How we count.

Every number on a report is a count of answers we saved, and every claim carries the quote, the page and the date it came from. This page is the whole method, including the parts where we say we don’t know.

What happens in a run

1
Your pages are read first
Before anything is asked, your site is read and three things are proposed: what you are, who you sell to, and who you compete with. You edit all three and confirm them. Nothing is read off the page and used as fact without you.
2
Questions come from your sub-category
Not the category. A sales pipeline CRM for small teams gets different questions from a CRM, and the buyer asks different things at each stage: discovering, comparing, deciding. You keep, edit or remove every question before it runs.
3
Every answer is saved whole
An answer is asked once per engine and stored as it came back, with its date. Nothing is summarised on the way in. Every count on your report is a count over those stored answers, and clicking it opens them.
4
Cited pages are opened and read
When an answer cites a page, that page is fetched and its text stored. A claim is only shown as backed when its words are on the page it was cited from. Pages that cannot be read are listed as unreadable, never guessed at.
5
Changes point at a line
Each problem carries what AI said, the passage on the cited page, and the nearest line on your own page. The change we suggest is a line, in a place, on a named page — or a question only you can answer.

What the words mean

The same six words are used on every screen, and each is a count over saved answers.

Named: Your product appears in the answer at all.
Offered: It appears as one of the options the buyer is given, not only as a passing mention.
Picked: The answer recommends it over the others, or makes it the subject of the recommendation.
Above you: A rival placed higher than you in the same answer. Counted only against the rivals you confirmed as the same league.
Recognised: The answer knows the product and describes it, in a question that named it.
Cited page: A page an engine pointed at in its answer. Ours or anyone’s.

What we will not say

Rules the report applies to itself, whether or not they make it look thinner.

“Never”, “none of”, “absent”

Not asserted from fewer than five answers. A small sample is reported as the count it is.

Trends and deltas

Not shown on a first run. Change is always against the run before it, never against a baseline we invented.

Scores out of 100

Not produced at all. A single number across questions at different stages hides the one you are losing.

Sentiment

Read per claim, from the words in the answer, and shown with the words. Never averaged into a mood.

What we do not measure

AI crawler visits

We have no access to your server logs.

Traffic from AI answers

We have no referral data.

Sign-ups or revenue that follow

We have no conversion link, and anyone claiming one is guessing.

Whether AI will change its answer

A re-run shows what changed. Neither we nor anyone else can promise the change.

If a number is not in the product, it is not claimed on this site either.

Your data

What is stored

Your domain, the pages we read from your public site, the questions you confirmed, the answers engines gave, and the public pages those answers cited.

What is never done

Nothing is published, posted or edited on your behalf. No page of yours is changed by us. Every change is yours to make.

Who sees it

Your report is yours. An agency account keeps each client’s report to that client.

See it on a real run

The sample report is a complete run on a public site, with every rule on this page applied to it.

Read the sample report →