120+ AI Shortlist Statistics for 2026: How B2B Buyers Use AI, and Which Firms Get Named (Updated October 2026)

Key takeaways

• 51% of B2B software buyers now start research with an AI chatbot more often than with Google. Half of software buyers now open an AI chat window before they open Google.¹
• Gen AI chatbots are the top source influencing B2B software shortlists, named by 54% of buyers. Being left out of the chatbot answer now costs more shortlist places than a weak review profile.¹
• 69% of B2B software buyers picked a different vendor than planned because of an AI chatbot. A vendor nobody planned to buy from can win the deal if the AI names it.¹
• Our data42 of 114 UK crypto tax firms (36.8%) were never named by ChatGPT, Perplexity or Google AI Overviews across 16 buyer queries. Across our three UK service studies, between a third and three quarters of firms never made an AI answer at all.²
• Our dataOnly 24.8% of UK boutique cyber security compliance firms (37 of 149) received a single AI citation, so about 75% were never cited. In cyber compliance, three in four small specialist firms never made an AI answer.³
• Our dataOfficial crypto protocol pages made up under 10% of AI citations: 8.8% on Perplexity and 7.7% in Google AI Overviews. What AI says about you is written mostly by other people's pages.⁴
• Fewer than 1 in 100 repeated runs of the same prompt returned the same brand list in ChatGPT or Google's AI. One screenshot of an AI answer proves almost nothing about your visibility.⁵
• Only 37.9% of URLs cited in Google AI Overviews also ranked in the organic top 10, roughly half the 76% Ahrefs measured in July 2025. A top-10 Google ranking is now a coin flip, at best, for an AI Overview citation.⁶

How many B2B buyers start with AI?

Start with the buyers. Every survey below asked a slightly different question to a different group, so the percentages won't line up neatly. They all point the same way, and the spread between them is useful in itself.

How many buyers use AI to research purchases

51% of B2B software buyers now start research with an AI chatbot more often than with Google. That is up from 29% in G2's 2025 report, a big jump in about a year even allowing for the change of survey.¹

71% of B2B software buyers rely on AI chatbots at some point in research, up from about 60% seven months earlier. So the remaining 29% are now the exception rather than the norm.¹

94% of business buyers now use AI somewhere in their buying process, up 5 points from 89% a year earlier. Forrester's full report is client-only and the sample isn't public, which is why this sits below G2 in our tiers.⁷

63% of technology buyers used AI in their purchase journey. TrustRadius surveyed in January 2026, so this is the most conservative of the recent figures.⁸

45% of B2B buyers used AI during a recent purchase (Gartner). Gartner's fieldwork ran in August and September 2025, which may explain why it sits well below the 2026 surveys.⁹

Half of US consumers already use AI-powered search. This is a consumer figure from a US panel of 1,927 people, useful as a ceiling check on the B2B numbers.¹⁰

Google hasn't gone anywhere

80% of B2B software buyers still use Google somewhere in the buying journey. Most buyers run AI and Google side by side, so the job is to be visible in both.¹

ChatGPT's share of B2B software buyers' AI chatbot use rose from 47% in 2025 to 62% in 2026. If you only have time to check one assistant, check ChatGPT first.¹

61% of business buyers use private AI tools provided by their own organisation. Answers from a company's private Copilot or ChatGPT deployment are invisible to any public tracking tool.⁷

Business buyers are twice as likely as consumers to use ChatGPT and four times as likely to use Microsoft Copilot. Copilot sits inside Outlook and Teams, which is where a lot of procurement work already happens.⁷

B2B buyers used an average of 7 information sources during a recent purchase. AI is one source among several, which is why third-party pages matter so much later on this page.¹¹

The typical B2B buying decision involves 13 internal stakeholders and 9 external influencers. Twenty-odd people touching one purchase means twenty-odd separate chances for an AI answer to shape the shortlist.¹²

67% of B2B buyers prefer a rep-free buying experience. Buyers who want to avoid sales calls lean harder on whatever the research tools tell them.⁹

AI now builds the shortlist

This is the part that matters for anyone selling. The shortlist forms before the buyer talks to anyone, and AI is now the biggest single input into it.

