State of AI Citations in UK B2B SaaS, 2026

When a buyer asks an AI engine “what’s the best contract management software for in-house legal teams in the UK”, they get a short answer with three to five names. Either your company is in that answer or it doesn’t exist for that buyer. We wanted to know how often UK B2B SaaS companies make it in, and whether the things the industry tells you to fix are the things that decide it.

How was the study run?

• 50 companies. Funded, growth-stage UK B2B SaaS across five categories: security and compliance (17), customer experience and research (9), property tech (9), legal tech (8) and finance operations (7).
• 250 questions. Five high-intent questions per company, written the way a buyer in that category would type them, for example “best document automation software for law firms UK”.
• One engine. Google AI Mode, logged out, market set to the UK, all run on 8 October 2026. One run per question.
• 634 pages crawled. About 13 pages per site, checked for crawler access, llms.txt, structured data, heading structure and links to outside profiles.
• Named means the company appears in the answer a buyer reads. We counted only what came back. Nothing is estimated.

We are not naming the 50 companies. This is about the category, not about calling anyone out.

How often are UK B2B SaaS companies named in AI answers?

Not often. 24 of the 50 companies were named in none of their five answers. Only 7 were named in three or more, and none in all five.

These are not unknown startups. Every one has a real product, funding and customers, and Google’s AI still recommended someone else.

Does a good technical score get you named?

Not on its own. We scored every site out of 100 for technical readiness; the average was 65. The correlation between that score and how often the company was named was 0.24, which is weak.

It does push in the right direction: sites scoring 75 or more were named 1.6 times on average, sites under 60 were named 0.8 times. But 5 of the 14 best-built sites were named zero times. The highest score in the sample, 95 out of 100, belonged to a company named in none of its five answers. One site that scored 53 was named in four.

The technical layer decides whether an AI engine can use your site. It does not decide whether it wants to.

Are UK SaaS sites blocking AI crawlers?

No. 49 of the 50 sites let every AI crawler in. The one exception blocks two training crawlers, CCBot and Bytespider, and leaves the answer crawlers open. Not one site blocked the retrieval agents that fetch pages while a model writes a live answer. Access is not the problem in this market. The AI gets in and finds very little worth quoting.

Does llms.txt help companies get named?

Not in this data. 21 of the 50 sites (42%) publish an llms.txt file. Those companies were named 0.95 times on average; companies without one were named 1.14 times. 52% of the llms.txt sites were never named, against 45% of the rest.

That does not mean llms.txt hurts — the sample is small and one engine is one engine. It does mean there is no reason to expect it to move anything on its own. More on what the file is for in llms.txt: what it actually does.

Are B2B SaaS sites written to be quoted?

Mostly not. AI answers borrow from pages that ask a buyer’s question and answer it in a short, self-contained paragraph. Across 7,663 headings on 634 pages:

• 13% were phrased as a question a buyer would ask. On the median site it was 11%, and four sites had none.
• 27% were followed by a direct answer of 25 to 90 words, the length AI answers tend to lift.
• 39 of 50 sites had no FAQ markup on any page we crawled.
• 9 of 50 had no structured data at all.

Can AI engines tell who these companies are?

Often not. Models confirm a company’s category partly by checking outside profiles such as LinkedIn, G2, Capterra and Crunchbase, which a site points to with sameAs links in its Organization markup. 26 of the 50 sites had no such links, and 35 had two or fewer. See entity clarity for why that matters.

Which categories are hardest to get named in?

Legal tech had the lowest rate. Its answers were dominated by a handful of names: Lexis+ AI (17 appearances), LEAP Legal Software (16), Spellbook (14) and Harvey (13). In security awareness, Hoxhunt appeared 23 times and KnowBe4 22 times. In property tech, Reapit appeared 17 times and Alto 14. When you are not in the answer, a competitor is, every time the question gets asked.

What should a UK SaaS company do with this?

1. Measure before you fix. Ask AI engines the five questions your buyers type and write down who gets named. That is your baseline.
2. Rewrite headings as buyer questions. Start with the homepage and main product pages. Under each, add a 40 to 60 word plain answer naming your company, the problem and who it is for.
3. Connect your entity. Add sameAs links to every real outside profile in your Organization markup. Under an hour.
4. Do not stop at the technical audit. A 95/100 site was not named once. Content that answers questions is what gets quoted.
5. Do not buy llms.txt as a strategy. Fine to have. It is not what moved anyone in this sample.

What does this study not prove?

• One engine, one run per question, one day. AI answers change between runs, so this is a snapshot of 8 October 2026, not a trend. See measuring AI visibility without fooling yourself.
• Five questions per company, written per category; they are not identical across companies.
• The 50 companies came from our own prospect research, not a random sample of UK SaaS.
• The readiness score is our own scoring model; the paid measurement protocol is published on /methodology.
• Correlation is not cause. A company named often may also be bigger, older or better known.

We will rerun the same questions next quarter and publish the change.