of UK startups are unreadable for the AI engines their buyers now ask.
370 UK post-funding B2B tech startups audited by Fast Growth Advisors with Positive Marketing. Two lenses: what prospects understand, what AI crawlers can read. Average message score 45.5/75.
September 2026
clarity score
AI readability
of the potential
Raising 50 million does not fix a fuzzy message.
370 UK post-funding startups audited. 68.1% are graded D or F on AI readability. Capital does not buy clarity. That is what this study measures.
12 figures to understand where UK B2B tech falls behind.
370 UK post-funding startups, two audit instruments, one September 2026 wave. What no one had measured until now.
Raising does not move the message forward proportionally.
146 UK startups with a disclosed amount. On the X axis, the amount raised on a log scale. On Y, the message clarity score. Spearman = -0.073. Raising 5 or 50 million makes no predictable difference.
Read the full report below, or download the PDF edition of The State of Message-Market Fit: UK B2B Tech 2026 (2 MB), written by Fast Growth Advisors with Positive Marketing.
What does the Message-Market Fit study measure?
Start with the definition. Message-Market Fit is the degree to which a company's message, what it says it does, for whom, and with what proof, is immediately understood by its market. Any startup that wants to sell, hire or raise should be able to answer four questions in under a minute: who we are, what we sell, what sets us apart from a competitor, and what concrete value we bring to the customer. If a single answer is missing from the website, the market tunes out.
In 2026, before a prospect ever lands on a homepage, an AI engine has often already read it, summarised it, and decided whether to recommend it.
Google AI Overviews, ChatGPT, Claude, Perplexity: the first reader of a startup's message is now, more often than not, a machine reading on a human's behalf.
So this study measures two things. First, whether prospects understand the message: an audit of 15 criteria, scored out of 75. Second, whether AI engines can read and restate that same message: a five-layer GEO audit scored out of 100, plus visibility measured directly on Google AI Overviews.
Two lenses, one object.
Everything is measured from 100% public data (website, LinkedIn, press, funding announcements), the way a prospect would before a first meeting, and the way a crawler does every day. Nothing a company says about itself in private enters the calculation.
Why do three in five startups struggle to be understood?
Across 370 UK post-fundraising B2B tech startups audited in September 2026, 61.4% score below 50 out of 75 on message clarity. These companies have convinced demanding investors and hired solid teams: the product holds up. What breaks is the translation, turning a real technology into a promise a buyer grasps in ten seconds.
The second lens is harsher still: 68.1% of these startups are graded D or F on AI readability.
A good product that tells its story badly loses customers, lengthens its sales cycles, and often weakens the next round. Now it also loses the recommendation of the machines that brief its buyers. The pages that follow quantify both gaps, sector by sector, and show where they form.
One report, six chapters, one question.
Why messaging is the most under-invested variable in UK B2B tech growth, and what it costs a post-fundraising startup when neither prospects nor AI engines can say what it does.
- 00Executive summary90 sec
- 01The POV: the first 45 seconds5 min
- 02The invisible cost of a fuzzy message4 min
- 03The data: 370 startups, two lenses12 min
- 04The 4 patterns of a failed message9 min
- 05Sector-by-sector X-ray6 min
- 06The 5% that get it right5 min
- 07The startup to-do list4 min
- 083 predictions for the next 12-18 months5 min
- 09Three anonymised portraits6 min
- 10FAQ7 min
- 11Appendices & methodology8 min
For readers in a hurry.
This summary distils the 12 key figures and 5 takeaways from The State of Message-Market Fit: UK B2B Tech 2026. Reading time: 90 seconds.
The 12 figures to remember
The 5 main takeaways
For those short on time, here is the essence.
- 01 Nobody reaches 85% of the potential. The best startup in the corpus scores 64/75, or 85.3% of what the grid allows, and it is alone up there. The average sits at 45.5. Even the strongest UK messages leave measurable clarity on the table.
- 02 Raising funds does not clarify the message. The Spearman correlation between amount raised and message score is -0.073, on the 146 startups with a disclosed amount. Capital buys engineers and pipeline, not narrative. A Series B with a generic message stays a Series B with a generic message.
- 03 Classic SEO is a solved problem. AI readability is not. SEO is the strongest of the 15 criteria at 4.16/5, while GEO is among the weakest at 2.56/5, and 68.1% of startups are graded D or F on the dedicated AI audit. UK startups have optimised for the search engine of 2015, not the answer engine of 2026.
- 04 Conversion is the weakest link in the chain. Lead generation scores 1.76/5 and pricing strategy 1.94/5, the two lowest criteria in the audit. Startups know how to describe themselves and differentiate. They do not know how to turn a convinced visitor into a conversation.
- 05 A good message and AI readability go together. The Spearman correlation between message score and GEO score is 0.547, and the top 5% of the corpus averages 70.1/100 on GEO against 51.2 for the whole population. Write clearly for humans, structure honestly for machines, and each lens reinforces the other.
Four questions. Forty-five seconds. It all plays out here.
Before the data, one exercise. Four questions, forty-five seconds, and two readers in mind: the human prospect, and the AI engine now reading on their behalf.
For twenty-five years, the discipline behind this test has been called the POV, Point of View. The equivalent of the elevator pitch, but made operational: not a slogan, not a marketing hook, a testable formulation. A founder who cannot answer these questions out loud, without hesitation, in front of a prospect, an investor, or a future colleague, has a Message-Market Fit problem. However good the product behind it.
The technique was formalised by Christopher Lochhead (former CMO of Mercury Interactive), with Dave Peterson and Al Ramadan, in Play Bigger (2016), one of the reference works on category design in B2B.
Different is better than better.Al Ramadan, Dave Peterson, Christopher Lochhead · Play Bigger (2016)
The data says UK startups fail this test exactly where it costs most.
