94% of B2B buyers now use AI to pick vendors. The data, unpacked

Claims about AI transforming B2B buying were easy to dismiss as vendor enthusiasm until 2026, when two substantial datasets arrived. The first is Forrester's 2026 Buyers' Journey Survey, which asked 18,000 business buyers worldwide about their most recent purchase. The second is G2's research into how B2B software buyers specifically use AI chatbots during evaluation, with a repeat measurement seven months apart that captures the trend as well as the level. Different populations, different methods, same conclusion: AI sits inside the buying process now, at every stage from first research to final business case.

This article walks through both studies number by number, then draws out what the figures change about how B2B marketing actually works: where the shortlist forms, who gets on it, and why half the market is absent from the conversation entirely. Every statistic is attributed, and where a figure needs a caveat, the caveat is attached. If you only take one number into your next planning meeting, take this one: 69 percent of software buyers changed their expected vendor because of what an AI chatbot told them.

A quick scope note. Forrester's survey covers business buying broadly, while G2's work focuses on software purchases, which is why the two sets of percentages should be quoted separately rather than blended. Both belong in the same argument; they answer adjacent questions.

Start with the headline: 94 percent of buyers used AI during their most recent purchase. At that level, AI use stops being a segmentation variable. There is no meaningful AI-using buyer persona to target, because the non-users round to nobody. The interesting information moves into what buyers do with the AI, and Forrester measured three activities in detail.

55 percent compared vendors in AI tools. Comparison is the activity where brands win or lose deals, and it now happens inside a conversation the vendors cannot see, on criteria the assistant chooses, informed by whatever sources the assistant retrieves. 54 percent researched products with AI, which means feature questions, pricing questions and how-does-it-work questions get answered by the model rather than by your carefully staged product tour. And 47 percent built internal business cases before vendor contact: nearly half of buyers assembled the justification, the numbers and presumably the shortlist before any vendor knew the deal existed.

That last figure deserves the most attention, because it relocates the moment of influence. The classic B2B funnel assumes the vendor participates in framing the problem: a rep discovers needs, a solutions engineer shapes requirements, marketing nurtures the evaluation. When the business case is drafted with AI assistance before first contact, the framing is over by the time you hear about the deal. Whatever the AI said about your category, and about you, is baked into a document circulating inside the buyer's company. Sales teams call this arriving late; the data says late is now the default arrival time.

G2's research narrows to software buying and adds motion to the picture. 51 percent of B2B software buyers now start research with an AI chatbot more often than with Google. Seven months earlier the same measurement was 36 percent. Fifteen points in seven months is not gradual channel drift; it is the fastest reallocation of first-touch attention since mobile, and it happened while most marketing teams were still debating whether to take AI search seriously.

Within the chatbot share, concentration is heavy. ChatGPT is the dominant chatbot for software research at 63 percent, 72 percent of buyers use ChatGPT during vendor evaluation, and 44 percent also use Perplexity during shortlisting. The workload buyers assign to these tools is analytic: comparing vendor strengths and weaknesses is the top AI use case in software research at 41 percent. Buyers are asking the machines to do exactly the comparative work that analyst firms, review sites and vendor comparison pages used to mediate. This is why comparison-shaped content has become the most valuable page type in B2B, a dynamic we break down in our guide to comparison pages in AI answers.

Note what the two engines imply for coverage. ChatGPT's dominance makes it the mandatory surface, and Perplexity's 44 percent shortlisting share makes it the most underrated one. Cross-platform citation analyses find only about 11 percent of domains are cited by both engines, so presence in one does not transfer to the other. Reachroller tracks both, separately, through official APIs, precisely because the buyer data says both sit inside real evaluations.

Usage statistics show attention; outcome statistics show money. G2 measured two outcomes that should reorganize B2B marketing priorities. First, 69 percent of B2B software buyers chose a different vendor than they originally expected because of AI chatbot output. The expected vendor, the one with the brand awareness, the retargeting budget and the incumbent advantage, lost the deal in a conversation it never knew was happening. Second, 33 percent of buyers bought from a brand they had never heard of before the AI named it. A third of purchases went to vendors whose entire first impression was made by a machine's recommendation.

Together these two numbers dismantle a comfortable assumption: that AI answers merely reflect existing brand strength, so strong brands can ignore them. The data shows the opposite. AI answers override prior brand preference in most evaluations where they participate, and they mint brand awareness from nothing for a third of buyers. The answer layer is an active participant in vendor selection with its own preferences, and its preferences are legible: they come from training data and from the sources the engine retrieves, which means they can be studied and, within honest limits, changed.