AI is the biggest influence on the shortlist

Gen AI chatbots are the top source influencing B2B software shortlists, named by 54% of buyers. No other source, including review sites, comes close.¹

Software review sites influence 43% of shortlists, vendor websites 36%, market research firms 36%, peers 33%, forums 30%. Review sites still matter because AI cites them, as the source section below shows.¹

Vendor salespeople influence only 19% of B2B software shortlists. A rep who joins after the shortlist is set has very little left to influence.¹

AI changes which vendor wins

69% of B2B software buyers picked a different vendor than planned because of an AI chatbot. This is the clearest commercial number on the page: AI is moving deals, not just saving research time.¹

33% of B2B software buyers bought from a vendor they had never heard of before AI surfaced it. For a smaller firm this cuts both ways, since the AI can introduce you to buyers who'd never have found you.¹

85% of B2B software buyers think more highly of a vendor when AI names it in an answer. Being named carries some of the weight a referral used to.¹

64% of B2B software buyers run into AI inaccuracies at least a few times a month. Buyers know the answers aren't always right, and they use them anyway.¹

The shortlist is set early, and it holds

94% of B2B buyers rank their shortlist before contacting any vendor. By the time a buyer fills in your contact form, you're already ranked.¹³

Buyers fill an average 3.6 of their 5.1 shortlist slots on day one, and buy from a day-one vendor 95% of the time. If you're not in the first set of names, you're very unlikely to get in later.¹³

The vendor ranked first before sales contact wins 77% of the time, down from 83% in earlier years. Coming second on a pre-contact list means losing about three times in four.¹³

Buyers make first contact with sellers 61% of the way through the buying process, earlier than the 69% seen before. This moved earlier in 2025, partly because buyers wanted to check vendors' own use of AI.¹³

B2B buyers evaluate 5.1 vendors on average, up from 4.5 in earlier years. Five is the average, but AI answers often show fewer, as the next two sections show.¹³

How many firms make the AI shortlist?

Surveys tell you buyers use AI. They don't tell you which firms the AI actually names. That's what our three UK studies measured, running real buyer questions through ChatGPT, Perplexity and Google AI Overviews and recording every firm named.

Our UK studies

Our data8 of 25 UK AI recruitment agencies (32%) were not named once across ChatGPT, Perplexity and Google AI Overviews for 11 buyer queries. These eight weren't obscure firms; Computer Futures is a big name in tech recruitment.¹⁴

Our dataPerplexity never named 68% of the 25 recruitment agencies; AI Overviews missed 64% and ChatGPT 44%. Perplexity was the hardest platform to get onto, so check it separately from ChatGPT.¹⁴

Our dataThree agencies (Harnham, Xcede, Understanding Recruitment) took 78% of all 218 AI mentions in UK AI recruitment, out of 25 agencies tested. When a category has a clear leader in AI answers, everyone else competes for the scraps.¹⁴

Our data42 of 114 UK crypto tax firms (36.8%) were never named by ChatGPT, Perplexity or Google AI Overviews across 16 buyer queries. Over a third of a specialist market missing from every AI answer is a lot of firms whose buyers are being sent elsewhere.²

Our dataChatGPT never named 79% of UK crypto tax firms; AI Overviews missed 74% and Perplexity 62%. ChatGPT was the toughest platform here, the reverse of the recruitment study, so the hardest platform changes by sector.²

Our dataAI recommendations for UK crypto tax were spread thinly: the top 3 firms took 19% of 204 AI mentions, against 78% in UK AI recruitment. Crypto tax has no dominant AI winner yet, so there's still room for a firm to become the default name.²

Our dataAsked how to spot a bad crypto tax accountant, ChatGPT and Perplexity named no UK crypto tax firms at all. On a trust question, the two big chatbots sent the buyer to professional bodies and checklists, not to firms.²

Our dataThe top 10 UK crypto tax firms held 38% of all citations recorded in the study. Even the leading firms each held a small slice, so no one owns this category in AI yet.²

Our dataOnly 24.8% of UK boutique cyber security compliance firms (37 of 149) received a single AI citation, so about 75% were never cited. Most firms in this pool came from the CREST register, so accreditation on its own didn't get them named.³

