Half of B2B software buyers now start their research with an AI assistant, not Google. Here is what the shortlist data shows.

The line was crossed in the last twelve months

G2's Answer Economy study, run in March 2026 across 1,076 B2B software decision-makers, found that 51 percent now begin purchasing research with an AI chatbot more often than with Google. In G2's April 2025 buyer survey the equivalent figure was 29 percent. Seventy-one percent said they rely on an AI assistant at some point in their research, 61 percent use AI search alongside Google rather than instead of it, and 53 percent said research done with a chatbot is more productive than traditional search, up from 36 percent seven months before.

The figure that should concern a vendor more than adoption is what the assistant does to the shortlist. In the same G2 study, AI chatbots ranked as the top source influencing shortlist creation, ahead of review sites, analyst firms and vendor websites. Sixty-nine percent of buyers said they chose a different vendor than they had initially planned because the chatbot recommended it, and one in three bought from a vendor they had never previously heard of. Eighty-five percent said they think more highly of a vendor when an AI names it in an answer.

The counterweight: buyers do not trust the answer blindly

Gartner surveyed 645 B2B buyers in August and September 2025 and presented the findings in May 2026. Forty-five percent had used generative AI in a recent purchase, primarily to research vendors and products, which is a lower adoption figure than G2's and a reminder that the population surveyed matters: G2 asked software buyers specifically, Gartner asked B2B buyers generally. More telling, 69 percent of Gartner's respondents said they prefer to validate AI-generated insights with a sales representative, and buyers were almost evenly split on where misleading information is more likely to come from: 51 percent said generative AI, 49 percent said sales reps. In G2's study, 64 percent reported encountering inaccurate AI recommendations often or very often.

So the honest picture is not that AI has replaced diligence. It is that AI now writes the first draft of the shortlist, and humans then verify it against reviews, peers and, late in the process, sellers. G2's respondents named citations from software review sites as the single most confidence-inspiring signal inside an AI answer. Verification flows through the same off-page sources the assistant read in the first place.

Why the first draft is close to the final draft

6sense's 2025 Buyer Experience Report, built on nearly 4,000 responses globally, measured what happens before a vendor knows the deal exists. Buying groups made first contact with sellers 61 percent of the way through their journey. Ninety-five percent of eventual winners were already on the buying group's day-one shortlist, and 80 percent of deals were won by the vendor the group favoured before any contact was made. The same study put the buying group at ten or more people evaluating around five vendors.

Put the two datasets together and the mechanism is uncomfortable for any vendor that relies on being found late. The shortlist is formed early, it is formed increasingly in conversation with an assistant, and by the time a seller is contacted the decision is four-fifths made. A vendor that is not in the AI's answer is not losing the deal at the demo. It is absent from the process entirely.

Where the evidence is thin

Three caveats, stated rather than buried. These are surveys of self-reported behaviour, not observed traffic, and self-report overstates new habits. G2 sells review-site presence and has a commercial interest in the answer-economy thesis, which does not make the data wrong but does make the Gartner and 6sense figures useful cross-checks from firms with different incentives. And absolute AI-referred traffic to most vendor sites remains small; the case rests on the growth rate and on the winner-take-most shape of a synthesised answer that names three to five products where a results page listed ten or more.

What a vendor should do, in order of payback

1. Measure whether you are named

Take the twenty questions your buyers actually ask, of the form "best X for Y-sized companies" and "X versus Z", run them across three assistants, three times each, and record who is named and which sources are cited. Most vendors have never seen this table. It takes an afternoon, and our free B2B audit records it for your category as standard.

2. Fix what the engines read

Because assistants assemble answers largely from off-page material, the work is mostly unglamorous: a steady, honest review base on the sites your category's answers cite; comparison and alternative pages that state plainly who the product is for and who it is not for; documentation and pricing pages that can be quoted accurately; consistent product naming everywhere. This is what our AI visibility, SEO and local search work covers for software companies.

3. Sharpen the category sentence

An assistant can only recommend a product it can classify. If your own materials describe the product five different ways, the engine will pick one, and possibly the wrong one. A tight definition of category, buyer and use case, the output of proper Positioning and ICP work, is now machine-readable positioning as much as it is messaging.

None of this guarantees inclusion in any answer, and we would distrust anyone who claims it does. What it does is put a vendor's evidence where the shortlist is now being drafted, and give the owner a measurement instead of a guess.