“Does Anybody Know Why We Weren’t On The List?”
That question hung in the air once, in a loss review, and then somebody moved on to the next slide.
We’d been cut before the comparison table existed, so the sales side had nothing to report; you can’t lose a deal you were never in, and you can’t debrief a conversation you weren’t part of. I said something about price (which is what I say when I have nothing), mostly to have said something.
You’ve sat in that meeting, haven’t you.
And the trouble with “buyers start somewhere else now” is that it’s easy to nod at and impossible to spend money on. It doesn’t say which stage it touches. So here’s a survey that asked the buyers directly, and it puts numbers on exactly that stage.
Is AI Actually Changing Which B2B Vendors Get Chosen?
It is. Of 1,076 B2B software buyers and decision-makers, 69% chose a vendor other than the one they had originally intended to go with, after AI guidance (G2 ran the study in March 2026 and published it on 15 April). And 33% bought from a vendor they’d never heard of.
The panel spans North America, EMEA and Asia-Pacific, from individual contributors up to the executive level, and everybody in it either holds purchasing authority or influences it. So their answers come from people who sign the purchase order or shape who signs it. Before the switching comes a change in where buyers start reading:
- 51% now more often begin their research with an AI chatbot than with Google (29% in April 2025)
- 71% lean on AI chatbots for software research, against 60% seven months earlier
- 63% use ChatGPT for B2B research, and 41% reach for a Deep Research-style feature routinely
If that were the whole story, it would be a story about surface area. One more place to show up in, one more line in the reporting, and the existing playbook stretches to cover it.
But the 69% and the 33% are a different animal. They say the destinations changed, not the routes.
The Stage That Moves Is The Shortlist
85% say they view a vendor more favorably when a chatbot names it.
Favorability of that kind is what gets lost when AI visibility is filed under “awareness”, because awareness is about recall — somebody who already knows the name retrieving it again later. A shortlist is a fixed number of slots, three or five, decided in one pass, and the only question it asks is whether the name is in them.
Miss the slots and there’s no comparison table, no approval memo, nothing. Nobody records an absence. There’s no artifact to review afterwards. That was the meeting I opened with.
The buyers’ side of it moved too: 80% say they now decide faster, and 83% say they’re more confident in the final choice. Deals close faster, and buyers feel surer. I doubt much room is left for a latecomer to argue its way back in.
The Case For Reading Those Answers On A Schedule
64% of the buyers run into inaccurate recommendations “often” or “very often” — the one figure in the study that reads like a case against all of this.
Inaccurate recommendations happen because the description a model holds about a company is stale, or has drifted in from another company. And the buyer can’t tell. They have one answer in front of them, not the trail behind it, and no obvious way to check it in the ten seconds they’re giving the question (and ten seconds is generous).
So the 64% is where measuring earns its keep. Go and read what’s being said about you, on a schedule somebody keeps, and the loss review stops being how you find out.
45% named the same reassurance above every other: a review-site citation inside the AI’s answer. Since G2 runs a review site, that’s the one number in the study pointing at its own author’s product — treat it as a claim the publisher had an interest in, not as a settled finding.
Put These Four Lines In The Footnote
I’d want the caveats traveling with the numbers if this goes into a deck.
- It’s a self-reported survey, not purchase logs
- The sampling method, the question wording and the country breakdown aren’t published, and the publisher states no limitations of its own
- The pool is global; there’s no Japan-only or single-market figure in it
- The year-on-year comparisons (“51%, up from 29%”) are against separate studies seven to twelve months earlier, not a tracked panel
The missing single-market breakdown is the one that bites in a budget meeting. A global average tells you which way the wind is blowing and nothing about your own street; present it as your street, and the first person who asks takes the whole slide down with one question.
What the survey is good for is the setup. I’d borrow the direction and measure the level myself — a smaller claim than the press release makes, and the one that survives being challenged.
Measure Whether The Name Is Still On The List
If one thing comes out of this survey, it’s what an AI visibility number should be counting. Recall was the old answer. Shortlist slots are the answer these buyers described. 69% and 33% are describing a change in the composition of shortlists, and an awareness metric can’t see that, because it was built to count people rather than slots.
In practice that means fixing a set of questions — the company name, the category, the problem it solves — and counting how often the name comes back in the answers, on a schedule somebody keeps. And reading what gets said when it does come back, since 64% of buyers are meeting bad descriptions somewhere, and I wouldn’t assume the descriptions of my own company are clean.
Then the question that comes round every planning cycle — “is the AI thing brand work or not” — answers itself. It’s shortlist work, so it belongs in the same budget line (the comparison-stage one, not the sponsorship one) as everything else that gets a company compared.
I’ve been round this survey once already, from the other end — the end where nobody has heard of you yet: Buyers Are Picking Vendors They’ve Never Heard Of. Thank ChatGPT.. That one is about where the opening came from, and it’s the one to read first. This one is about what to count once you’re through the opening, and how to hand the number to a finance person without losing the slide.
I didn’t think the front of the funnel had moved this far. Maybe I’m reading more into one survey than it can hold. But I’d rather find out by counting than by sitting through another loss review with nothing to say.
Sources
- G2, “New G2 Research: Half of B2B Software Buyers Now Start Their Research With AI Chatbots”, PR Newswire, 2026-04-15 (survey of 1,076 B2B software buyers and decision-makers across North America, EMEA and Asia-Pacific, from individual contributors to executives, all with purchasing authority or influence; fieldwork March 2026. 51% more often start research with an AI chatbot than with Google, against 29% in April 2025; 71% rely on AI chatbots for software research, against 60% seven months earlier; 63% use ChatGPT for B2B research and 41% routinely use Deep Research-style features. 69% selected a different vendor than originally intended after AI guidance and 33% bought from a previously unknown vendor; 85% view AI-named vendors more favorably; 80% report faster decisions and 83% greater confidence in the final choice; 64% encounter inaccurate recommendations often or very often, and 45% name a review-site citation in the answer as their main reassurance. Self-reported survey data rather than purchase records; the sampling method, question wording and country breakdown are not published and the publisher discloses no limitations; the pool is global with no single-market figures; year-on-year comparisons are drawn from separate earlier studies rather than a tracked panel. G2 operates a software review site.)