Across the 370 companies audited, lead generation scores 1.76 out of 5, the lowest mark of the fifteen criteria, and pricing strategy sits at 1.94. The conversion path averages 2.25. AI readability, the geo criterion, 2.56.
Four weak points. The same four questions.
A fourth, often underestimated reader: the analyst ecosystem. Gartner, Forrester, IDC and their vertical equivalents spend their days briefing enterprise buyers on the shortlists to consider in each category. When a CIO or a VP Marketing prepares a major purchase, their first reflex is rarely a search engine: it is a call to their go-to analyst. If the startup has not formulated a legible POV, the analyst literally cannot position it. Not in a public report, not in an inquiry response, not in an annual briefing.
And in 2026, a fifth reader has joined the queue: the AI engine.
It summarises your category before the buyer ever reaches your site. Same test, stricter format: what is not written in plain HTML text does not exist for it.
Working glossary
Six terms used throughout the report. Defined once, here.
- Message-Market Fit
- A startup’s ability to make its market understand, in under 10 seconds, what it does, for whom, and why to choose it over a competitor. Measured here across 15 criteria, scored out of 75.
- GEO (Generative Engine Optimisation)
- The discipline of making a website readable, quotable and correctly restated by AI engines such as ChatGPT, Perplexity and Google AI Overviews. Audited here across 5 layers, scored out of 100.
- AI crawler
- The robot an AI engine sends to fetch your pages (GPTBot, ClaudeBot, PerplexityBot). Most of them do not execute JavaScript: they read the raw HTML, nothing else.
- llms.txt
- A plain-text file at the root of a website that tells AI engines what the company does and where its key pages sit. In this study, 33.5% of sites have one.
- AI citability
- The likelihood that an AI engine cites or recommends the company when answering a buyer’s question in its market. Measured here on Google AI Overviews, from none to strong.
- The two lenses
- The reading grid of this report. Lens one: is the message understood by prospects (score out of 75)? Lens two: is it readable and correctly restated by AI engines (GEO score out of 100)?
Acronyms used
- POV (Point of View)
- the position a company defends, the claim a competitor cannot take over.
- ICP (ideal customer profile)
- the target customer described by observable criteria rather than by an industry.
- ARR (annual recurring revenue)
- the yearly recurring revenue, the common basis for comparing subscription startups.
- VC (venture capital)
- venture capital, and by extension the funds that deploy it.
- HR (human resources)
- the function, named here as the buyer of a solution.
- IDC (International Data Corporation)
- a technology market analyst firm, alongside Gartner and Forrester.
- AIO (AI Overviews)
- the AI-generated answers Google places above its search results, the engine measured in this study.
- WAF (web application firewall)
- the security layer that filters a website’s traffic, and sometimes challenges AI crawlers along the way.
- NPS (net promoter score)
- a customer recommendation indicator, measured on a zero to ten scale.
- CAC (customer acquisition cost)
- what it costs to win one new customer, marketing and sales included.
- HQ (headquarters)
- a group’s head office, as opposed to its country subsidiaries.
The invisible cost of a fuzzy message.
The cost of a fuzzy message shows up on no dashboard. In 2026 it is paid twice: the prospect does not understand, and the machine reading on their behalf does not understand either.
B2B buyers now ask ChatGPT or Google AI before they visit a website. Whatever the engine restates, or fails to restate, becomes the first version of your message a buyer ever meets.
You no longer control the first impression. The machine does.
We measured it. Of the 370 UK B2B tech startups audited, 40.3% have no AI citability at all: asked standard buyer questions in their market, the engine never cites them. And 38.6% appear in no Google AI Overviews answer on their own market queries. They simply do not exist in the response.
The mechanism sits in the audit’s weakest layer. Content clarity, as an AI engine extracts it, averages 2.7 out of 20. The message may exist on the page; once scripts, carousels and widgets are stripped away, almost none of it survives extraction.
Your website is still your top salesperson. It now briefs the machines too.
It is read before the first call, checked between two meetings by the buying committee, and summarised by AI engines for buyers who will never scroll your homepage. If neither the human nor the machine gets your message in ten seconds, the deal is compromised before it began.
We analysed 370 UK B2B tech startups, audited in September 2026. What we found exceeds our initial assumptions.
"The AI engine reads your website before your prospect does."
This report exists because the way B2B buyers read has changed faster than the way startups write.
In 2026, a buyer’s first question rarely lands on your homepage. It lands in ChatGPT, in Perplexity, in Google AI Overviews. The engine fetches your pages, extracts what it can, and answers on your behalf. Your message now has two readers: the prospect, and the machine reading for them.
The machine came first. That is the shift this UK edition measures.
We audited 370 UK B2B tech startups through both lenses: is the message understood by prospects, and is it readable and correctly restated by AI engines? The two scores tell one story, and we tell it here without complacency, because the data leaves little room for it.
Why this partnership? Fast Growth Advisors brings the method: fifteen message criteria, five layers of AI readability, visibility measured on real market queries. Positive Marketing, AI-powered B2B tech PR and marketing experts based in London since 2009, brings what no audit can: the daily practice of making UK tech companies heard, by journalists, by analysts, and now by the engines.
For founders, one piece of good news survives the diagnosis. Your message is a variable you control, and you can fix it without raising another pound.
The data. What no one had measured in the UK.
Methodology, score distribution, the funding question, the fifteen criteria, machine readability, measured visibility in Google AI Overviews. All of it, verified at the source.
Two instruments, one object.
The Message-Market Fit framework sets out the what: a defensible position, a testable value proposition, a repeatable sales narrative. What remains is the how.
This study reads each of the 370 startups through two instruments, both built by Fast Growth Advisors and both pointed at the same object: the message.
- The prospect lens: a detailed audit of 15 operational criteria, each scored out of 5, 75 points in total. It asks whether a buyer landing on the website understands who this is for, what it does, why it wins, and what to do next.
