Nine of 40 UK B2B software companies were named in none of the 24 shortlist answers they received from four AI platforms, according to The AI Shortlist 2026, a report Recrawled published on September 29, 2026. The same research reports YouTube in 25% of Google AI Overview answers about the companies and in none of the 360 ChatGPT answers.

In Short

A UK agency put 1,608 questions of the kind software buyers type to ChatGPT, Gemini and two Google search tools, then checked which British software companies came back in the answers. Nine of the 40 companies it studied closely were never recommended on any of the four tools when a buyer asked who to consider, even though every one of the 40 was recognised once its name was typed in. So a company's place in an AI answer depends on how the question is worded and which tool is asked, and the results describe one day, September 4, 2026, rather than a permanent ranking.

A one-day snapshot of 1,608 questions

Recrawled describes itself as an answer engine optimisation (AEO) agency, and The AI Shortlist 2026 was researched by two of its co-founders, Jess Brookes and Charles Törnros. The study began with 200 UK-headquartered B2B software companies, each with between 20 and 150 staff and selling a specialist product to other organisations. They were spread across seven sectors: contact centre and customer experience, HR and workforce, compliance and risk, fintech and insurtech, supply chain and field service, healthtech and edtech, and other specialist software. According to the report, each company was checked against its own website, its filings and its review-site listings rather than pulled from a list. Forty were then studied in depth, in 33 buyer categories across six of the seven sectors.

The questions went to ChatGPT, Gemini, Google AI Overviews and Google AI Mode, each in default mode, signed out and with the location set to the UK. The run took place on September 4, 2026. In all, 1,608 questions were put: 168 to the sectors (six broad questions per sector, on each of four platforms) and 1,440 about the 40 companies. Every studied company received nine questions, six asking which companies to shortlist and three asking about the company by name. On four platforms that comes to 36 answers per company, 24 to the shortlist questions and 12 to the by-name ones. Of the 1,608 questions, 1,600 produced answers. The remaining eight were by-name questions for which Google displayed no AI Overview, and the report records that outcome as a result in its own right.

The six shortlist questions covered different buying situations. Three say UK or name a UK buyer. One drops the country and names the size of the buyer instead, one starts from the market leader's name, and one is a head-to-head between the leader and its nearest rival. In the report's example, the UK-specific wording asks about court bundling software for a UK law firm, while the size-based version asks about court bundling software for mid-sized law firms.

The executive summary carries its argument in a heading: "Being recognised by AI is not the same as making the shortlist." When checked by name, all 40 companies were correctly identified by at least one platform, according to Recrawled. Asking which companies to consider produced a different picture. Twenty-nine were named on at least one shortlist question that specified the UK, two were named only when the question did not say UK, and nine were named in none of their 24 shortlist answers, on any platform. No company appeared in all 24. The most any managed was 22.

Wording changed the result

Four of the six shortlist questions tested the same companies in different buying situations. A company counts in the report's chart if any of the four platforms named it anywhere in the answer; the final column is the appendix's count of individual answers.

Question typeCompanies named by at least one platformIndividual answers naming the company
UK-specific shortlist question63% (25 of 40)36% (173 of 480)
Mid-sized buyer, no "UK"43% (17 of 40)22% (35 of 160)
Alternatives to the market leader13% (5 of 39)4% (6 of 156)
Market leader against nearest rival0% (0 of 38)0% (0 of 152)

One company was excluded from the alternatives question and two from the head-to-head question because their own product was the subject. For the same reason the answer-level figures add up to 948 rather than 960: 12 answers in which a company's own product was the subject are left out. The company was named in 11 of those 12, so across all 960 answers it was named 225 times, or 23%.

The first two question types were equally specific, according to the report, so the drop was not simply the result of narrower wording. In one category, removing "UK" was enough for three of the four platforms to switch to companies based outside the UK. A buyer asking for alternatives to the market leader is "explicitly looking beyond the market leader", yet only 13% of the companies studied appeared in those answers. The fourth question set the market leader versus its nearest rival. None of the other 38 companies showed up in any of the four platform answers, according to Recrawled; AI sometimes introduced another brand, just never one of the companies studied. Because that prompt already names two vendors, the result measures whether a third name found room in the answer, which is narrower than general invisibility.

The platforms disagreed

Of the 31 companies named on at least one platform, Gemini never named 13 (42%), Google AI Mode 11 (35%), Google AI Overviews six (19%) and ChatGPT five (16%). Across the same 240 shortlist questions, Google AI Overviews named the studied companies 69 times and AI Mode 44. Google's two AI experiences therefore differed substantially from each other, according to the report.

Other datasets point toward platform divergence as well. The Semrush AI Visibility Index, covered by PPC Land on June 30, 2026, analysed 126 million US prompts from January to April 2026 and found that only 36 of more than 1,200 brands kept top-100 status on every platform in every month studied. In that index the mention-source overlap, which Semrush reports by platform, was 64% on AI Overviews and 30% on Gemini, and the average number of brands named per response ran from 3.3 on ChatGPT to 5.2 on AI Mode.