Our dataBoutique-eligible firms received just 27.5% of AI and search citations for UK cyber compliance queries. The rest went to larger firms, excluded firms, government bodies and directories.³

Our dataOne firm, Evalian, was cited in 6 of 7 UK cyber compliance buyer queries. Evalian also had awards and NCSC and CREST credentials that AI platforms could find on other sites.³

Other studies

73% of cybersecurity vendors got zero ChatGPT citations when buyers asked for vendor recommendations. This is a vendor benchmark with the full report not yet public, but it matches our cyber boutique numbers one tier up.¹⁵

87% of 500 UK business websites failed to appear in AI answers for their main commercial search terms. The study comes from an AI visibility agency and the question set isn't published, so treat the exact figure with care.¹⁶

Only 8% of UK businesses appeared consistently across all major AI platforms. Appearing on one platform is far easier than appearing on all of them.¹⁶

76% of UK businesses were outranked by competitors in AI-generated answers. Most firms have a named competitor sitting in the answer where they'd want to be.¹⁶

89% of brands never appeared in AI answers to category research questions, even though AI described 96% of them accurately when asked directly. AI knows who these brands are. It just doesn't recommend them when buyers ask for options.¹⁷

Asked for the best accountant in six UK cities, ChatGPT, Gemini and Perplexity named 109 firms; 85% were named by only one engine and just one firm by all three. Run the same question on three engines and you'll mostly get three different lists of firms.¹⁸

For business solicitors, 74% of the 61 firms named were picked by only one engine; 3 firms (4.9%) were named by all three. Solicitors got more agreement between engines than accountants, but still nowhere near consensus.¹⁹

No Big Four accountancy firm and no Magic Circle law firm was recommended in any city. Size didn't buy a place in these answers, which is good news for smaller firms.¹⁹

How long is an AI shortlist?

A Google results page shows ten links. An AI answer usually shows a handful of names, and fewer still once the question gets specific. That's the real squeeze.

• Ask "which agency should I use" and you get four to six names.
• Ask "who would you contact first" and you get two or three.

Our dataAI shortlists shrank as buyers got closer to a decision: about 5.8 agencies per answer at the specialisation stage, falling to 2.7 at the decision stage. The decision-stage questions are the ones that lead to a phone call, and they leave the least room.¹⁴

Our dataAI answers named an average of 5.1 recruitment agencies on ChatGPT, 4.8 on Perplexity and 3.6 in Google AI Overviews. AI Overviews gave the shortest lists, so a place there is the hardest to win.¹⁴

83% of technology buyers shortlisted three products or fewer. Buyers shortlist about as many vendors as an AI answer names, so being outside the answer often means being outside the shortlist.⁸

ChatGPT named 5 brands per response on average, against 8.3 for Perplexity, 8.5 for Claude and 9.2 for Gemini. These are API results, which tend to name more brands than the consumer apps people actually use.²⁰

OpenAI's search model averaged 3.3 citations per B2B buyer prompt, with a maximum of 7. This used OpenAI's search model through an API on 100 prompts, so treat it as a rough guide.²¹

AI Mode mentions 3.3 entities per response on average against 1.3 for AI Overviews. AI Mode gives more room, but most people still meet the AI Overview first.²²

59% of AI Overviews name no brand or person at all, against 35% of AI Mode responses. In over half of AI Overviews there's no brand to compete with at all, because none is named.²²

Ranking on Google doesn't get you on the AI shortlist

A lot of firms assume their Google rankings carry over into AI answers. Our data says they often don't, and so does every large study of AI Overview citations.

Firms that rank but don't get named

Our dataHarvey Nash appeared in Google's organic top 20 for 9 of 11 recruitment buyer queries but held a 6.3% AI recommendation share, with zero ChatGPT mentions. Harvey Nash was visible to anyone who Googled and close to invisible to anyone who asked an assistant.¹⁴

Our dataEvery one of the 42 crypto tax firms that AI never named did show up in Google's results; 44 other firms were named by AI without ranking on Google at all. Ranking was no protection here, and not ranking was no barrier.²

Our dataAndersen LLP ranked on Google for 43.8% of crypto tax queries but led AI recommendations; CountDeFi ranked for 12.5% and was never named by AI. Andersen was named more than any other firm despite modest Google coverage.²