- The machine lens: a GEO audit across 5 layers scored out of 100 (schema.org, semantic HTML, content clarity, crawl directives, agent comprehension), completed by measured visibility in Google AI Overviews on three market queries per startup.
Two lenses. One object.
The first asks whether a human prospect understands the message. The second asks whether an AI engine can fetch it, extract it and repeat it. A startup can pass one and fail the other. Many do, and the gap between the two is where this report earns its keep.
Corpus & volumes
| Analysis | Volume |
|---|---|
| UK startups audited, 15 criteria, score out of 75 | 370 |
| Individual criterion scores recorded | 5,550 |
| GEO audits, 5 layers, score out of 100 | 370 |
| Google AI Overviews visibility checks | 370 |
| Startups with a public raise amount | 146 |
| Sector buckets analysed | 11 |
| Top 5% cohort | 18 |
No startup exceeds 85% of the potential.
Across the 370 startups audited, the average message score sits at 45.5/75, 60.7% of the potential. The median is 47. The standard deviation, 8.72, says the pack is tight: most of UK B2B tech is clustered around a message that half works.
The best score in the corpus is 64/75. That is 85.3% of the maximum. Nobody goes higher.
61.4% of startups score below 50/75. 22.2% fall below 40, the zone where the message stops functioning as a sales tool. At the other end, 10.5% reach 55 or more, and only 7 startups clear 60.
| Band /75 | Startups | Interpretation |
|---|---|---|
| 60-64 | 7 | Excellent: defensible differentiation, legible proof |
| 55-59 | 32 | Good: a few conversion gaps |
| 50-54 | 104 | Fair: functional message, little pull |
| 45-49 | 89 | Fair: clarity without conversion |
| 40-44 | 56 | Insufficient: unclear on target or value |
| 35-39 | 38 | Weak: generic or brochure message |
| 30-34 | 24 | Weak: positioning not legible |
| 25-29 | 13 | Critical: rework required |
| 20-24 | 3 | Critical: rework required |
| 15-19 | 2 | Critical: rework required |
| 10-14 | 2 | Critical: rework required |
Read the ceiling, not just the floor.
A market where the best available message stops at 85.3% of the potential is a market where clarity is still a vacant advantage. The first movers will not need to be perfect. Just legible.
Raising does not clarify the message.
One limit first, stated in full: a public raise amount exists for 146 of the 370 startups, 39.5% of the corpus. Every figure in this block covers those 146 companies and no one else. Treat it as a strong signal on a partial sample, not a census.
On that sample, the Spearman correlation between the amount raised and the message score is -0.073. Statistically, nothing.
The median raise is £18.65M, and knowing whether a startup raised more or less than that tells you almost nothing about whether its message works.
Money buys engineers, offices and ad spend. It does not buy a sentence that sells.
Average score by funding series
| Series | n | Score /75 |
|---|---|---|
| Seed | 18 | 45.7 |
| Series A | 46 | 47.0 |
| Series B | 23 | 45.8 |
| Series C | 14 | 46.6 |
Flat. From Seed (45.7) to Series C (46.6), the line barely moves.
Series B startups average 45.8, marginally above the Seed cohort they left two rounds and tens of millions of pounds ago. Whatever a growth round professionalises, the public-facing message is not reliably on the list.
So what does move the score?
| Commercial stage | n | Score /75 |
|---|---|---|
| Early-stage | 30 | 35.3 |
| Emerging | 115 | 43.8 |
| Growing | 175 | 47.7 |
| Mature | 49 | 48.5 |
Commercial maturity, and here the gradient is real: 35.3 for early-stage startups, 43.8 for emerging ones, 47.7 for growing ones, 48.5 for mature ones. Startups that sell every day sharpen their message because prospects force them to.
Commercial traction follows the message. The raise does not.
Where UK B2B tech falls behind.
The 15 criteria of the detailed audit are not equal. Some are broadly under control, others fail almost everywhere. The full ranking, weakest first, across 370 audits:
The 15 criteria, in figures
| Criterion | Average /5 |
|---|---|
| Lead generation | 1.76 |
| Pricing strategy | 1.94 |
| Conversion path | 2.25 |
| AI readability (GEO) | 2.56 |
| Content marketing | 2.75 |
| Proof points | 3.07 |
| Calls to action | 3.19 |
| ICP clarity | 3.21 |
| Target audience | 3.25 |
| Consistency | 3.34 |
| Value proposition | 3.35 |
| Trust & social proof | 3.45 |
| Clarity | 3.46 |
| Differentiation | 3.75 |
| SEO | 4.16 |
Start at the strong end. SEO averages 4.16/5, with 81.9% of startups scoring 4 or more. Differentiation follows at 3.75 (70.5% at 4 or more), then clarity at 3.46. UK startups know how to rank and how to sound different.
The classic playbook is learnt. The next one is not.
Now the weak end.
Lead generation averages 1.76/5, and 72.2% of the corpus scores 2 or less on it. Pricing strategy sits at 1.94, the conversion path at 2.25, AI readability at 2.56 (44.9% at 2 or less). Everything that turns a visit into a conversation, or a crawl into a citation, is underbuilt.
The middle of the table tells the same story more quietly: proof points at 3.07 and content marketing at 2.75 are the substance criteria, and they trail the style criteria, value proposition (3.35), consistency (3.34), trust and social proof (3.45). These startups can describe themselves. They struggle to prove themselves, and they struggle to be chosen.
Where 370 startups land.
Same distribution, now as a picture.
Each bar is a 5-point band of the message score out of 75. The red zone on the left holds the 22.2% of startups below 40.
The centre of gravity is mediocrity, not disaster.
89 startups sit in the 45-49 band and 104 in the 50-54 band. This is the crowd: a message that names a category, lists features, shows a few logos, and gives a prospect no strong reason to choose it over the next tab. Below them, 56 startups at 40-44 and 38 at 35-39 slide toward the red zone. Above them, the air thins fast: 32 startups at 55-59, then 7 at 60-64, and nothing beyond.
