Almost nothing is ready for AI search. We scored 850,000 sites to be sure.

Since the start of 2026 we have pointed the same audit at different slices of the web and published each as a SAVI edition, the State of AI Visibility Index. One reads the entire world. One reads the UK. One reads a single English region. Two read individual professions, accountants and dentists, right down to the practice. Different populations, different sample sizes, one question throughout: how ready is each site to be picked by an AI assistant?

Any single edition is a snapshot. Put them side by side and something more useful appears, a pattern that holds no matter how far you zoom in. The AI visibility gap is not concentrated somewhere convenient you can steer around. It is everywhere, at every level of magnification.

The one number that matters

When someone asks an AI assistant to recommend a business, it puts forward two or three names, not ten. To be one of them, a site has to clear the top band of our 0 to 100 scale, "AI-Ready", which starts at 80. Here is how many websites get there.

Those are not typos. Worldwide, roughly one site in five hundred is ready. Across the entire UK cohort, the number rounds to none. And the single signal that most directly earns an AI citation, laying your answers out in the simple structured format the engines lift from, is missing on effectively every business we have measured, in every edition.

The same story at every scale

This is the part that makes the benchmark worth reading as a whole rather than one report at a time. Zoom from the global web down to a single profession and the average creeps up a little, but the ceiling never does. Almost no one is ready, wherever you look.

Sources: SearchScore SAVI editions, 2026. Each site is scored across 250+ AI search visibility signals. "AI-Ready" is a score of 80 or higher. City editions (London, Birmingham and Manchester accountants) show the same shape at smaller samples.

Why the world looks worse than the professions (it doesn't)

Read the table too quickly and you might conclude UK accountants and dentists, at 53, are twice as good as the global web at 34. They are not. This is the single most important caveat in the whole benchmark, and it is a lesson in reading data honestly.

The international average is drawn from 850,000 sites of every kind, the entire long tail of the web, including abandoned pages, thin listings and sites that were never trying to be found. The UK sector editions are curated cohorts of real, trading professional practices with staffed websites. A higher average there reflects a cleaner population, not superior work. The like-for-like comparison is not the average. It is the ceiling, and the ceiling is the same everywhere: almost nobody clears 80.

The pattern behind the numbers

Look under every edition and the same two-part story repeats, almost word for word.

The door is open. In every cohort, the overwhelming majority of sites let the AI in, the crawler directives are fine and the pages are technically reachable. Across editions, "technical basics" is consistently the strongest area we measure, scoring in the mid-70s. Whatever platform businesses build on handles that part for them.

The table is bare. Then, in every cohort, the same sites fail the parts that actually win the recommendation: labelled, structured information an AI can lift, and clear signals of a known, trusted brand. "Structured data" and "brand authority" are the two weakest areas in edition after edition, scoring in the mid-30s. The content exists. It is simply never labelled in a way a machine can read, quote and trust.

That is why the fix is cheaper than it looks. Nobody in this data is losing because their advice is bad or their business is weak. They are losing because their site was built to be read by people and never relabelled for machines, and the relabelling is largely a technical job that can be done once.

What it means for your business

The strategic read is almost optimistic. In a market where 99.8% of sites are not AI-Ready, the bar to stand out is astonishingly low. An AI assistant naming three businesses in your category is choosing from a field where almost no one has done the work. The few who do are not better accountants, or better dentists, or better anything. They feed the machine better.

The concrete steps are the same in every edition we have published:

• Say plainly what you do, who for and where, in text specific enough for a machine to match to a real question.
• Label your answers in the structured format AI quotes from, the single highest-leverage change most sites can make.
• Name the real, credentialed people behind your content, in a way the machine can read, not an anonymous corporate voice.
• Earn genuine mentions on sources the AI already trusts.
• Keep the same business described consistently everywhere it appears.

None of that is a new file format or a growth hack. It is the fundamentals, and this benchmark is the evidence that they are still unclaimed almost everywhere.