Our dataXcede earned 55 AI mentions (68.8% of the leader's total) while appearing in Google's top 20 for only 5 of 11 queries. Xcede shows it works in both directions: strong AI presence with middling rankings.¹⁴

Our dataXEC Recruitment was named 12 times by ChatGPT and 6 times in AI Overviews but never by Perplexity, across the same 11 queries. Same questions, same week, and the platforms disagreed completely about this firm.¹⁴

What the large studies found

Only 37.9% of URLs cited in Google AI Overviews also ranked in the organic top 10, roughly half the 76% Ahrefs measured in July 2025. Ahrefs links the drop to Google expanding each query into many sub-queries before it writes the answer.⁶

31% of AI Overview citations came from pages outside the organic top 100. Pages that don't rank at all for the query are getting cited, which is an opening for firms with good niche content.⁶

BrightEdge's weekly panel put the share of AI Overview citations from the organic top 10 at about 17%, flat across Feb 2025 - Feb 2026. BrightEdge measures differently from Ahrefs, which is why the published overlap figures range so widely.²³

About 90% of AI Overviews contain at least one top-10 URL, even though only 32% of keywords showed keyword-level overlap. Most AI Overviews include at least one top-10 page, but your top-10 page often isn't the one chosen.²³

88% of Google AI Mode citations don't rank in the organic results for the exact-match query. AI Mode goes further from the rankings than AI Overviews do.²⁴

Google's AI Mode and AI Overviews cite the same URLs only 13.7% of the time for the same query. Even Google's two AI products don't agree on sources, so treat them as separate channels.²²

AI Mode and AI Overviews reach semantically similar answers 86% of the time despite citing different pages. The answers sound alike while citing different pages, so you can't tell who was cited just by reading them.²²

What happens to clicks

Google users clicked a traditional result on 8% of visits when an AI summary appeared, against 15% without one. This is 2025 data from 900 US adults, but it's still the cleanest measure of real behaviour.²⁵

Only 1% of visits to a page with an AI summary involved clicking a link inside the summary. Being a linked source in an AI Overview brings very little traffic, so the value is in being named.²⁵

Users ended their browsing session after 26% of searches with an AI summary, against 16% without. More searches end on the results page, which means the AI answer is all many people read.²⁵

Where AI gets its shortlist from

If AI isn't reading your website to decide whether to recommend you, what is it reading? Mostly other people's pages about you, or about your competitors.

Your own site is a small part of it

Our dataOfficial crypto protocol pages made up under 10% of AI citations: 8.8% on Perplexity and 7.7% in Google AI Overviews. YouTube and Reddit alone made up 27.5% of AI Overview citations, more than three times the protocols' own share.⁴

99.99% of 49,391 citations in category research answers pointed to third-party sites, not the brand's own domain. The sponsored write-up comes from the study's owner, but the number matches ours from a different sector.¹⁷

Only 4 of 150 brands earned a citation to their own website in category research answers. Out of 150 brands, four got their own site cited. That's the scale of the problem.¹⁷

Brand websites supply only 5-10% of the sources AI search answers draw on. McKinsey's figure comes from AI Overview source analysis in consumer categories.¹⁰

Only 4.6% of citations in B2B buyer answers pointed to vendor or product pages; TechRadar was cited in 23% of responses. Comparison and 'best of' pages did most of the work in B2B software answers.²¹

55% of pages cited in B2B buyer answers had 'best' or 'top' in the title and a quarter were comparison pages. If there isn't a 'best X for Y' article that names you, the AI has very little to go on.²¹

Our dataWhen AI platforms linked to a UK cyber compliance firm, 49% of links (50 of 103) went to the firm's own website, a much higher share than in crypto protocol answers. Service firms get their own pages linked far more often than crypto protocols do. Our sample is small, so treat it as a pointer.³

Our dataIn Google AI Overviews, 39 of 74 links to UK cyber compliance firms (53%) used the firm's page as a source without naming the firm in the answer. Google used these firms' pages to write the answer and then left their names out of it.³