The shape matters commercially. When most of a market sounds the same, the buyer, human or machine, defaults to whoever is easiest to summarise.
The five GEO layers: where machine readability breaks.
The GEO audit scores each site on five layers, each out of 20. Together they answer one question: when an AI engine fetches this website, what survives?
Content clarity: 2.7 out of 20. The most broken layer of the five.
It means the message exists but the machine cannot extract it. The promise sits in an animated hero, the proof in a JavaScript carousel, the pricing behind a toggle. A human visitor pieces it together. An extractor throws it away, and what reaches the AI engine is a page that says the company exists without saying why it matters.
The other layers frame the diagnosis. Crawl directives average 17.1/20: the doors are open, almost everyone lets the crawlers in. Semantic HTML reaches 13.4, agent comprehension 11.6, schema.org just 6.0. So the crawler gets in, parses the structure, and finds too little machine-legible substance to quote.
The doors are open. The shelves are half empty.
Two thirds of the market is unreadable for AI engines.
The five layers roll up into a grade per site. The distribution is blunt: 21 sites earn a B, 97 a C, 168 a D, 84 an F. Not a single A. 68.1% of UK B2B tech startups sit at D or F, the grades where an AI engine struggles to say what the company does, let alone why it wins.
| AI readability grade | Startups |
|---|---|
| B | 21 |
| C | 97 |
| D | 168 |
| F | 84 |
The hygiene markers
Three simple checks confirm the grade sheet. 43% of sites score 2 or less out of 20 on schema.org: the structured layer that tells a machine "this is the product, this is the price, this is the company" is missing for nearly half the market. An llms.txt file, the emerging convention for addressing AI crawlers directly, exists on 33.5% of sites. A robots.txt exists on 95.9%: basic hygiene is fine, the AI-specific layer is not.
Who is locking the door?
52 sites block at least one AI crawler. GPTBot is refused by 32 sites, ClaudeBot by 30, OAI-SearchBot by 18, PerplexityBot by 18, and Googlebot itself by 18.
Blocking Googlebot is rarely a strategy.
In many of these cases the culprit is a WAF or bot-protection layer serving challenge pages to every non-human visitor, the AI crawlers included. The startup never decided to be invisible. Its infrastructure decided for it, and nobody checked.
| Crawler blocked | Sites |
|---|---|
| GPTBot (OpenAI) | 32 |
| ClaudeBot (Anthropic) | 30 |
| OAI-SearchBot (OpenAI) | 18 |
| PerplexityBot | 18 |
| Googlebot | 18 |
What Google AI Overviews actually say.
Readability is a prediction. Visibility is a result.
Each of the 370 startups was probed on three market queries in Google AI Overviews, the questions its own buyers would ask, and the answers were scored for presence, recommendation and accuracy.
61.4% of startups appear at least once in an AI Overview across their three queries.
48.1% are actively recommended. Encouraging, until the citability column: for 40.3% of the corpus, 149 startups, citability is nil. The engine has nothing quotable to work with. 61 more are weak, 73 moderate, and only 87 reach strong citability.
| Citability | Startups |
|---|---|
| None | 149 |
| Weak | 61 |
| Moderate | 73 |
| Strong | 87 |
Then the brand test: does the AI describe the company the way the company describes itself? The verdict is accurate for 53.5% of startups, 198 out of 370. It is only partially accurate for 46.2%.
Nearly half the market is presented to buyers with a message it did not write.
The double penalty.
Cross the two lenses and the market splits into four.
Message score out of 75 on one axis, GEO score out of 100 on the other, cut at their medians: 47 and 53.
The two scores are related but far from redundant: the Spearman correlation between them is 0.547.
Clear writing helps machines, but it does not guarantee machine readability, and the quadrants prove it.
- Strong message, strong GEO: 132 startups (35.7%). Legible for humans and machines. The group the rest of the market will be compared against, by both.
- Strong message, weak GEO: 66 startups. The message works in the room and dies in the crawl. The cheapest problem to fix in this entire study.
- Weak message, strong GEO: 60 startups. Technically impeccable delivery of a message that does not persuade. Machines can quote them accurately, to no effect.
- Weak message, weak GEO: 112 startups (30.3%). Below both medians.
Inside that last quadrant sits the group this chapter has been building toward: 73 startups, 19.7% of the corpus, combine a message below the bar with zero citability in AI answers. Unconvincing for the humans who visit. Invisible to the machines that answer for everyone else.
That is the double penalty. One startup in five in UK B2B tech is currently losing the argument twice, in front of two different readers, for the same underlying reason: the message was never engineered to be understood.
The rest of this report is about the ones who escaped it, and how.
Anatomy of a failed message.
Four recurring patterns across UK B2B tech. Each one measured, each one independent. Their combination, which is common, is especially destructive for pipeline.
Four patterns, measured across 370 audits.
The patterns below are the most frequent failures of a broken Message-Market Fit in the UK corpus. For each, the figures come straight from the 15-criteria scoring, followed by the mechanism identified and anonymised examples drawn from the audits.
One of the four is new this year. It concerns what machines read, not what people read.
"The technology talks, the prospect does not understand."
The founder has mastered the science. They can hold a room of engineers or technical investors for 45 minutes. On the website, that mastery becomes specialist-to-specialist language: the buyer's business problem never appears.
The gap shows in the scoring.
Value propositions average a respectable 3.35/5. Lead generation averages 1.76/5, the weakest of the 15 criteria. The message exists on paper. It just never reaches the person who signs.
Typical examples (anonymised): a biomanufacturing startup whose homepage headline reads "Biomanufacturing the future." A prospect cannot tell in five seconds what it sells, to whom, or what problem it solves. The real B2B offer, a premium ingredient for meat producers, sits buried on internal pages. An enzyme technology company speaks the language of enzyme science throughout, with no clear, immediate benefit offered to a buyer. Scores: 25/75 and 20/75.