A measurement caveat belongs here. AI answers are probabilistic: SparkToro measured under a 1 percent chance that two identical ChatGPT runs return the same brand list. The 69 percent switch rate does not mean the same rival wins every run; it means the answer layer as a whole moves buyers. For a brand, the response is statistical too: measure your mention rate across repeated runs, and work to raise the rate rather than to win a single lottery draw. That is exactly how Reachroller's tracking is built: repeated scheduled runs, stored answers, and a score that only counts literal brand mentions in the text.

Set the demand-side data against the supply side and the gap is stark: 51 percent of B2B tech brands have zero citations across ChatGPT, Perplexity and Gemini, per G2's research. Half the vendors in the market contribute nothing to the conversations where 51 percent of buyers now start research and 69 percent change their minds. Their features, their pricing, their differentiation: none of it exists where the shortlist forms.

For invisible brands this is a quiet emergency, because absence does not produce a signal. No lost-deal report says the buyer's chatbot never mentioned us. Pipeline just thins, win rates drift, and the expected-vendor advantage erodes deal by deal while every dashboard stays green. The first step out is simply to look: run your buyers' questions and see whether you exist in the answers. We published a manual procedure in our step-by-step AI visibility audit, and the same baseline takes minutes with a tool.

For visible brands, the same statistic is an opening. If half your competitors are absent from the answer layer, the share of voice available to whoever shows up is enormous, and the 33 percent never-heard-of-them-before figure proves unknown challengers convert that visibility into revenue. Early movers in empty channels compound their advantage; the buyer data says this channel is half empty and filling with buyers.

The traditional B2B funnel assumed the vendor could observe and influence each stage: awareness through ads and content, consideration through nurture, decision through sales. The 2026 data describes a different shape. Awareness now happens inside AI answers, where 33 percent of buyers meet their eventual vendor for the first time. Consideration happens there too, with 55 percent comparing vendors in AI tools. Even the decision scaffolding moves upstream, with 47 percent building business cases before contact. The visible funnel, the one your analytics can see, starts after the important choices.

This does not make existing marketing worthless. It re-prices it. Content still matters, but its most valuable reader is now a retrieval system deciding what to tell a buyer, which changes how pages should be written: direct claims, real statistics, honest comparisons, named sources. The Princeton and Georgia Tech GEO research, published at KDD 2024, quantified this: adding quotations, statistics and cited sources boosted content visibility in generative engine responses by up to about 40 percent, while keyword stuffing ranked near the bottom. Brand campaigns still matter, but their effect should be checked against a new question: when buyers ask an assistant about your category, does the assistant know what you want it to know? The traffic that does arrive from AI answers rewards the effort, as the conversion studies in our review of the AI conversion data show: Adobe measured AI-referred traffic converting 42 percent better than non-AI, and Semrush found roughly 4.4 times standard organic.

The operational translation is a loop, and it is deliberately unglamorous. Define the questions your buyers ask. Run them across the engines buyers actually use, repeatedly. Read the answers, note which name you, which name rivals, and which sources each answer cites. Publish precise, citable pages for the questions you lose. Get them indexed, wait for retrieval to catch up, and recheck. Reachroller automates every step of that loop: tracked questions across five engines via official APIs, evidence-grounded scoring with branded questions excluded, a publish-ready fix page for each lost question with slug, title, schema and indexing steps, and a recheck that confirms whether the answer flipped. The manual version costs a day a month; Starter costs $29.

Big statistics invite big overreactions, so three grounding notes. First, 94 percent used AI is a breadth number, and it includes light-touch use. The depth numbers, 55 percent comparing vendors and 47 percent building business cases, are the ones that justify budget, and they are strong enough on their own. Second, these are self-reported surveys, which capture what buyers say they did; self-report tends to overstate shiny behaviors somewhat, though the seven-month jump from 36 to 51 percent in G2's repeat measurement is the kind of trend self-report bias does not fabricate.

Third, AI influence does not eliminate the rest of the journey. Buyers still visit websites, still take demos, still negotiate. What changed is who makes the first cut. The answer layer now performs the triage that determined which vendors got a chance to run their playbook at all. Losing the triage is invisible and fatal; winning it feeds every downstream motion you already do well. That is the precise sense in which AI visibility, the subject of our full primer on AI visibility, decides deals you never see.

The 2026 buyer studies will be revised, and next year's percentages will differ. But direction, magnitude and mechanism all point the same way, from two independent research houses. The shortlist moved into the machine. The only strategic question left is whether your brand is in the answers when it forms.