How we score

The checker runs 18 checks across four pillars, totalling 100 points. Fourteen are read from your pasted page source. Five are questions you answer, because a browser cannot fetch files from another domain.

Pillar 1 — Machine-readable (29 points)

Pillar 2 — Answer-shaped (27 points)

Pillar 3 — Reachable (25 points)

Four of these are self-reported, because same-origin policy prevents a browser reading another domain's root files.

Pillar 4 — Trustworthy (19 points)

(Trustworthy sums above 19 because the self-reported entity question is scored into this pillar from the question set.)

The AI Visibility Score™

One number from 0 to 100, built from four dimensions. Each answers a different question, and each is assessed by its own subset of the twenty-one signals listed below.

The H and A marks are the point of the whole framework. The first two dimensions end with a human reading something an assistant wrote about you. The last two end with software doing something without a human present. Almost every tool in this category measures only the first half.

The dimensions are weighted by how many signals feed them, which is why they do not carry equal maximum points. That asymmetry is deliberate and explained further down.

Offerings, not websites

An assistant is never asked whether a website is good. It is asked who does this specific thing — emergency HVAC repair, commercial fitouts, a bathroom renovation in the inner west. It answers by deciding whether it can confidently recommend a particular service, and that decision is made service by service.

Which means a single score for a whole business hides the thing you most need to know. A firm can be entirely recommendable for one offering and invisible for another, and one number averages the two into something true of neither.

So the unit of measurement is the offering — a service, a product, a topic cluster, a customer intent — and a website gets as many scores as it has offerings.

How the rubric divides

The twenty-one checks do not all belong at the same level, and pretending they do would be the easy mistake.

An offering's score is the domain foundation plus its own result. That has a consequence worth stating plainly: a domain problem caps every offering at once. If there is no Organization schema, no offering can score above the low seventies no matter how well written its page is. Fix the foundation first, then the weak offerings.

What a result looks like

Domain foundation                          24 / 33

Emergency callouts        ████████████░░   88   Agent Ready
Preventative maintenance  ██████████░░░░   74   Recommended
Commercial fitouts        ██████░░░░░░░░   47   Recognised
Refrigeration             ████░░░░░░░░░░   31   Discoverable

Overall                                    60   Recognised

The overall number is a roll-up, and it is the least useful line in that block. It tells you the average of four very different situations. The four lines above it tell you that emergency callouts will get recommended and refrigeration will not, which is the thing you can act on.

How offerings are found

From your own site, in this order: the navigation and its structure, the sitemap, page URL patterns, Service or Product schema where it already exists, and repeated heading language across page clusters.

It will sometimes be wrong. A site with a flat structure and generic navigation is genuinely ambiguous, and we would rather show you the grouping we inferred and let you correct it than quietly score the wrong clusters. Every report lists the offerings it detected and the pages assigned to each, and there is a line to tell us when we have split or merged something badly.

When this does not apply

A single-service business. A sole-trader electrician with one service has one offering, and the offering view is the site view. Nothing is lost, but nothing is gained either.

A site under about ten pages. There is not enough there to cluster. You get the domain foundation and one offering.

Both still get a score and a fix list. They just do not get the breakdown, and the report says so rather than inventing four categories to fill a template.

The five levels

A number on its own means little. The bands give it a position you can say out loud.

Most websites we have run land between 35 and 60. Getting from Recognised to Recommended is usually a day of work; getting to Agent Ready takes deliberate effort, because almost nothing on the open web is built for it yet.

The bands are our definition, not an industry standard. They exist so a score becomes a sentence — "we are Recommended but not Agent Ready" — rather than a number without a reference point.

Every score we report is drawn against two threshold lines: 61, where a website becomes structured enough to be named in an answer, and 81, where software can act on your information rather than only quote it. A score means little on its own; against those two lines it tells you what to do next.

What we do not publish yet is an average. We have not run enough websites to state one honestly, and an invented benchmark would be worth less than no benchmark. When the sample is large enough to mean something, the median and the top decile will appear here, with the sample size beside them.

Where that will come from. Two places, both consented. Every paid audit is a data point. And under the free checker there is an unticked box offering to contribute your score — six numbers and a platform name, no address, no page content, no identifier. What is sent, in full. If you would rather not, do not tick it; the checker sends nothing at all.

How the four component scores are built

Every check is scored against the pillars it serves, so one structural fix moves several scores at once. That is the whole argument for doing the work once, made arithmetic.

A pillar's score is the points earned against the points available for that pillar, so the totals differ — Find you has the most checks behind it because it is the best-understood of the four.

The grades

• 85–100 — AI-Visible ✦. Readable, understandable, safe to cite.
• 55–84 — Findable, with gaps. You will appear sometimes. The gaps explain why not more often.
• 0–54 — Invisible to AI, fixable. Nothing here is hard. Work down the list.

What the score is not

It is not a prediction. It measures whether your page is legible and citable, which is a necessary condition for being named, not a sufficient one. A perfectly structured page in a category you have no authority in will still lose to a better-known competitor.

We would rather publish a rubric you can argue with than a black box you have to trust.

Our own scores

We run this on our own pages and publish the results, including the bad ones.

The first build of this site scored 75/100 on its own checker. Titles were running to 84 characters because we were appending the brand name to everything. There was no machine-readable date on any page. There were no contact details anywhere except a link to a contact page. All three were found by the tool, on us, before launch.

They are fixed now, and every page here passes all fourteen on-page checks. That is the standard we think is reasonable to hold ourselves to before charging anyone for advice.

Changes

Weights change as the field does. Every change is recorded in the changelog.