Which third parties AI trusts

Our data42.4% of AI and search citations for UK cyber compliance queries went to names that were never candidates: enterprise firms, government bodies, review platforms and directories. That was the largest single share, ahead of both boutique firms and the firms the study excluded.³

Our dataSix professional bodies (ICAEW, ACCA, CIOT, ICAS, AAT, IAB) were cited in their own right in UK crypto tax answers. For regulated professions, the professional body's register may be the most direct route into AI answers.²

Our dataAccountancy firms made up 53% of the 382 sources and names AI platforms gave for UK crypto tax queries; tax software, government pages, forums and directories made up most of the rest. Software tools, HMRC pages and forums took the other half of the answer.²

Chambers, The Legal 500 and ReviewSolicitors were cited by all three engines in business solicitor answers. These three directories came up across engines, so a strong profile on them is worth the effort.¹⁹

Legal-sector AI citations concentrated in a handful of prestige directories, while SaaS citations spread across 10,000+ domains. In legal, a few directories decide most of what AI says; in SaaS there are thousands of possible sources.¹⁷

Wikipedia made up 47.9% of ChatGPT's top 10 cited sources and Reddit 46.7% of Perplexity's. These are shares of each platform's top 10 sources, not of all citations, and the data is now over a year old.²⁶

Reddit leads Google AI Overviews' top cited sources at 21.0%, ahead of YouTube (18.8%), Quora (14.3%) and LinkedIn (13.0%). Google's AI leans on its own YouTube and on community sites more than on firm websites.²⁶

In B2B CRM and sales software, TechRadar alone held 8.86% of category citations while vendor sites were largely absent from the top 10. One review publisher holding that share means a single article can decide a lot of AI answers.²⁷

ChatGPT's Reddit share of responses fell from nearly 60% to about 10% in six weeks in 2025, after a model change. One model update cut a whole source category's share in weeks, which is why single-channel strategies are risky.²⁷

Only about 11% of domains are cited by both ChatGPT and Perplexity. ChatGPT and Perplexity mostly read different sources, so winning one doesn't win the other.²⁷

More than 65% of AI search sources in consumer goods and financial services are publishers, user-generated content or affiliate sites. Publishers, user-generated content and affiliates decide most of these answers.¹⁰

What predicts getting named

Brands mentioned on fewer than 2,000 indexed web pages were named in AI answers just 3% of the time. There seems to be a threshold of being talked about elsewhere before AI starts naming you.¹⁷

Referring domains (0.49) and third-party web mentions (0.45) had the strongest correlation with AI brand mentions. Mentions on other sites mattered about as much as links did.¹⁷

Our dataSchema didn't predict AI citation: the only DeFi protocol with JSON-LD schema scored zero, while one with no schema ranked second. Schema didn't decide citations in our crypto data; third-party coverage did.⁴

45% of buyers say a review-site citation is the most confidence-inspiring signal in an AI answer. G2 runs a review site, so it has an interest here, but the figure fits everything else in this section.¹

74% of technology buyers use reviews to inform purchase decisions. Reviews feed the buyer directly and feed the AI answer too.⁸

Only 13% of technology buyers used analyst reports to make purchase decisions, down sharply since 2022. Analyst reports have lost most of their pull, which leaves more room for the sources AI reads.⁸

In local business answers, Google AI Mode cited the firm's own website 90% of the time and Google reviews or profiles 81%. For local businesses AI Mode does read the firm's own site; B2B category questions behave very differently.²⁸

Why one check isn't enough

Run the same question twice and you'll often get a different answer. That changes how you should read any AI visibility check, ours included.

Answers change from run to run

Fewer than 1 in 100 repeated runs of the same prompt returned the same brand list in ChatGPT or Google's AI. The authors sell AI tracking and the study isn't peer reviewed, but nobody has published a result that contradicts it.⁵

Fewer than 1 in 1,000 runs returned the same brands in the same order. Exact rank position in an AI answer is noise, so don't pay anyone to track it.⁵

Leading headphone brands appeared in 55-77% of 994 responses despite almost no repeat lists. The leaders kept turning up even while the full list changed, which is why appearance rate is the measure to watch.⁵