Video citations separate Google from the other platforms

According to the report, YouTube appeared in 25% of Google AI Overview answers and 19% of AI Mode answers about the 40 companies. For Gemini the share was 2%, and ChatGPT showed none in its 360 answers. The wording shifts within the document, though: pages 5 and 10 say YouTube "appeared", while the appendix says the answers "cited at least one video". Counted by citations rather than answers, YouTube was 3.3% of the 7,704 citations across every answer the researchers read, and by distinct pages it was 4% of 2,750 pages on 1,223 websites.

Measurements elsewhere differ by platform, market and method. Meltwater's analysis for August 2026, reported by PPC Land on September 23, tracked eight systems including ChatGPT and ranked YouTube the most-cited platform across all eight. Recrawled's zero for ChatGPT sits apart from that finding. The datasets differ in query type, market, period and measure, and the report does not address Meltwater's work. Tinuiti's Q2 2026 report, in the same article, found that YouTube citations in AI Mode more than quadrupled between January and April 2026, and in AI Overviews more than doubled.

Eight of the 40 companies had their own pages cited in answers that recommended somebody else, across 19 answers, according to the report. One vendor selling software that checks suppliers' security had published a guide to the best software in its category. AI cited the guide in nine shortlist answers. In four of them, it used the vendor's content but recommended other companies. In the report's usage, cited means the answer displayed the page as one of its sources, while named means the company itself appeared in the answer.

The report's source breakdown shows where the cited material came from. Of the 2,750 distinct pages, 67% were on software vendors' own websites, 9% on review directories, 4% on YouTube, 3% on official and industry bodies, 3% on press, blogs and forums, and 14% on other sites. Pages were classified by website, with vendor sites matched against every brand named in the run and a random 100 read by hand as a check. Around a quarter of the pages cited in each company's shortlist answers sat on the websites of the most frequently named companies in its category.

A separate study points at a similar gap between being cited and being chosen. Scrunch, which sells AI visibility monitoring, analysed roughly 10,000 AI-cited URLs and found that when a brand's own ranked list was cited, the brand's recommendation rate rose from about 4% to roughly 7%, and its mention rate from about 19% to 39%. A competitor was recommended instead of the author 24.3% of the time. PPC Land reported the study, which covered data collected from May 17 to June 30, 2026, in late September.

Review counts did not track naming

Across the 40 companies, the correlation between review volume and AI naming was -0.06, which the report calls effectively zero. The company named most often had 19 Capterra reviews and no G2 listing, and it appeared in 22 of its 24 answers. One company with 637 reviews was named in just one of 24. Among the ten companies named most, the typical company had two to three reviews across G2 and Capterra; among the nine never named, the typical company had 13. Counts were taken from G2 and Capterra on the day of the run, and a missing listing counted as zero.

The report is careful about what this does and does not show. Finding no link, it says, is "not proof there is no link anywhere". Among the companies named most, Recrawled says it repeatedly saw specific third-party proof and measurable facts in the answers and pages that named them, which it describes as "an observed pattern, not evidence that those signals caused the naming".

Buyer behaviour gives some context. A Semrush survey of 519 B2B buyers who use AI tools for work, published by PPC Land on July 20, 2026, found that 92% said AI had shaped their vendor shortlist and 66% had spotted vendors missing from AI answers. Only 7% said brand recognition determines whether they notice a vendor in AI responses, against 53% who prioritise a close match with their specific use case. And 38% checked reviews on G2 or similar sites after receiving an AI recommendation.

Dominant brands, and room beneath them

Of the 33 buyer categories, 61% had one dominant brand appearing in at least three-quarters of the answers. Even so, 50% of the 40 companies took first position on at least one buyer question. Each category received the same six shortlist questions on four platforms, so 24 answers, read once for each company in it.

The limit of a strong position shows in one example. A vendor that was not the leader in its category was named in 16 of its 24 answers, on all four platforms, in much the same terms each time. When the buyer asked for alternatives to the leader, or compared the leader with its nearest rival, it was named in none of those eight answers.

The appendix adds a wider view. On the six broad sector questions, separate from the shortlist set, 23 of the 200 companies were named at all, and global brands dominated. Contact centre and customer experience software returned 54 different brand names, one of them in 17 of that sector's 24 answers; fintech and insurtech returned 98.

Accuracy once the buyer knows the name

The sixth finding moves from visibility to correctness. Of the 40 companies checked by name on four platforms, one in three got at least one answer about the wrong business, two in five had review questions answered using employee-review sites, and one in five had pricing come back inconsistently or without a source. Thirteen of the 40 shared a name with an unrelated business or product.

The report lists four examples:

  • A UK desk-booking vendor was answered as a restaurant and cocktail bar in Northampton, by ChatGPT.
  • A UK contact-centre vendor was answered as a Continental car tyre, by Gemini and Google AI Overviews.
  • A UK insurance-data vendor was answered as a Dyson robot vacuum cleaner, by Gemini and Google AI Mode.
  • A UK cyber-risk vendor was answered as a cannabis producer, by Gemini.

