Four in a hundred. That is how many of the 2,174 companies a machine could read on 1 September 2026 scored 70 or better on how well an AI system can understand their website. Not one of them was a FTSE 100 company. Two of them are based in the South West.

Those figures are from the September run of the Crane Index™, the benchmark we publish every month across nineteen industries. It is a first-party measurement, free to quote, and anyone can re-run it on their own site in about a minute.

What is actually being measured

Every month we approach the leading organisations in each industry and score their websites out of 100 on the three things a machine has to do before it can recommend you to anybody. Can it reach and read the page. Can it act on what it finds there. Can it confirm, from the site itself, who the company is. The three combine into one composite score.

What this measures is legibility to software, not how a site looks or how much traffic it gets, and it matters now because of how people increasingly ask. Put a question to an assistant and you do not get ten blue links to choose between. You get two or three names. Being one of those names depends on whether a machine could read you well enough to be confident about what you do.

The September run approached 3,178 organisations. 2,174 returned enough for a machine to score at all. Across the nineteen industries the typical median was 44 out of 100. The strongest field was Technology and Software, at a median of 57. The weakest was Automotive, at 40.

The top of the September table

OVO Energy is headquartered in Bristol. Wessex Internet is a full-fibre provider built for rural connections, based in Dorset and serving Dorset, Somerset, Wiltshire and Hampshire.

Both out-score every FTSE 100 constituent we were able to read. Of the FTSE 100, 79 were approached and 40 returned a page a machine could score. The strongest of those was GSK, at 64. None reached 70. The UK’s biggest online retailers tell the same story: 142 storefronts approached, 77 readable, none at 70, the field led by Office at 68, with Seasalt Cornwall fourth at 63.

Structure, not spend

The reason a rural broadband company can lead a national industry table is that this score responds to structure rather than budget. Nearly everything the Index measures is a decision about how information is arranged, and nearly none of it is expensive.

Does the site say plainly, in text rather than inside an image, who the company is, what it sells and where it operates. Is that identity written the same way everywhere it appears, so that a machine reading three sources arrives at one answer instead of three. Can the pages that matter be reached and read without running any JavaScript, given that plenty of the systems now reading the web do not run it. Does a page answer the question in its own heading in the first sentence, rather than four paragraphs down. Are the claims and credentials on the site linked to somewhere that verifies them.

None of that is a redesign. Most of it is a week of unglamorous work that no one will ever compliment you on.

What this means if you run a creative business

Two things, pointing in different directions.

The first is about your own firm. Agencies and studios sell judgement, and judgement is exactly what a machine finds hardest to verify from a website. If your site is a beautiful showreel with the substance carried in images, video and PDFs, you have built something that reads brilliantly to a human being and thinly to everything else.

The second is about your clients, and it is the more interesting of the two. In eighteen of the nineteen industries we measure, even the leader scored below 80. The top of almost every table in the country is genuinely unoccupied. The work that moves it is information architecture, copy that answers before it flourishes, and consistency across everything a business publishes. That is not a new discipline anyone needs to hire for. It is what good creative and content people already do, pointed at a reader nobody briefed them on.

Check where you stand

Any site can be run against the same three reads at thecraneconsultancy.com/crane-index-scan. It is free for three different sites a day and there is no account to create. The full tables, the method and the reports are at thecraneconsultancy.com/research, published under a Creative Commons licence so anyone can quote or republish them.

The score itself is the least useful part. What is worth your time is the list of specific things a machine could not work out about you, because that list is short, it is boring, and almost nobody has done it yet.

Every table is re-scored on the first of each month, and the movement becomes public record. For now, the top of almost all of them is unclaimed.

Nic Crane is founder of The Crane Consultancy and publisher of the Crane Index™, the monthly AI readability benchmark.

Every month, our research programme reads more than 3,200 UK company websites the way an AI system does, scores what it finds, and publishes the results. It covers 19 industry benchmarks, from the FTSE 100 and ecommerce to law, travel, property, healthcare and manufacturing. We built it in Bristol, it is free to read, and I want to explain what it measures and why a business anywhere in the South West should care.

