We scanned 62 websites for AI readiness. Not one passed.

There’s a particular silence that follows showing a business owner their website’s AI-readiness score. Not at the number itself — they can handle a number. It comes at the second slide, the one where their number sits inside everyone else’s. We finally have that second slide: the first completed scan of every domain that has ever run through LoopCite, plotted. It is not a flattering picture, and the most useful part of it is not the average.

Between June and August 2026 we pulled the first scored free scan for every domain in our database — 62 of them, one point each, so a site scanned five times counts once. The headline took checking three times because it seemed too clean: not one of the 62 hit all five readiness targets. The best any site managed was three out of five, and only two sites managed that. Thirty-seven of the 62 hit none at all.

That is not a story about bad websites. Most of these are perfectly good sites, built by competent people, that rank fine on Google. It’s a story about a bar that moved while everyone was busy clearing the old one.

What “ready” means, and where the bar comes from

A readiness scan scores five areas, each with its own target — the point where that area stops holding you back. GEO (AI) Readiness: can AI engines reach, read and understand the site. SEO: the classic foundations, which still feed AI answers. Trust & Integrity: the security and credibility records engines lean on. Multilingual: whether you exist for customers who don’t ask in English. Agent-Commerce: whether an AI agent could actually transact with you.

Two things worth being upfront about before any number gets quoted. The targets are per area, and there is no single overall pass mark — the low-80s figure is derived from the five targets and how they’re weighted, not declared as a threshold. And this is readiness, not visibility: nothing here measures whether ChatGPT or Google’s AI actually names these businesses. Readiness is what a site controls on its own pages. Being cited is earned out in the wider web over months. They are different questions, and anyone blurring them is selling you something.

The shape of it

Median 62. Range 19 to 78. Twenty-three of the 62 — 37% — came in under 50. The whole distribution sits left of the goal, and there is no right tail: the single highest first score across 62 independent businesses was 78.

But the more useful detail is the shape. It has two humps — a cluster in the 40s, a cluster in the 60s, and a real trough between them. That survives 10-point bands, 5-point bands and a density estimate, so it is in the data rather than an artefact of how it’s drawn.

Two humps means two different problems wearing the same label. The upper cluster is roughly 11 points short. Those sites are close — someone has done the work, the structure is largely there, and what’s missing is a specific, finite list: a schema graph, a crawler rule, a security header, a translated page. That’s a weekend, not a rebuild. The lower cluster is roughly 40 points short, which is a different conversation entirely — usually something structural, like content that only exists after JavaScript runs, or no machine-readable description of what the business does at all.

Anything that treats “AI readiness” as one problem with one answer will be wrong for about half the room.

Why nothing got past 78

A ceiling that clean across 62 unrelated businesses usually means it isn’t about the businesses. It isn’t. It’s arithmetic. Two of the five areas — the two newest, which between them carry a fifth of the score — sit near their floor for almost every site in the sample, and once they’re pinned there they cap everything above them.

Run the numbers and it’s stark: a site that leaves those two areas at the sample median tops out in the low 70s even if it hits every one of the other three targets exactly. Perfect scores across all three classic areas still don’t clear the mid-80s. There is no amount of conventional excellence that gets you past it, because the composite is weighted and two of the weights aren’t in play. The top of the distribution says the same thing from the other direction: the four highest-scoring sites are the only four that scored meaningfully on both of the newest areas.

That isn’t a flaw in the measurement. It’s the clearest statement in the data of where the open ground is — the gains available on the things nobody is working on are larger than the gains available on the things everybody is.

Area by area: the gap is widest where it’s newest

SEO is the strongest area — and it’s still short. A median of 80 against a target of 90, with only 12 of 62 sites reaching it and eighteen coming in under 70. Two decades of an entire industry teaching this discipline, tooling built for it, agencies selling it — and the median site is still a full 10 points off on the oldest, best-understood area in the set. That sets a fairly sober expectation for the rest.

Trust is the quiet failure. A median of 66 against a target of 90, and only 2 of 62 sites reached it. What’s striking isn’t the average, it’s the uniformity: the middle half of all sites sits between 58 and 76. There is no cluster of sites that are good at this. Nobody is winning trust because almost nobody is looking at it — it’s the area with no industry attached, so it quietly goes unattended.

GEO Readiness carries the largest share of the score and has a median of 69 against a target of 80. Twenty-three sites scored under 50, which means that for more than a third of this sample the problem isn’t polish — AI engines are having genuine trouble reaching, reading or understanding the site at all.

Multilingual sits on the floor — a median of 20 against a target of 60, with 53 of the 62 sites scoring exactly the same number. That’s what an area looks like when it simply isn’t being attempted. And Agent-Commerce barely exists yet: a median of 30 against 60, on the newest and most future-tense question of the five.

The pattern is consistent and points one way: the older the discipline, the better the scores. SEO, then GEO, then trust, then multilingual, then agent-commerce. The web is well optimised for a reader that stopped being the only reader.

Why competent people have sites that score 40

It’s tempting to read a chart like this as a competence story. It isn’t, and getting that wrong leads to the wrong fix. Every site in the lower cluster was built to a brief that was correct when it was written: look good, load fast, convert visitors, rank on Google. Not one of those goals is wrong now. They’re just no longer sufficient, because a second reader arrived — one that doesn’t scroll, doesn’t look at your design, can’t wait for your JavaScript, and doesn’t infer from context what your business does. It reads what’s machine-readable, decides whether you’re quotable, and moves on.

Nothing in a normal business’s toolkit reports on that reader. Analytics won’t. Your rankings won’t. The site looks fine, converts fine, and quietly fails a test nobody mentioned was being administered. That’s the real finding here — not that sites are bad, but that the failure is silent. A median of 62 across 62 businesses, none of whom knew that was their number until they ran a scan.

Read this before you quote any of it

These caveats are the price of publishing your own numbers, and they cut against the story in places. This is readiness, not visibility. It is not a random sample — 62 domains that came through our scanner, over-representing South African small and mid-size businesses, self-selected rather than surveyed. It is historical by construction, because a “first” score is the oldest one on record for that domain and several sites have improved since. Three of the 62 are ours, scanned in June at 36, 40 and 40 — we were not exempt from our own finding; excluding them moves the median from 62 to 64. Two scans hit a bot challenge and may have measured a block page rather than the site. The multilingual signals read the homepage, so a site with a translated page deeper in but an English-only front door scores low. And n = 62 is small — treat every figure here as a direction, not a constant.

What it means if you run a website

Mostly this: your competitors are also short. Every one of these 62 businesses is behind, and so is everyone they compete with. The whole field is early, which makes this a land-grab window rather than a catch-up scramble — and land-grab windows have the irritating property of closing quietly.

The practical read is that the gap is heaviest exactly where there’s no incumbent to call. If your problem is SEO, there’s an entire industry to hire and it’s good at its job. But credibility records, machine-readable identity, language coverage, agent-readability — 2 of 62, 53 on the floor, almost nobody started. Those aren’t crowded lanes, and the work in them is measured in hours rather than quarters.

None of it guarantees an AI will cite you. Be wary of anyone who says otherwise — engines decide what to quote and no vendor controls that. What you can control is whether you’re eligible: reachable, readable, understandable, trustworthy. Readiness moves fast, often within days. Being consistently recommended is slower and compounds over months.

A median of 62, a ceiling of 78, and not one site in 62 clearing all five. Right now that isn’t a competitive disadvantage — it’s a room full of people who haven’t checked.