"The website says it, the machine cannot read it."
This is the pattern the UK edition puts front and centre, because it is new and because almost nobody is watching it. The message is on the page. A human sees it. The AI engines that increasingly brief buyers do not.
The mechanism: content injected client-side via JavaScript.
Most AI crawlers do not execute JavaScript, so everything the script draws is invisible to them. Across the 370 sites audited, average content clarity as measured on the machine-readable layer is 2.7 out of 20, the lowest of the five GEO sub-scores, and 68.1% of startups are graded D or F on AI readability overall.
Typical examples (anonymised): a recruitment platform serving 94% of its homepage client-side. Crawlers get a near-empty page whose only server-rendered signal is a warm but vague tagline. A robotics company injects 92% of its content the same way, making a genuinely clear positioning effectively invisible to OpenAI, Anthropic and Perplexity. Scores: 16/75 and 23/75.
"Proof exists, but not on the page."
The UK corpus is polarised on trust. 52.2% of startups score 4/5 or above on social proof. And 23.2% sit at 2/5 or below, with almost nothing shown at all.
What makes the low end frustrating is that the proof usually exists. Real clients, real partnerships, real review scores. They live in press releases, on review platforms, in funding announcements. The homepage shows none of it: no logos, no case studies, no quantified results.
Typical examples (anonymised): an enzyme technology company with a genuine partnership with a global chemical group, never mentioned on its own site. A recruitment platform holding a 4.7/5 rating on a major review site, invisible on its homepage. The buyer who checks only the website concludes the proof does not exist.
"No path from interest to conversation."
A prospect arrives, understands, believes.
Then looks for the next step and finds nothing to hold on to. No pricing, no demo booking, no form. Sometimes just a raw email address at the bottom of the page.
- • Lead generation scored 2/5 or below: 72.2% of sites
- • Conversion path scored 2/5 or below: 54.9% of sites
- • Pricing strategy: 1.94/5 on average
- • Conversion path: 2.25/5 on average
Typical examples (anonymised): a robotics company whose only conversion mechanics are a contact email and a chatbot. No pricing, no demo CTA, no form. An enzyme technology company in the same situation, raw email addresses only. Both sit near the bottom of the ranking, at 18/75 and 20/75.
When the logo disappears, what do we hear?
A simple exercise: take the first thing each homepage shows, remove the company name, and ask what business this is.
The four openings below are real, drawn verbatim from the UK audits. All four companies raised institutional money.
Four real openings, four audited UK homepages, September 2026
Biomanufacturing the future.
A more human job search
Privacy Policy
Account Suspended
The first two are polished.
They are also interchangeable: no buyer, no outcome, no proof. The last two are not messaging failures so much as abandonment, and the audit treats them the same way a prospect does. Across the corpus, the weak openings follow one implicit template.
+ [ a category word (platform / solutions / technology) ]
+ [ no buyer, no number, no name ]
→ Zero outcome. Zero proof. Zero perceived difference.
The language of the best vs the language of the worst.
The audits record what each homepage says first. Read side by side, the 39 startups scoring 55/75 or above and the 82 scoring under 40 speak two different languages.
The best name three things in their opening lines: the buyer, the job, the proof. A workforce platform talks about frontline teams. A construction tool talks about site progress. A payments company talks about selling software globally.
The worst name none of them. They describe their own technology, their own ambition, their own category. The buyer has to translate.
The best · the words of the buyer
The worst · the words of abstraction
The best startups talk about the outcome for the prospect. The worst talk about their own work. This shift in point of view, from "I" to "you", from process to benefit, is one of the fastest transformation levers identified in the field.Fast Growth Advisors × Positive Marketing · UK lexical review, September 2026
Sector-by-sector X-ray.
Every sector has its pathology. PropTech and HR lead. AI & Deep Tech, the most crowded group in the corpus, sits below average on both lenses.
Eleven sectors, two lenses.
First, the operational sectors lead. PropTech & Construction posts the best average message score of the eleven groups, 50.8/75, with a GEO score of 54.0/100. HR & Future of Work follows at 49.6/75 and takes the top spot on machine readability, 55.8/100. Both sell to buyers who want a working answer to a concrete operational problem, and it shows in the copy.
Neither is glamorous. Both are legible.
Next, the paradox. AI & Deep Tech is the most crowded group of the corpus, 85 startups, and one of the weakest: 43.7/75 on message, 47.3/100 on GEO, below the average on both lenses. The deeper the technology, the more the message gets lost inside it.
Health & Bio (42.1/75) and Supply Chain & Mobility (41.0/75) confirm the pattern: the harder the science or the operations, the softer the pitch.
Three sector reads to remember
Why PropTech and HR do better
Their buyers are operators: site managers, HR directors, people who live with the problem daily and penalise vagueness on sight. That buyer pressure forces these startups to name the use case, the user and the outcome earlier than others.
The AI paradox: the cobbler's children go barefoot
AI & Deep Tech startups build the engines that now read everyone else's websites. Their own sites average 47.3/100 on AI readability, under the 51.2 corpus mean, and 43.7/75 on message. The symptom is a headline that describes a technology instead of naming a buyer and a result. Descriptive. Categorical. Undifferentiated.
The engines they ship can read anything. Except, too often, their makers.
Fintech: readable by machines, average for humans
Fintech scores 55.4/100 on GEO, near the top of the table. On message it sits at 46.5/75, barely above the corpus average of 45.5. The technical plumbing is right (structured data, clean HTML, open crawl paths), the argument on top of it stays ordinary. A machine can restate what these companies do. A prospect still struggles to say why this one rather than the next.
Amount raised vs narrative clarity.
A reasonable hypothesis: better funded startups afford better marketing, so their message should be clearer. The UK data does not support it.
