The first Crane Index™ Industry report publishes on the first of September, and a fresh one on the first of every month after. It puts 3,241 organisations across nineteen benchmarks through one identical test: how clearly each website reads to the automated readers that AI engines such as ChatGPT, Gemini and Google's AI Overviews now send ahead of a person. When a buyer, a patient, a client or a student asks one of those engines a question, the answer is built from the sites the engine can read cleanly, and it arrives before anyone opens a brochure. How well a machine reads your pages is what we call your AI legibility, and name recognition does not decide it. Construction does.
The benchmarks run from the FTSE 100 and the UK's biggest online retailers to seventeen measured industries: financial services, insurance, legal and professional services, technology, telecoms, healthcare, automotive, manufacturing, energy, consumer goods, media, travel, hospitality, transport, property and education. An eighteenth, the UK's universities, is built and takes its first measured run in the September scan, so it carries no score yet. Each organisation is scored the same way, on its own live website, with nothing self-reported and nothing that can be bought.
Machines read your rendered text, not your reputation
Language models work from the text your pages actually render, so legibility is a property of how a site is built rather than how well known its owner is. State in plain prose what you sell, to whom and on what terms, and an engine lifts those facts and repeats them. Bury the same facts behind a filter, a menu or a search box that only responds to a human clicking through it, and a machine reading top to bottom never reaches them; the question then returns a thinner answer, or a competitor's name. A low score does not mean a business is invisible to these engines. It means the engine finds less to lift from the site itself, so its answer leans on whatever else it can find.
Three questions, and how to run them yourself
The report scores three behaviours, and you can put the same three questions to an engine today:
- Retrieval. Ask a buyer's question in your category. Does the engine find your business at all?
- Extraction. Does it pull what you sell, to whom and on what terms, correctly?
- Naming. Does it name you in the answer, or describe your offer while crediting someone else?
An engine that cannot read your offer cleanly still answers; it fills the gap by inference, and an invented price or a mis-stated term is worse than silence to someone who cannot easily check it. You never see the enquiry that goes elsewhere, because it was settled inside a sentence you had no hand in writing. You can see where your own site stands in about a minute with the free scanner at thecraneconsultancy.com/crane-index-scan/: one address, all three reads, a composite out of 100.
How the report is formed
Twenty benchmarks, each with a list of recognisable organisations of substance, 3,241 in all. On the first of every month each site on each list goes through the same three measured reads, scored out of 100, and each benchmark is published with its strongest sites named. The first industry-wide report lands on 1 September 2026, and because the scan repeats monthly, every standing is re-measured and re-earned each month after that. Nothing is self-reported: a household name is read the same way as an independent, from its own live pages.
Old SEO, pointed at a machine
Getting read by a machine is not a new discipline. As the web fills with text that says a great deal and states little, a site that describes itself plainly stands out to the engine sorting the two, and Google's own guidance now treats scaled, low-value content as spam. Several tools sell this as a brand-new field with a monthly fee and a proprietary visibility score only the vendor can read. The honest position is that no language model consults a vendor's dashboard when it builds an answer; it reads your page. So is our own work, linked below, the same subscription under another name? No, because of what changes hands. A recurring fee buys a number you cannot verify. The work we do is a fix that finishes: plain prose that states what you sell, and Schema.org structured data that labels it for Google's retrieval and AI Overviews, are one-time construction jobs inside ordinary AI search and technical SEO work. You revisit that text when your business changes, not when a vendor ships an update. Run the test on your own site before you approve another quarter's spend.