City of Hope appeared in 69 of 71 ChatGPT answers about West Coast cancer care (97%) but ranked first in only 25. Appearing nearly every time and coming first only now and then is a perfectly good result.²⁹

Day to day, roughly two thirds of the sources AI search engines cite change (source Jaccard 0.34-0.42). Most of the sources behind an answer change from one day to the next.³⁰

Day-to-day brand overlap ran at Jaccard 0.45-0.59 across AI engines. Brand names are more stable than sources, but still far from fixed.³⁰

Same-day source overlap was lowest for ChatGPT (0.233) and highest for Gemini (0.505). ChatGPT was the least consistent about sources and Gemini the most.³⁰

45% of AI Overview citations change between generations. Google's own data shows close to half the AI Overview sources rotating between generations.²²

Some questions are steadier than others

Comparison prompts produced the most stable answers: the same lead brand came first 91% of the time. Comparison answers stick, so being named in one is worth a lot.²⁰

Purchase-intent prompts had the lowest brand overlap between runs (40%) and lead-brand stability of just 59%. Ready-to-buy questions churn the most, which is exactly where you'd want stability.²⁰

Cloud computing, the most B2B category tested, had the lowest run-to-run brand overlap at 36%. This was the most B2B category in the test, and it was the least stable.²⁰

How many checks you need

A single run gives a brand-visibility standard error of 0.37; seven runs per prompt are needed to get it under 0.10. Seven runs per prompt gets you a usable number; one run doesn't.³⁰

Tracking over a 2-4 week rolling window brings the standard error down to 0.05 or below. Track over weeks, not on a single day.³⁰

Buyers double-check anyway

94% of buyers who use AI fact-check its answers at least some of the time. Buyers treat AI as a starting point and then go looking for confirmation.⁸

69% of B2B buyers prefer to check AI-generated insights with a sales rep. The salesperson's job is moving towards confirming or correcting what the AI already told the buyer.¹¹

51% of B2B buyers think they are more likely to get misleading information from gen AI, against 49% from a sales rep. Buyers trust AI and salespeople about equally, which means not very much.¹¹

5.2% of 728 AI business recommendations could not be confirmed as an active business trading in the named town. This was local-business data, but it shows AI will sometimes recommend firms that may no longer be trading.²⁸

Who is measuring this?

Most firms aren't measuring any of this, and the tools that do measure it see only part of the picture.

Few firms track it

Only 16% of Fortune 500 consumer brands systematically track their AI search performance. That's from roughly 30 Fortune 500 consumer-brand CMOs, a small group, but if big brands aren't tracking, most B2B firms won't be either.¹⁰

27% of marketers consistently track whether their brand appears in AI answers; 25% don't track it at all and 12% didn't know they could. Fewer than a third have a consistent process.³¹

91% of UK businesses showed no sign of an AI visibility strategy. This was judged by an agency selling AI visibility work, so read it as a rough guide.¹⁶

92% of 120 UK independent business websites had no AI crawler policy and about 6% had correct business-type structured data. Most small UK sites haven't told AI crawlers anything at all.²⁸

Analytics undercount AI traffic

70.6% of AI visits in Loamly's database arrived without a referrer, so analytics tools file them as direct. This replaced an earlier, unmeasured '60%' estimate that is still widely quoted.³²

ChatGPT referral traffic rose 130% month on month in June 2026 after inline citations rolled out, as GA4 added a dedicated AI Assistant channel. GA4's new AI Assistant channel makes this traffic easier to see from mid-2026 onwards.³³

ChatGPT sends 92.4% of trackable LLM referral traffic across 166 sites, after 12.8x growth in 19 months. If you're checking only one AI platform's referrals, it's ChatGPT.³⁴

Monthly LLM referral sessions grew 9.9x to 644,478 by May 2026, despite a 50% one-month drop in November 2025. AI referral traffic is growing fast but swings sharply from month to month.³⁴

Perplexity referral traffic is down 61% from its March 2025 peak and Copilot's down 96% since August 2025. Platforms rise and fall quickly, so don't build a plan around one of them.³⁴

Claude referral sessions grew 64x and overtook Perplexity in March 2026. Claude grew from a tiny base, so the multiple looks bigger than the actual traffic.³⁴