One caveat before the number, stated in full: we could match a disclosed amount for 146 of the 370 startups, a coverage of 39.5%. Read what follows within that limit.
Median raise in that subset: £18.65M. Correlation between amount raised and message score: Spearman -0.073, n=146.
Statistically, nothing. Raising more does not buy a clearer sentence.
Money buys engineers, offices and paid campaigns. It does not buy the decision about what to say, to whom, in what order. On this data, that decision is as rare after a £30M round as before it.
The 5% that get it right.
Four shared decisions. And one trait the rest lack: the best win on both lenses at once.
Eighteen startups, the top 5% of the 370 audited, score 57/75 or higher. Their average GEO score is 70.1/100, against 51.2 for the corpus as a whole. The clearest messages for humans are also the most readable for machines.
Three of them are already public in this study and can be named. Synthesia leads at 64/75, with a GEO score of 73/100. Paddle follows at 63/75 and 78/100, the best AI readability of the top group. Sona posts 62/75.
Different sectors: AI video, payments, workforce software. Same four decisions.
Name your target with a number, not a sector.
The startups above 57/75 rarely open with "companies" or "your teams". They open with a named role and a count: how many customers of exactly that kind already rely on them. A finance lead, an operations manager, a frontline rota planner. The visitor knows within three seconds whether the page is for them.
One figure does two jobs at once: it names the target precisely (a role, not a company size) and it proves traction in the same breath. It is the most economical formulation in B2B messaging.
Assert a differentiator your competitors cannot copy.
Among the top 18, the differentiator is structural, not tonal.
A payments company that takes on global tax liability as merchant of record. A video platform whose avatars come from its own research lab. A rota tool built for frontline shifts rather than desk calendars. Each claim commits to an architecture or a business model.
Terms like these are not jargon to sound clever. They are proof of positioning: a rival cannot borrow the sentence without borrowing the product decision behind it.
Replace your adjectives with numbers.
"Leading", "innovative", "trusted": adjectives cost nothing to write, so they are worth nothing to the prospect. The top 5% put counts where the rest put superlatives. Customers served, stated as a figure. Review scores with the platform named. Hours saved per week, in hours.
Concrete beats confident. Every time.
Create your category, don't join a saturated one.
The rarest and most powerful move: not positioning inside an existing shelf, but naming a new one. The best UK performers sell an AI video studio rather than another video tool, a merchant of record rather than another payment gateway, frontline workforce management rather than generic HR software.
A category you invented has one occupant. Comparison ends before it starts, and both lenses reward it: prospects retain the label, AI engines restate it verbatim.
The to-do list for UK startups.
The data ranks the work by itself. Machine readability first, conversion second, proof in HTML third. And the first is measured in days, not months.
Three priorities, ranked by the data.
| # | Weak point measured across 370 sites | Result |
|---|---|---|
| 01 | Content clarity for AI engines (GEO layer) | 2.7/20 |
| 02 | Schema.org structured data (GEO layer) | 6.0/20 |
| 03 | Sites with an llms.txt file | 33.5% |
| 04 | Sites blocking at least one AI crawler | 52 |
| 05 | Lead generation (score /5) | 1.76 |
| 06 | Pricing strategy (score /5) | 1.94 |
| 07 | Conversion path (score /5) | 2.25 |
| 08 | Proof points (score /5) | 3.07 |
Priority 1: make the site readable for AI engines
The weakest scores of the whole audit sit on the machine side. Content clarity for AI engines averages 2.7/20. Schema.org markup averages 6.0/20. Only 33.5% of sites publish an llms.txt file. And 52 sites block at least one AI crawler, often without knowing it, because the WAF challenges the bot before the page ever loads.
These are configuration fixes, not strategy projects. Publish the structured data, add the llms.txt, check what the WAF actually serves to GPTBot and ClaudeBot, put the pitch in server-rendered HTML. Days of work, not months.
Few line items in marketing cost so little and gate so much.
Priority 2: repair the conversion machinery
The three weakest of the 15 message criteria tell one story. Lead generation averages 1.76/5, with 72.2% of startups at 2 or below. Pricing strategy averages 1.94/5. Conversion path averages 2.25/5. A prospect who understood the pitch still cannot act on it.
The fix list is unglamorous: a pricing page that says something, a call to action per page, a demo path that takes minutes rather than a discovery call to book. Editorial and product-marketing work on the existing site.
Priority 3: put the proof in HTML
Proof points average 3.07/5.
Passable for a human skimming the page. Weak for a crawler, because so much of the proof lives in logo carousels, review widgets and animated counters that never reach the extracted text. A claim without proof in plain HTML is, for an AI engine, a claim without proof.
Numbers in text. Client names in text. Results in text. The rest is decoration.
Where the majority sits.
The striking finding: the majority is not failing, it is stalling. 193 of the 370 startups score between 45 and 54 out of 75. Above the broken messages. Below the 57/75 threshold where the top 5% starts, where targets carry numbers and proof lives in the HTML.
For that core, the gap to the next tier is not a rebrand, it is one targeted piece of work from the list above: the machine readability sprint for most, the pricing page and demo path for the rest.
The to-do list is short. The queue at the top is not.
What changes in 12-18 months.
Three converging forces are redrawing the rules: the AI-assisted buyer, the standardisation of GEO, and a widening gap between startups machines can read and startups they cannot.
The B2B buyer delegates the first read to AI engines.
Google AI Overviews, ChatGPT, Perplexity. The first pass over your market is increasingly done by an engine, not a person. The buyer arrives later, with a shortlist already formed.
This study measured what that first read returns. 61.4% of the 370 startups appear at least once in Google AI Overviews. 48.1% are recommended.
And 40.3% have zero citability: nothing in their public presence gives an engine a reason, or the raw material, to quote them.
Those startups are not weaker companies. They are invisible at the exact moment the shortlist forms. In 12-18 months, that moment is where UK B2B deals start.
GEO becomes a standard discipline of UK B2B marketing.