The employee-review problem has a quotable case. Google AI Mode reproduced an employee-review site verbatim in response to a reviews question for one vendor: "Was a great company, on the decline." Elsewhere in its answers, according to the report, AI Mode had called that vendor's product "the gold standard".

Same-name confusion has appeared outside B2B software. In July 2026, a Google AI Overview told searchers that Anna Mae's Bakery and Restaurant in Millbank, Ontario was closing on July 31; the closure belonged to a different business of the same name in Illinois.

Why the findings matter to marketing teams

The report opens with three figures from other organisations. According to 6sense's 2025 B2B Buyer Experience Report, covering nearly 4,000 buyers, 95% of purchases went to a vendor already on the buyer's Day One shortlist. Forrester's 2026 report The State Of Business Buying, with nearly 18,000 buyers, found that 94% of business buyers used AI while buying. TrustRadius, in its 2026 B2B Buying Disconnect Report of 1,862 buyers, found that 83% of technology buyers shortlisted three or fewer products, with an average shortlist of 2.7. Recrawled's framing is that the list is short, forms early, and now passes in part through AI.

Other numbers point the same way. Bain data cited in a LinkedIn marketing blog post, and covered by PPC Land on May 26, 2026, put the share of B2B buyers who start with a day-one vendor list at 86%, with 81% buying from it. Whether marketing teams can see their position on that list is a separate question. An index from Agentcy found that 81% of 104 senior B2B marketing leaders consider AI visibility a blind spot and that only 10% can consistently connect AI-driven touchpoints to revenue. PPC Land reported it on March 5, 2026.

Software vendors have started to build measurement for the gap. HubSpot's AEO tool, reported on April 14, 2026, tracks brand visibility in ChatGPT, Gemini and Perplexity and costs $50 a month as a standalone product. Cloudflare opened an AEO Visibility Dashboard in early access on August 6, 2026, covering Claude and GPT models. Recrawled's study differs in form: one fixed day, a defined sample, and separate counts per question type and per platform, which the report argues a single blended score would hide.

Recrawled's recommendations and commercial position

Each finding in the report ends with a recommendation box. Recrawled recommends building visibility around the buying situations that matter rather than one broad category question, and tracking platforms separately instead of as a blended score. It also recommends content that shows why a company belongs on the shortlist for the problems it solves, specific proof rather than review volume as the goal, consistent category, pricing and product claims across sources, and a focus on the situations where a company has the strongest reason to win.

The authors also sell services in this area. According to the report, Recrawled offers diagnosis and setup, repair and optimisation, and management and continuity, plus a free AI Shortlist Assessment covering ChatGPT, Gemini, Google AI Overviews, Google AI Mode and Claude, delivered by email within three working days. The companies studied were not contacted and took no part in the research, according to the report.

What the data does not show

The report states its limits. It is "a snapshot of one day", and 40 companies are "a sample, not a census". Recrawled notes that the same questions asked the next day may produce different answers on all four platforms, and that nothing in the study predicts what any platform will say next. SISTRIX's 17-week measurement across three platforms and six countries gives a sense of the movement involved: AI Mode rotated 56% of its citations each week and ChatGPT 74%, while AI Overviews stayed completely stable for just over half of queries, according to the analysis PPC Land reported on April 30, 2026.

Recrawled also separates citation from cause. The platforms report which pages they cited, not why they named the companies they named, so a cited page is an observation and not an explanation of a recommendation.

Two presentation points stand out in the document itself. The executive summary counts 29 companies named on at least one UK-specific shortlist question, while the first bar in the wording chart shows 25 for the UK-specific question type, and the report does not set out the four-company difference in so many words. The YouTube figure is described as appearance in one place and as citing at least one video in another. Neither changes the headline count of nine companies named in none of their 24 shortlist answers, but both leave open exactly which answers each percentage covers. Fuller methodology is available from Recrawled on request, according to the report.

Timeline

Summary

Who: Recrawled, a UK answer engine optimisation agency, through its co-founders Jess Brookes and Charles Törnros. The subjects were 200 UK-headquartered B2B software companies with 20 to 150 staff, of which 40 were studied in depth. The platforms queried were ChatGPT, Gemini, Google AI Overviews and Google AI Mode.

What: A report titled The AI Shortlist 2026, based on 1,608 questions and 1,600 answers. Nine of the 40 companies were named in none of their 24 shortlist answers. Named share fell from 63% to 43% to 13% to 0% as the question moved from a UK-specific shortlist to a no-country version, to alternatives to the leader, to a leader-versus-rival comparison. YouTube appeared in 25% of AI Overview answers and none of ChatGPT's 360, and review volume showed a correlation of -0.06 with naming.

When: The research was run on September 4, 2026 and published on September 29, 2026.

Where: The questions were asked with each platform in default mode, signed out and with the location set to the UK, about UK B2B software companies.

Why: Recrawled set out to learn what gets a B2B software company into an AI shortlist, citing figures that buyers form short lists early and increasingly use AI. The report's authors sell AEO services, and the study is a one-day sample of 40 companies, which Recrawled itself describes as not a census.