Across 28 B2B SaaS companies, homepages drew 31.9% of LLM-referred sessions and 82% of AI-attributed leads. Buyers arriving from AI often go straight to the homepage, so make it explain who you serve.³⁵

ChatGPT sent 72.2% of LLM-referred sessions to 28 B2B SaaS sites; overall AI session-to-lead rate was 0.13%. Few AI visits turned into leads on these sites, which is why being named matters more than the clicks.³⁵

34.6% of AI-referred traffic to SaaS sites lands on search and category pages. In SaaS, a third of AI visits land on search and category pages.³⁴

What's coming next

OpenAI began showing ads in ChatGPT to logged-in US adults on the Free and Go tiers on 9 February 2026, placed beneath answers. Paid placements now sit under some ChatGPT answers, next to the organic recommendations.³⁶

44% of AI search users say it's now their primary and preferred source of insight. A big share of AI search users have made it their main way to find things out.¹⁰

McKinsey estimates $750 billion of US revenue will flow through AI-powered search by 2028. This is a US forecast and depends on adoption assumptions.¹⁰

Brands that don't adapt could lose 20-50% of their traditional search traffic. McKinsey's estimate covers traditional search traffic, not total demand.¹⁰

What these statistics mean in practice

Check the decision-stage questions first

Awareness questions get long lists. The questions buyers ask just before they pick up the phone, like "who would you contact first", get two or three names. If you only test a few questions, test those, on ChatGPT, Perplexity and Google separately.

Run each question more than once

A single answer is one draw from a changing set. Run each question several times across a few days and count how often you appear. And remember that answers vary by user and login, so the answer a prospect sees may not match yours.

Work on the pages AI already reads

Your own site is a small share of what gets cited. The work that moves the numbers is getting named on the comparison articles, directories, review sites and professional body registers that already show up in your category's answers. Look at which sources appear for your questions and start there.

Don't assume your rankings will carry you

Every one of the 42 crypto tax firms AI never named was visible on Google. Treat AI visibility as a separate channel with its own checks. If you want to see where you stand, our AI visibility reviews run your real buyer questions across the main platforms.

Frequently asked questions

What percentage of B2B buyers use AI to research vendors?

Depending on the survey, between 45% and 94%. G2's March 2026 survey of 1,076 software buyers found 71% use AI chatbots during research and 51% start there before Google.

Do AI chatbots influence vendor shortlists?

Yes. In G2's 2026 survey, 54% of buyers named AI chatbots as an influence on their shortlist, more than any other source, and 69% picked a different vendor than planned because of AI.

How many firms get recommended by AI?

Usually a small share. In our UK studies, 32% of AI recruitment agencies and 37% of crypto tax firms were never named, and about 75% of boutique cyber compliance firms got no AI citation.

Does ranking on Google mean AI will recommend you?

Not reliably. Ahrefs found only 37.9% of AI Overview citations come from the top 10 organic results, and in our crypto tax study every firm AI ignored still appeared on Google.

Why does my firm show up in some AI answers and not others?

AI answers change between runs. SparkToro found the same brand list repeats less than 1 in 100 times, so track how often you appear across many runs rather than relying on one check.

Methodology

We checked 166 candidate statistics against their original publications and kept 126. The other 40 were dropped. Some had no traceable origin and were only ever quoted second-hand. Others were superseded by a newer or more direct figure, came from vendor pages with no published method, or added nothing the page didn't already say. Several widely repeated figures fell into the first group.

Every statistic is tiered. Strong means we read the original source and it states a clear sample, period and method. Moderate means a single source with a plausible method but a small sample, vendor ownership or self-reported data. Weak means illustrative only. Key takeaways use strong-tier figures only.

Our own figures come from three aivisibilitygap.com studies run between June and August 2026 (UK AI recruitment agencies, UK crypto tax accountants and UK cyber security compliance consultants) and the 2026 crypto protocol study on cryptocontent.dev. Each ran verbatim buyer-journey questions across ChatGPT, Perplexity and Google AI Overviews, alongside Google organic results, and recorded every firm named. AI answers vary by user, session and day, so these are snapshots. The cyber figure counts firms with no AI citation, while the other two count firms with no AI mention.

