Nearly half of Britain's biggest brands cannot be read by a machine at all
In July 2026 we scanned 221 of the country's most significant websites through the Crane Index™, our three-scanner AI-visibility suite: the full FTSE 100 and the 100 largest UK online retailers, measured on 19 and 20 July. Of the 221 sites we approached, 106 could not be scored at all. Their own systems refused an automated reader, gave it no response, or served pages that do not exist until scripts run. Of the 115 sites a machine could read, the median score across the three reads was middling on findability, reasonable on whether an agent can act, and only 21 out of 100 on whether a machine can confirm who the brand is. These are the largest, best-funded organisations in Britain, and on the measure that now decides whether ChatGPT, Perplexity or Google's AI Overviews can find, trust and name them, size bought very little. All figures are first-party measurements; the full dated result set is archived, and we will re-run the scan and report the movement.
The sixty-second read
- ▸The finding: across 221 of Britain's biggest brands, 106 could not be read by a machine at all, and the 115 that could sort cleanly into four kinds with one shared weakness.
- ▸The money: when a buyer asks an engine to shortlist suppliers, an illegible brand is not in the answer. The demand does not vanish; it goes to a legible competitor for free, while paid channels buy it back at a rising cost.
- ▸The fix: one structured-entity habit runs through the worst of every group. The checks around it fail together almost perfectly, so a single, cheap correction clears a whole bundle at once.
Test your own site first: the free scanner takes a minute. The two underlying studies cover the FTSE 100 and the UK's biggest retailers in full.
The four kinds of brand
We did not decide these groups in advance. We handed the 115 readable brands to a clustering method with no labels and let it find its own structure from the three scores alone. It settled on four kinds, and the boundaries are the algorithm's, not ours.
The largest group, 41 brands, we would call competent but faceless: the plumbing works, an agent can move around the site, and yet the brand does almost nothing to say who it is, scoring a median 12 out of 100 on representation. A second group of 30, mostly retailers, are slick machines with no identity: strong on operability, still thin on self-representation. A third group of 32 are the relative leaders, rounded across all three reads, and even they reach only 36 out of 100 on representation. The fourth group is the uncomfortable one: 12 brands, many of them household names, sit at the bottom on every axis. They are, in the plainest sense, invisible.
The pattern under all four is the same. British brands have built sites that a machine can operate but cannot vouch for.
The blind spot they share
Read across the three scanners and one axis trails badly in every group. Findability sits at a pooled median of 49 out of 100 and operability at 66, but representation, whether a machine can confirm the brand is a real, single, describable entity, sits at 21. The failures are not scattered oddities either. They are the same handful of specific, fixable things missing from almost everyone at once: 99% of the readable brands fail to span a recognised registry and profile, 98% never mark up what they actually sell, 98% make claims with no sourcing a machine can check, and 94% carry no named, credentialed authorship. The basics of access mostly pass. The layer that lets a machine trust and repeat your commercial facts is where the market is absent.
That is worth sitting with, because it reframes the problem. A brand is not usually losing an AI-visibility race to a sharper rival. It is missing what nearly everyone is missing, which means the ground is open to whoever moves first.
One fix runs through the worst of it
The most useful thing the clustering surfaced is a habit nobody encoded. A small group of structured-entity checks fail together, brand by brand, at correlations as high as 0.98 out of a possible 1.0. Having a defined entity travels with having entity markup, which travels with a consistent identity, which travels with valid structured data. In practice a site either has this layer or it has none of it.
For a business owner that is the good news buried in the bad ones. It means the single most valuable move is not a long programme. It is one correct piece of structured data that declares who you are, links your identity to the profiles that verify it, and states plainly what you sell. Done properly, it clears a correlated bundle of failures at once and lifts more than one of the three reads. It is the first thing we would put on any site, ahead of the slower work of building genuine authority.
Size is no defence
The instinct is that the biggest brands must be the safest. The data says the opposite. The names furthest from the pack, at the very bottom of the readable set, include some of the largest in Britain: Amazon UK, Ocado, Experian and BAE Systems all scored a composite of around 20 out of 100. Heavy applications that render only after scripts run, and aggressive bot defences that turn away automated readers, are precisely what breaks an AI reader, and scale is what funds the migrations, acquisitions and multi-brand tangles that fracture one company into many. Retailers, whose living depends on being found, did modestly better than the blue-chips on every read and by the most on representation, and even they reached only a median of 25 out of 100. The clearest performer in the whole set was White Stuff, a mid-sized retailer that scored well across all three reads for the simple reason that its site presents one coherent, well-described identity.
What this means for you
Three plain questions decide most of it. Can a machine read your site at all, or does your own technology turn the reader away before it starts. Can it confirm who you are from your own pages, without guessing. And does it know what you sell, in words it can quote back without error. For nearly every brand we measured, at least one of those answers is no.
None of this needs a rebuild. It needs the entity layer put right, then the longer work of representation. The free Crane Index scan will tell you which of the three reads is holding you back, in a minute and with no obligation. If you would rather we simply fixed it, that is the work we do.
This is a snapshot, not yet a trend: we hold one scan for each cohort, taken on 19 and 20 July 2026, and the figures throughout describe the 115 brands that returned a full score, never the full 221. We will re-run both cohorts and publish the movement, so the four kinds become a tracked index rather than a single photograph.