Why measure AI readability at all

Buyers have started asking AI assistants for recommendations and receiving written answers instead of lists of links. Those answers are assembled from websites the AI systems could actually fetch, parse and trust. Which meant a question nobody could answer when we started: how readable are British company websites to the machines now doing the recommending? Opinion was everywhere. Measurement was nowhere. So we built the instrument.

What The Crane Index actually measures

Each site gets three machine reads. Retrieval asks whether an AI system can fetch the site and extract its meaning: the facts in the code, the structured data, the clarity of what this business is. Execution asks how the site behaves when automated agents call: whether they are served, throttled or turned away at the door. Representation asks whether the machine’s picture of the business matches reality across the open web: consistent identity, verifiable credentials, links that hold up. The three reads combine into a composite score, and each industry’s table is published with every figure measured on a stated date.

The rules we hold ourselves to

Nothing is estimated. If a read could not be completed, it is reported as unmeasured rather than guessed, and no composite is published for it. Equal scores are published as joint standings rather than split by a tiebreak. And the whole programme re-runs monthly, so a standing is a dated fact rather than a permanent badge. Sites move up when they improve and down when they decay, which is precisely what makes the record worth watching. A subscription cannot alter a public score; the research is editorially independent of anything we sell.

What it means for a business in Bristol or Bath

The companies in the benchmarks are household names, but the findings transfer directly to any business. The characteristics that make a FTSE 100 site readable to AI are the same ones that decide whether a Bristol firm gets cited when someone asks an assistant for a recommendation: facts in the code, doors open to agents, an identity that checks out. None of them require a famous brand. Most of them are fixable in weeks, and because the reads are measurements rather than opinions, you can prove the fix landed.

Where to look first

The full tables are free at our research library, industry by industry. If you want the same three reads run on your own site, the scanner is free too and takes about a minute. And I write up what the measurements are teaching us each week in The Crane Standard, published every Tuesday morning. British businesses spent twenty years learning to be visible to Google. The next few years are about being legible to the systems that answer on Google’s behalf, and the ones treating that as a measurement problem are already pulling ahead.

Nic Crane is the founder of The Crane Consultancy, the Bristol built practice behind The Crane Index, working with businesses across the South West on AI search, paid media and technical SEO.

Ask three systems how your marketing performed last month and you will get three answers. Google Analytics says one thing, the ad platforms claim another, and the invoices in your accounts software tell a third story. Most businesses live with the disagreement, glance at whichever number looks healthiest, and quietly stop trusting all of them. That erosion of trust is more expensive than any single bad campaign, because it means decisions get made on feel.

The disagreement is structural, not sloppy

Each system counts on its own rules. Ad platforms attribute sales to themselves generously, because their attribution models are written by the party being marked. Analytics tools count what their tracking could see, and tracking loses a slice of reality to consent choices, blockers and browsers. Your accounts count what actually happened, but weeks later and with no idea which channel caused what. None of these systems is lying. They are answering different questions, and the trouble starts when a report treats them as interchangeable.

What one source of truth actually means

It does not mean another dashboard. It means choosing, once, the definitions your business runs on, and forcing every number through them. Revenue net of VAT, because that is the money you keep. Spend as the platforms actually charged it, daily, reconciled against what left the bank. Profit on the same definition in every channel, so a pound from paid search and a pound from email are finally comparable. Once the definitions are fixed, the disagreements become explainable gaps rather than mysteries, and the number in the Monday meeting is the same number your accountant would produce.

We built our client reporting, Crane Intelligence, on exactly this principle after years of watching platform dashboards flatter themselves. The test we hold it to is blunt: the numbers must agree with the bank. If a figure cannot be measured, it is reported as unmeasured. It is never estimated into the gaps, because one invented number quietly poisons trust in all the real ones.

The weekly habit that changes behaviour

Reporting that arrives once a month arrives too late to change anything. The rhythm that works is small and weekly: true spend against budget, revenue on your definition, and anything that moved enough to deserve a decision. Twenty minutes, same morning every week. Businesses that adopt this stop having month end surprises, not because performance improves overnight but because drift gets caught while it is still cheap to correct.