The data shows a solved problem sitting next to an unsolved one. Classic SEO is the strongest criterion in the whole audit: 4.16/5, with 81.9% of startups scoring 4 or more. Machine readability, the GEO criterion, averages 2.56/5.
The same thing happened to SEO fifteen years ago. A niche obsession became a budget line, a hire, a dashboard.
Expect GEO to follow that path, faster. The playbook is already legible: schema.org, llms.txt, server-rendered proof, content written to be extracted. Teams that treat it as a standard discipline now will be measured against a population where 68.1% still grade D or F.
The two quadrants drift apart.
Split the 370 startups on both medians, message score and GEO score, and two corners dominate. 35.7% are strong on both lenses. 30.3% are weak on both.
The strong corner compounds: a clear message that machines can also read gets extracted, quoted, recommended. The AI engine treats each query as a question asked by the buyer, and it answers with the startups it can actually parse.
The weak corner pays for its pipeline the expensive way: outbound, paid acquisition, events, because the recommendation channel does not carry them.
No identifiable question answered. No machine-readable proof. No citation.
The engine can extract the claim and repeat it. So it does.
Startups that fix both lenses today are not just improving 2026 conversion. They are choosing which corner they occupy in 2028.
Three anonymised portraits.
To make concrete what the statistics describe. No company is named and some details have been blurred. The patterns stay faithful to the cases audited in September 2026.
The startup whose product is strong and whose website is a challenge page.
The company is real, funded, and its homepage does its job. A visitor sees a named buyer, a concrete promise, credible proof. The human lens has little to complain about.
Now replay the visit as a machine. OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot: every one of them receives a verification challenge page instead of the content. The server even answers HTTP 200, so nothing looks broken in the logs.
What the prospect sees: a polished, convincing site.
What the machine sees: a security riddle it will never solve. It moves on and cites a competitor.
The lesson: nobody decided this. No marketer chose to hide the company from AI engines. A web application firewall shipped with a default, and the default quietly became the positioning. This study found 52 sites blocking at least one AI crawler, several of them through exactly this kind of challenge page.
The site that speaks to scientists, not to buyers.
The homepage headline is an abstract manifesto about building the future of its field. Elegant. Buyer-agnostic. A prospect cannot tell in five seconds what the company sells, to whom, or against what problem.
The actual B2B value proposition exists. It is buried on internal pages, in the vocabulary of the founders' scientific discipline, never surfaced on the hero.
What the prospect sees: an impressive research organisation, with no obvious way to become a customer. No pricing signal, no demo path, no case study, raw email addresses as the only door.
What the machine sees: the same gap, amplified.
An engine asked "who solves this problem" finds no sentence on the site that answers it, so the company never enters the answer.
The lesson: the site was written for peers, partners and grant committees, not for the person holding the budget. It is the most common pattern in the study: differentiation scores high (3.75/5 on average), conversion mechanics score lowest (lead generation at 1.76/5). Knowing what makes you different is not the same as giving a buyer a reason, and a way, to act.
The strong message that JavaScript hides.
This portrait is the cruellest of the three.
The messaging work has been done. Clear promise, named audience, quantified proof: the audit scores the message above 50/75, comfortably in the top half of the population.
The problem is the delivery vehicle.
Between 40 and 50% of the page content is injected client-side by JavaScript. Crawlers from OpenAI, Anthropic and Perplexity do not execute JavaScript. For them, half the site does not exist.
What the prospect sees: everything. The browser renders the full experience, and nothing ever looks wrong.
What the machine sees: a skeleton. Often the navigation, a heading, a footer. The proof points, the customer logos, the pricing logic, the differentiators: all injected, all absent from the bytes a crawler receives.
The lesson: message quality and machine readability are two different investments, and the second can silently cancel the first. The two scores correlate (Spearman 0.547) but they are not the same measurement. The fix is not a rewrite. It is a rendering decision: put the sentences that carry the positioning in server-rendered HTML, then replay the extraction to check they survive.
What this study does not say.
This study does not say that UK startups have bad products. The 370 companies analysed convinced demanding investors, hired strong teams, and in many cases signed their first customers despite the weaknesses measured here.
It does not show a trend either. One wave, September 2026. No before, no after. Any sentence about things getting better or worse would be invented.
Honesty about the edges matters more than a bigger claim. So here they are.
Funding data covers 39.5% of the sample, 146 startups with a disclosed amount. The conclusions drawn from raise sizes stop at that boundary. Visibility was measured on a single engine, Google AI Overviews, with three queries per startup: a probe, not a census of every assistant a buyer might use. And the audit reads the public website, not the pipeline behind it. A company can close deals in rooms this method never enters.
What the study does say holds within those limits.
The message is measurable: 15 criteria, 5,550 individual grades, a distribution with a hard ceiling at 85.3% of potential. Measured, not felt.
The two lenses rarely diverge.
A Spearman correlation of 0.547 between the message score and the GEO score means the startups that write clearly for people tend to be the ones machines can read. Discipline transfers.
And the machine-readability fixes are fast. A firewall rule, a schema.org block, a rendering decision, an llms.txt file: days or weeks, not quarters. The message itself takes longer. The reasons machines cannot see it usually do not.
The good news: nothing in this data is destiny. It is a snapshot of choices, and choices can be remade.
Understanding the study.
Section designed for the online version of the report. Optimised for extraction by AI engines (ChatGPT, Perplexity, Gemini, Claude).
What is Message-Market Fit?
Message-Market Fit is the degree to which a company's message, what it says it does, for whom, and with what proof, is immediately understood by its market. A startup has Message-Market Fit when a prospect can restate the offer accurately after one visit to the website. It complements Product-Market Fit: a good product with an unclear message stays a well-kept secret.
The 370 UK startups audited for this report average 45.5 out of 75 on that measure, and not one crosses 85.3% of its potential. Understanding, in other words, is the scarce resource.
