“We’re not losing more deals.” If you’ve sat through a quarterly pipeline review lately, you’ve heard that sentence, and you’ve probably said it.
Win rates are roughly where they were. The proposal-to-close ratio hasn’t collapsed. Nothing on the dashboard is visibly bleeding.
But there are accounts you’d normally have been called into, and there’s no record of anyone calling.
I used to reach straight for the deck when this happened. Bad test score, so go buy nicer pens.
The catch is that if you’re losing before the shortlist exists, no amount of deck polishing moves the number. So somebody went and asked 1,076 buyers what actually happens in that gap.
”Nothing Broke. So Why Are We Missing From Deals?”
Here’s the short answer. 69% of B2B software buyers say they chose a different vendor than they’d originally planned, based on advice from an AI chatbot. And 33% — one in three — bought from a vendor they hadn’t known about before.
The research is G2’s, published on 15 April 2026. G2 is the software review site, which matters later, so hold onto that.
They surveyed 1,076 B2B software buyers and decision-makers in March 2026, across North America, EMEA, and APAC, from individual contributors up through VP and above. They also interviewed 39 B2B software marketers alongside it.
Where People Start Looking Changed. That Part Isn’t The Story.
Start with the setup numbers, because they’re the ones everyone quotes.
51% say they now more often begin software research with an AI chatbot than with Google. A year earlier, in April 2025, that was 29%. Twenty-two points in twelve months.
71% use AI chatbots somewhere in their research (60% seven months before), and 53% call it more productive than traditional search (up from 36%). ChatGPT leads the pack at 63%.
Fine. But this is still the boring half. If all that changed is where people look things up, the same names should keep winning — you’d just be found through a different door.
The 33% Belongs To Whoever Isn’t Famous Yet
The interesting half is that the names changed too.
For as long as anyone’s measured it, the B2B shortlist has been more or less bounded by prior awareness. You pick from the handful of vendors you can already name, and everyone else is out before the evaluation starts. Not rejected — never considered.
That’s the wall the 33% goes through. When something else assembles the candidate list, there’s suddenly room to enter from outside the reader’s memory. Which makes it a number for the challenger, not the incumbent.
And the endorsement seems to carry weight: 85% say they view a vendor more favorably once an AI chatbot brings it up. So getting named isn’t just one more impression. It lands closer to a recommendation from a third party than to an ad.
The pace picked up too. Four in five say AI chatbots made their purchase decisions faster, and 83% say they felt more confident in the final call.
If You’re On The List, Check What The Sentence Actually Says
Now the uncomfortable one. In the same survey, 64% say AI chatbot recommendations are inaccurate often or very often.
You can read that as “so the AI can’t be trusted”, and plenty of people will. But that reading leaves you with nothing to do on Monday.
Here’s the other reading, and I think it’s the useful one. The place where a third of buyers let in a company they’d never heard of is the same place where two buyers in three say they keep hitting a wrong description.
Which means getting dropped on a wrong description could be happening at roughly the same rate as getting in on a right one. That’s the actual exposure, and it’s the part nobody’s watching.
If you don’t know what’s being said about you there, this kind of loss doesn’t even show up as a loss. Nobody called. There’s no lost-deal report for a call that never happened.
How Much Should You Discount All This?
Three caveats, and they’re real ones.
G2 runs a software review site. One finding in particular — that 45% named citations from software review sites as the most reassuring signal — is a question the publisher has money riding on. Discount that one hard.
It’s also a self-reported survey, not purchase logs. “I chose differently because of the AI” is a memory being reconstructed after the fact, and people are generous to themselves about why they did things.
And the sample spans North America, EMEA, and APAC. It isn’t any single market’s number, so don’t drop it onto your own without a shrug.
Maybe I’m giving one vendor’s survey more weight than it’s earned. But 69% and 33% aren’t the kind of margins that flip sign when you discount them; they just get smaller.
Write Ten Comparison Questions, Then Count How Often You Come Up
AI visibility looks like an awareness project. On this data it’s a pipeline project, and it sits upstream of everything sales touches — because whether you’re on the list gets settled before the first conversation.
And your awareness tracker can’t see it. “They didn’t know us” and “they knew us and didn’t list us” produce the same flat line in a monthly deck, which is exactly why it’s gone unnoticed this long.
So, one thing to do. Write down ten comparison questions your buyers actually type — not “best CRM”, but the real sentence with the conditions in it (“CRM for a 40-person sales team that has to sync with NetSuite”). That list is worth more than the measurement it feeds, and you can draft it this afternoon.
Then run them monthly and count what share of the answers name you. A single reading tells you nothing; the shape over a few months is what you’d actually take to a budget conversation.
If you want the other half of this — what the people who arrive through AI do once they’re on your site — that’s the piece on why AI traffic buys better than it looks. Good next stop.
Sources
- G2, “New G2 Research: Half of B2B Software Buyers Now Start Their Research with AI Chatbots”, 15 April 2026, prnewswire.com (survey of 1,076 B2B software buyers and decision-makers conducted March 2026 across North America, EMEA and APAC, spanning individual contributors through VP and above, plus interviews with 39 B2B software marketers. 69% chose a different vendor than originally planned based on AI chatbot advice; 33% purchased from a vendor they had not previously known; 51% more often start software research with an AI chatbot than with Google, against 29% in April 2025; 71% use AI chatbots for research, against 60% seven months earlier; 53% find them more productive than traditional search, against 36%; ChatGPT is the most-used assistant at 63%; 85% view vendors mentioned by AI chatbots more favorably; four in five report faster purchase decisions and 83% greater confidence in the final decision; 64% say AI chatbot recommendations are inaccurate often or very often; 45% name citations from software review sites as the most reassuring signal. Limits: the publisher operates a software review site and has a commercial interest in the review-citation finding; the data is self-reported survey response rather than purchase logs; the sample spans North America, EMEA and APAC rather than any single national market)
