Crane Index™ Research
Deterministic, archived, and built to be checked. This is the full method behind the Crane Index™, the exclusions we apply, and how anyone can re-run a score in about a minute.
What the Index measures
The Crane Index™ gives every organisation the same three deterministic reads of its home page, the front door a machine meets first. Each read is a rule-based check, not a judgment, and returns a score out of 100. The composite Crane Index is the blend of the three.
The market is settling on a name for this work: generative engine optimization (GEO), sometimes answer engine optimization (AEO), the discipline of being found, read and cited by AI search rather than only by classic search. The Crane Index measures the site-side half of it, whether a machine can read and represent you, because that is the part you control and can fix.
- AI Search ReadinessCan the engines that read live sites, such as AI Overviews, Perplexity and ChatGPT with search, resolve the organisation and extract what it does and sells.
- Agent ReadinessCan an automated agent get in at the edge and find anything machine-operable when it arrives.
- Brand RepresentationCan a machine confirm from the site itself which real-world organisation it is.
What this is, and is not: an on-page read of the home page, the part an organisation controls on the page. It is not a crawl of the whole site, and it does not watch AI answers or track whether an engine currently cites the organisation. It measures whether a machine can read and represent you, which is the part you can fix.
The scale
Every score sits in one of four bands, so a table can be read at a glance.
How to read a table
Read the columns against each other and the story sharpens. Rank is by the composite Crane Index™; the three reads follow. A low score says a site is hard for a machine to read, not that the business is anything other than excellent at what it does. Sites a reader could not enter are excluded as unmeasured rather than scored zero, so the medians describe only the most legible end of the field. The single habit that separates the top, valid Organization structured data and one resolvable identity, is naming consistency and accurate markup, not a rebuild.
How a cohort is built
Each benchmark is a defined cohort, the FTSE 100, the biggest online retailers, or an industry of around a hundred and seventy leading organisations. The constituent list is assembled from public sources, sanity-checked by a person, and pinned alongside the archived results so it is inspectable rather than convenient. If we cannot resolve an organisation to a single scoreable parent domain, that is itself an AI-legibility finding, and it is reported as unmeasured rather than guessed.
Anyone can ask for a domain to be added. A requested domain joins only after a human review, and the bar is simple: a real trading organisation of substance in that industry. A site built to sit well in a benchmark rather than to serve customers does not qualify, and requested-in constituents are recorded as such in the archived lists, so the composition of every cohort can be audited.
What we exclude, and why
Exclusions are applied by site type, and before any score is seen, never because a domain scored low. The Index's authority is that it is measured and re-runnable by anyone, so we never omit a domain for its result. We do set aside site types where the benchmark is not a fair test, and we disclose it.
Search engines and chat-box homepages are excluded. A search engine's or an AI assistant's home page is a query box, not a content site, so the question this Index asks, can a machine read and cite your content, does not apply. The line is the domain type, not the company: an AI company's corporate site is a normal content site and stays in, only the query or chat-box domain is set aside.
Unmeasured, not failed
Some sites cannot be read by an automated reader at all: a few refuse one outright, some give no response, and some serve pages that are empty until scripts run, which is how a site looks to the many machine readers that do not execute them. These are reported as unmeasured and set aside, never scored zero. It follows that the medians in every table are drawn from the most legible end of the cohort by construction, so the true picture is unlikely to be better than what we publish. Our reader is one unfamiliar bot, so a site that verifies crawlers by network may admit engines it recognises while refusing ours, which is exactly why a blocked site is reported as unmeasured, not as failed.
Re-run, archived, and checkable
Every cohort is re-scanned on the first of each month, and the full dated result set is archived, so the benchmark is a living record and the movement is published, not just a snapshot. Every score in every table can be re-measured live in about a minute on the same free scanners, so nothing here has to be taken on trust.
Integrity of the measurement
Three properties keep the tables honest. First, no one can affect anyone else's score: the scanners read only the target's own public site, and nothing a third party submits touches another organisation's row. Second, the monthly scan compares the page served to our reader with the page served to an ordinary browser, and a site that shows the machine a materially different, schema-rich page is flagged for review rather than banked. Third, we re-measure a sample of leading rows mid-month, unannounced; a moved score is usually a site that changed, and where re-measurement and the published table disagree we say so. Equal composites are published as joint standings, with the alphabet deciding only the order rows are listed in, never the claim.
If your organisation is named here
A score is a measurement of a page on one dated day, nothing more, and a low score says a site is hard for a machine to read, not that the business is anything other than excellent at what it does. If you are named in a cohort you can do three things at once: re-measure your own score live to confirm it, ask us to re-scan you if your site has changed, and add or correct a domain for the next monthly run. A named organisation is never left without a right of reply.
Licence and how to cite
The Index and its published figures are released under a Creative Commons Attribution 4.0 licence: free to quote, chart and republish with attribution. The clean citation is:
The Crane Index™, the AI visibility benchmark. The Crane Consultancy. thecraneconsultancy.com/research/
For the press pack, headline findings and downloadable assets, see the newsroom. For a question about the method, write to hello@thecraneconsultancy.com.
Questions about the Crane Index
What does the Crane Index measure?
How readable and well-optimised a site is for AI, drawn live from its real pages across three deterministic reads: whether an AI engine can retrieve and cite it, whether an agent can act on it, and whether it is described accurately. Each is scored out of 100, and the composite is the Crane Index.
Why does AI readability matter now?
Because discovery is moving. In 2026, Google reported that its AI Mode in Search had passed a billion monthly users. What an assistant can read of an organisation comes down to the machine-readable layer of its site, the structured data Google Search Central documents.
How often is it re-run?
Every month, on the first, so the movement is a matter of public record rather than a single snapshot. Every constituent is named, ranked and archived, and each score can be re-run live against the site as it stands today.
The senior read,
once a week.
Most marketing newsletters are noise on a schedule. This is not that. Once a week I send a short, considered read on the shifts that actually change what you should do, across the AI shift, PPC, technical SEO and analytics. Filtered, weighted and signed off by me before it goes out. The judgement is mine, and now you can hear every edition in my own voice.
Nic CraneFounder, The Crane Consultancy
Join readers of The Crane Standard.