers to the same question are related, so we work out the margin by resampling whole questions, not single answers (a cluster bootstrap, 10,000 times). If the range crosses a letter boundary we say so: "B, could be C".

When we say something changed

We compare the same questions before and after, and call it a change only when the whole range sits on one side of zero. Anything else says "no clear change yet", with the number of answers behind it. We never describe noise as a result.

What we don't do

We don't promise a ranking. Nobody controls what an AI says.
We don't edit answers. Every answer is stored exactly as it came back, with a fingerprint so any change would show.
We don't rewrite old grades. A new method gets a new version number, and old grades keep theirs.