Questions worth asking this week

Does your reported revenue figure match what your accountant would calculate for the same period? Does reported ad spend reconcile to your bank statement, to the penny? And when a number is unknown, does your reporting say so honestly, or fill the silence with an estimate? If any answer makes you wince, the fix is not a prettier dashboard. It is definitions, reconciliation and the discipline to publish only what was measured.

Marketing has spent two decades earning a reputation for numbers that do not add up. The businesses quietly winning are the ones that decided theirs would.

Nic Crane is the founder of The Crane Consultancy, working with businesses across Bristol and Bath on marketing measurement, paid media and AI search. He writes a weekly note on these subjects, The Crane Standard, published every Tuesday.

Bath businesses are spoilt for choice on agencies and starved for ways to tell them apart. Every pitch deck promises transparency, every case study glows, and every proposal has a dashboard in it. The only reliable way I know to separate a measurement culture from a presentation culture is to ask questions the deck was not built to answer. These are the five I would ask, having sat on both sides of the table.

One: where does this revenue figure actually come from?

Ask them to trace one number in their sample report back to its source. Platform reported revenue, analytics revenue and what lands in the bank are three different figures, and they rarely agree. An agency with a measurement culture will tell you which one they report, why, and what its known gaps are. An agency selling theatre will say the word integrated and move on.

Two: tell me about something you advised a client not to spend

Any agency can spend money. The valuable ones stop you spending it badly. Ask for a real example of budget they handed back, a channel they recommended pausing, or a campaign they killed on the evidence. If every story ends with spend going up, you are hearing a sales history, not a management history.

Three: will your numbers agree with my accountant’s?

This is the quiet test most reporting fails. Marketing figures gross of VAT, on platform attribution, against bank reconciled reality will drift apart within a quarter, and the meeting where the owner notices is always uncomfortable. Agree the definition of revenue and profit before you sign, and expect reporting built to match it. We call this having one source of truth, and we consider it the foundation of everything else.

Four: who reads what my money actually bought?

In paid search, ask who reads the search terms. In SEO, ask who looks at the queries you were actually found for. Automation and AI now decide much of where budgets flow, and the platforms’ defaults favour the platforms. The protection is a named human whose job is to audit what was bought, on a schedule, with the findings shown to you. Ask for the name.

Five: what happens when AI answers instead of Google ranking?

Buyers increasingly ask ChatGPT and Google’s AI features for recommendations and receive a single written answer rather than a list of links. Ask a prospective agency what they are doing about it. You are not listening for buzzwords, you are listening for measurement. Can they tell you whether AI systems can read your site today, and can they prove a change moved the needle? We publish monthly research measuring exactly this across thousands of UK company websites, because we think the honest answer starts with a number, not an opinion.

The pattern behind all five

Each question has the same shape: it asks the agency to show its working. Good agencies in Bath and Bristol will enjoy these conversations, because rigour is what they sell. The ones who bristle have told you something useful too. Marketing spend is one of the few costs in your business that can be measured to the pound. Choose a partner who behaves as if that is true.

Nic Crane is the founder of The Crane Consultancy and works with businesses across Bath, Bristol and the South West on paid media, AI search and marketing measurement.

If your paid media report leads with impressions, clicks and click through rate, you are being shown activity, not results. A Bristol business paying for Google Ads or Meta ads deserves one thing above everything else: a clear, honest account of whether the money spent came back as profit. Everything else in the report is decoration.

I have spent years running paid media for businesses across the South West, and the gap between what agencies present and what owners actually need has never been wider. Here is what I believe you should expect, whether you work with us or anyone else.

Spend you can reconcile to the bank

The first test of any report is whether the spend figure is true. Not a rounded number, not a pacing estimate, the actual daily cost the platforms charged, matching what leaves your bank account. If your agency cannot show you spend to the penny by day, they are reading the same dashboard you could read yourself. This sounds basic. It is remarkable how rarely it happens.

Revenue on your definition, not the platform’s

Google and Meta will both happily claim the same sale. Each platform grades its own homework, on its own attribution rules, and the sum of their claimed revenue is routinely more than you actually took. A serious agency reports revenue on one definition you have agreed, net of VAT where that is how you think about money, and is upfront about what is measured and what is inferred. If a number cannot be measured, it should say so, not be estimated into the gaps.