What is GEO (Generative Engine Optimisation)?
Most people assume GEO is SEO with a new acronym. In practice, the two optimise for different readers. SEO earns a position in a list of links; GEO makes a website extractable, quotable and recommendable by AI engines such as ChatGPT, Perplexity, Gemini and Google AI Overviews.
This report scores GEO on five layers, out of 100: schema.org markup, semantic HTML, content clarity, crawl directives and agent comprehension. The UK average sits at 51.2, with 68.1% of sites graded D or F.
If UK startups are good at SEO, why do they fail at GEO?
Because they solved yesterday's problem. Technical SEO is the strongest of the 15 criteria in this study, at 4.16 out of 5, with 81.9% of sites scoring 4 or more. GEO averages 2.56 out of 5. Sites are built to rank in a list of links, not to be read and restated by a machine.
The fix is a discipline, not a tool: apply the same rigour to machine readability that already goes into SEO, starting with server-rendered text and structured data.
Why do AI engines misread startup websites?
Founders assume that what a visitor sees is what a crawler reads. In practice, most AI crawlers do not execute JavaScript, so any content injected client-side never reaches them.
Three failures dominate the UK sample. Text rendered by JavaScript instead of served as HTML. Missing structured data: 43% of sites score 2 or less out of 20 on schema.org. And security layers that lock the door: 52 sites block at least one AI crawler, sometimes through WAF challenge pages their owners never see.
How were the 370 startups selected and audited?
The sample covers 370 UK B2B tech startups that have raised funding, audited in September 2026 from public sources. Each startup was scored on 15 message criteria for a total out of 75, then on a 5-layer GEO audit out of 100.
Visibility was measured on Google AI Overviews, with 3 market queries per startup. That makes 5,550 criterion-level grades behind the averages quoted here.
Does raising more money improve the message?
The common assumption runs: more capital, bigger marketing team, clearer message. In practice, the data is flat. Across the 146 startups with a disclosed amount (39.5% of the sample), the Spearman correlation between money raised and message score is -0.073.
Funding buys many things. Clarity is not one of them.
What is llms.txt and does it matter?
llms.txt is a plain-text file at the root of a website that tells AI engines what the company does and where its key pages live, a younger cousin of robots.txt. Only 33.5% of the UK sample has one, so most sites leave the machine to guess.
Our recommendation: adopt it, it takes an afternoon, but only after the essentials. Server-rendered text and schema.org markup carry more weight.
Who is behind this study?
Fast Growth Advisors, the Message-Market Fit firm, designed the methodology and conducted the 370 audits. Positive Marketing, a B2B tech PR and marketing agency based in London since 2009, is the UK partner of this edition.
The report is written by Hervé Dhélin (Fast Growth Advisors), with Positive Marketing (London).
How do I get my own score?
A national average tells you where the market stands, not where you stand. Fast Growth Advisors runs individual Message-Market Fit diagnostics on the same 15-criteria grid used in this report, with a prioritised action plan attached.
Contact: herve@fast-growth.fr · Site: fast-growth.fr.
Methodology, limits, sources.
A1 · The 15 criteria of the detailed audit
Each criterion is scored out of 5 points. Total score out of 75.
| # | Criterion | What is measured |
|---|---|---|
| 01 | Message clarity | Immediate, effortless understanding of the homepage |
| 02 | Value proposition | Clear, concrete, non-generic prospect benefit |
| 03 | Target audience | Target named explicitly (persona, sector, size) |
| 04 | ICP clarity | Ideal Customer Profile visible, readable segmentation |
| 05 | Differentiation | Defensible differentiator, not copyable by a competitor |
| 06 | Cross-channel consistency | Alignment across website / LinkedIn / email / deck |
| 07 | Proof points | Figures, client results, concrete evidence on the page |
| 08 | Trust & social proof | External reputation: logos, testimonials, G2, Capterra, press |
| 09 | Calls to action | Clear CTAs, differentiated by persona |
| 10 | Conversion path | Clear path from discovery to getting in touch |
| 11 | Lead generation | Forms, gated content, quizzes, calculators |
| 12 | Pricing strategy | Pricing visible or a range communicated |
| 13 | Content marketing | Blog, resources, active thought leadership |
| 14 | Technical SEO | Meta, Schema.org, sitemap, Core Web Vitals |
| 15 | GEO | llms.txt, structured FAQ, AI-extractable content, crawler access |
A2 · Methodological limits
Single wave. Every startup was audited once, in September 2026. This report contains no trend claims, and none should be read into it.
Funding coverage. Disclosed raise amounts exist for 146 of the 370 startups, 39.5% of the sample. Every funding statistic in this report carries that caveat.
One engine measured. Visibility was tested on Google AI Overviews only, with 3 market queries per startup. Other assistants (ChatGPT, Perplexity, Claude) were not measured and may behave differently.
Public website only. The audit captures what is visible online at the date of analysis. Sales conversations, decks and product experience sit outside its reach, and a startup may have rebuilt its site since.
Grid version 2. The 15-criteria grid used here is the v2 instrument. Its scores are not comparable with scores produced by other grids or earlier editions.
A3 · External sources
- Sample built from public UK startup and funding databases; company websites and public sources audited in September 2026.
A4 · About Fast Growth Advisors
Fast Growth Advisors is the Message-Market Fit firm. Founded by Hervé Dhélin, it audits and advises B2B startups and scale-ups on positioning clarity, Go-to-Market alignment and the readability of their message, for prospects and for AI engines alike. The methodology and the 370 audits behind this report are its work.
Contact · herve@fast-growth.fr
Site · fast-growth.fr
A5 · About Positive Marketing
Positive Marketing is a PR and marketing agency for B2B tech companies, based in London and founded in 2009. AI-powered B2B tech PR and marketing experts, the agency serves international clients and is the UK partner of this edition of the report.
Site · positivemarketing.com