We measured 221 of Britain's biggest websites. The league table is wide open.
In July 2026 The Crane Consultancy published its first two Crane Index™ research benchmarks: the FTSE 100 and the UK's top 100 online retailers, 221 sites approached in all (the FTSE 100 resolves to 79 scannable primary domains, and the retail cohort is drawn deliberately wide at 142 sites so the top 100 sits inside it whichever ranking you prefer), each given the same three reads. Can the engines that answer buyers, AI Overviews, Perplexity, ChatGPT with search, find and cite you? Can an automated agent reach your site and act on it? And can a machine confirm, from your own pages, who you are? Three scores out of 100, and a composite: the Crane Index™. The headline from both cohorts is the same, and it is not that somebody is winning. Not one site in either cohort scored 70. Nearly half would not let an automated reader in at all. Britain's biggest websites are competing ferociously for humans and, measurably, almost not at all for machines.
How the Index measures, in plain terms
Each scan reads a site the way a machine does, not the way a visitor does. The AI Search read resolves the company as an entity, extracts what it sells and to whom, and checks whether those facts survive being repeated; it weighs entity coherence, structured data, and whether the copy contains concise, extractable answers. The Agent read checks the door and the plumbing: whether the crawlers AI engines send are admitted, and whether there is anything machine-operable when they arrive. The Representation read asks the question boards assume is unaskable: can a machine confirm from your own site which real-world company you are, via identity markup, disambiguation links and authority signals. Every check is deterministic, every result is archived with the date, and any score in either table can be re-measured in about a minute on the free scanners, by you or by your competitors. That is deliberate. Research you cannot verify is marketing, and we wrote these benchmarks to be checked.
What the two cohorts revealed
The FTSE 100 scan found that budget does not buy legibility: of the 38 sites a machine could read, the median AI Search score was 48, nothing cleared 80, and the representation read collapsed to a median of 15. The retail scan found the door problem: 44 of the UK's biggest storefronts refused an automated reader outright, and of the 77 sites that could be read, the top of the entire field was Riverford, the Devon veg-box company, at 62, with Amazon UK's storefront at 20. Different sectors, same mechanism. In both cohorts, every one of the ten strongest sites carries valid Organization structured data, and not one of the bottom ten does. Size never predicted the score. Structure always did.
The one-minute test, and what to do if you fail it
In Perplexity, or ChatGPT with search turned on, paste this: "Using only [yourdomain.com], name this company's three main products or services and the specific buyer each one is for." If it misses one, muddles one, or invents a buyer, you now know what the engines shortlisting your category know. The fix is not a rebuild: it is valid Organization and Product markup matched to your visible copy, one name rendered identically everywhere, and a deliberate decision about which machine readers your site admits. Both reports carry the full ranked tables, named, with every company's category-level breakdown one tap deep, and a board-ready PDF for circulating internally. The league table is wide open; on this evidence, the first company in any category to take machine legibility seriously gets to own it.