Proof that someone is reading the search terms

Automation has made it easy to spend money and hard to know where it went. Broad match keywords and automatically created assets can quietly widen what you pay for, and the platforms now switch some of these features on by default. The discipline that protects you is old fashioned: a human being regularly reading the actual search terms you paid for, cutting the waste, and telling you what they cut. Ask to see it. The reaction to that question tells you most of what you need to know.

Budget pacing, not month end surprises

Overspend discovered on the last day of the month is not a reporting problem, it is a supervision problem. Daily pacing against a monthly budget, with someone accountable for the trajectory, is the difference between a managed account and a monitored one. We built our own paid media discipline, Crane Engine, around exactly this: true daily cost, budget pacing, and guardrails that flag drift while there is still time to act, because we could not find reporting we trusted enough to run client money on.

The question that cuts through everything

Ask your agency this: if we halved the budget next month, what would we lose, and how do you know? An agency that manages your account on evidence will have a considered answer with numbers behind it. An agency that manages your account on dashboards will talk about visibility. Bristol has no shortage of capable agencies, and the good ones will welcome the question, because it is the one they ask themselves every week.

Paid media is the most measurable marketing there is. Hold it to that standard and it will earn its keep. Let it drift and it will happily spend whatever you give it.

Nic Crane is the founder of The Crane Consultancy, a Bristol based practice running paid media, AI search and technical SEO for businesses that want their marketing numbers to agree with their accounts.

When someone asks ChatGPT or Google’s AI Overviews to recommend a firm like yours, the answer is written in seconds. That is the new front door. The old question was where you ranked. The new question is whether the machines assembling the answer can read you at all.

I run The Crane Consultancy, and every month we measure how readable company websites are to AI systems. It has changed how I think about search entirely, and I want to share the practical version for businesses here in Bristol and across the South West.

Why AI search is different from ranking

A search engine gives a person ten options and lets them choose. An AI assistant gives them one answer, perhaps with two or three sources cited. If your site is one of the sources, you inherit the trust of the answer itself. If it is not, you are invisible in a way a page two ranking never was, because the buyer never sees a list at all.

The systems writing those answers do not browse the way a person does. They fetch your pages, parse the raw code, and try to work out what your business is, what it sells, and whether the facts hold together. A site that looks polished to a human eye can be close to unreadable to a machine.

The three things that decide whether AI can use you

First, retrieval. Can an AI system actually get your content and understand it? That comes down to how your pages are built, whether your key facts sit in the code or are painted in later by scripts, and whether structured data tells the machine plainly who you are.

Second, behaviour. When AI agents visit, does your site let them in or does the firewall turn them away? Plenty of businesses block the very systems they want to be recommended by, without ever knowing.

Third, representation. When a machine reads your site, does the picture match reality? Consistent names, addresses, credentials and links across the open web are what let an AI system say this business is real and worth citing.

What we found measuring 3,241 UK websites

Our research programme, The Crane Index™, runs these three reads every month across 19 industry benchmarks, from the FTSE 100 to travel, legal, property and ecommerce. Every figure is measured on a stated date, published, then re-measured the following month, so the standings genuinely move.

On 6 August 2026 we scanned 3,241 UK company websites. 1,021 could not be scored at all: 577 refused the request outright, 326 never answered, and 118 hid their content behind scripts. Of the 2,220 that could be read, 20 scored 70 or better.

The pattern that matters for a local business is simple: readability to AI has very little to do with how big or expensive a website was to build. It is a set of specific, fixable characteristics, and businesses that fix them early are being cited while their competitors wait.

What to do this quarter

Get a measured baseline before touching anything. You can run a free scan of your own site and see the same three reads we use in the published research, with a score for each. Then fix in order: make your core facts machine readable, check nothing is blocking AI systems at the door, and tidy the public record so every mention of your business agrees with your own site. None of this is glamorous. All of it is measurable, which means you can prove it moved.

Bristol has one of the strongest creative and digital communities in the country, and the businesses here that treat AI readability as a measurement problem rather than a mystery will take a lead that compounds. The buyers have already changed how they ask. The only question is whether the answer includes you.

Nic Crane is the founder of The Crane Consultancy, working with businesses across Bristol, Bath and the South West on AI search, paid media and technical SEO.