Corrections, 18 October 2026. Cyber compliance figures were recalculated from the study's published dataset: 149 firms (not 143), 229 citation records including the surviving Q3 answers, and population shares of 27.5% / 30.1% / 42.4% (previously 21.6% / 24.4% / 54.0%). The crypto protocol official-source share now uses the study's version 2 recount (8.8% / 7.7%, previously 1% / 4%). The crypto tax trust-stage finding now refers to the one question where no firms were named.

Third-party data on this page was collected between March 2025 and September 2026. Statistics were verified on 13 October 2026. This page is refreshed every three months; the next update is due in January 2027.

Sources

1. G2, 2026 AI Search Insight Report (The Answer Economy) (Apr 2026)
2. aivisibilitygap.com, UK Crypto Tax study, AI Visibility Gap: UK Crypto Tax Accountants 2026 (Jul 2026)
3. aivisibilitygap.com, UK Cyber Compliance study, AI Visibility Gap: UK Cyber Security Compliance Consultants 2026 (Aug 2026)
4. cryptocontent.dev, Crypto Protocol study, 2026 AI Citation Visibility Study: 50 Crypto Protocols (May 2026)
5. SparkToro / Gumshoe.ai, AIs are highly inconsistent when recommending brands or products (Jan 2026)
6. Ahrefs, AI Overview citations and the top 10 (Mar 2026)
7. Forrester, 2026 Buyer Insights: Buyer Usage of GenAI (Buyers' Journey Survey) (2026)
8. TrustRadius (HG Insights), 2026 B2B Buying Disconnect Report (Jul 2026)
9. Gartner, Gartner sales survey: 67% of B2B buyers prefer a rep-free experience (Mar 2026)
10. McKinsey, New front door to the internet: Winning in the age of AI search (Oct 2025)
11. Gartner, Gartner Survey: 69% of B2B buyers turn to sales reps to validate AI insights (May 2026)
12. Forrester, The State of Business Buying, 2026 (Jan 2026)
13. 6sense, 2025 B2B Buyer Experience Study (Nov 2025)
14. aivisibilitygap.com, UK AI Recruitment study, AI Visibility Gap: UK AI Recruitment Agencies 2026 (Jul 2026)
15. GrackerAI, State of AI Search Visibility in Cybersecurity 2026 (Apr 2026)
16. LLM Listed, UK AI search visibility study (via DIGIT / Decision Marketing) (Jun 2026)
17. Victorious, Q2 2026 Quarterly Search Report (Jul 2026)
18. Whito, AI accountant recommendations (Aug 2026)
19. Whito, AI solicitor recommendations (Aug 2026)
20. Conductor, AI Brand Recommendation Study (2026)
21. Datadab, What ChatGPT actually cites (May 2026)
22. Ahrefs, Are AI Mode and AI Overviews Just Different Versions of the Same Answer? (Dec 2025)
23. BrightEdge (via API Serpent comparison), Weekly AI Search Insights (2026)
24. Moz, AI Mode citations study (Feb 2026)
25. Pew Research Center, Google users are less likely to click on links when an AI summary appears (Jul 2025)
26. Profound, AI platform citation patterns (Jun 2025)
27. Goodie (via 5W State of AI Citations 2026), State of AI Citations 2026 (May 2026)
28. Whito, AI business recommendation statistics, Edition 1 (Sep 2026)
29. SparkToro / Gumshoe.ai, AIs are highly inconsistent when recommending brands or products (Jan 2026)
30. University of St. Gallen (Schulte, Bleeker, Kaufmann), Don't Measure Once: Measuring Visibility in AI Search (GEO) (Apr 2026)
31. Page One Power, Brands are flying blind in AI search (Mar 2026)
32. Loamly (via David Rosam), The AI Traffic Attribution Crisis (Feb 2026 update) (Feb 2026)
33. seoClarity, ChatGPT's inline citations surge AI referral traffic (Jul 2026)
34. Previsible, 2026 State of AI Discovery Report (Jul 2026)
35. TripleDart, What Page Types Do LLMs Cite? A Study of 28 B2B SaaS Companies (2026)
36. OpenAI (via Search Engine Journal), Testing ads in ChatGPT (Feb 2026)