Most B2B software buyers now let AI agent shortlists, the ranked vendor lists a bot builds on its own, make the first cut before a human vendor ever hears about them. G2 says 82 percent sourced recommendations from AI chatbots, and Gartner expects agents to run a quarter of enterprise software purchases with no human involved by 2028.
Key takeaways
- 82 percent of B2B software buyers used an AI chatbot for recommendations in the past two years, and half say AI’s biggest value is narrowing the field, according to G2.
- Review sites (38 percent) now rival AI chatbots (37 percent) as the top influence on shortlists, in the same G2 survey of over 1,000 buyers.
- Only 9 percent of buyers are comfortable letting an agent execute a purchase, and just 2 percent allow a fully autonomous buy, even though 61 percent already use or plan to use agents somewhere in the process.
- Gartner predicts AI agents will intermediate more than $15 trillion in B2B spending by 2028, with one in four enterprise software purchases made without a human in the loop.
How real are AI agent shortlists in B2B software buying right now?
They’re already the default research step, not some future scenario. G2’s July 2026 survey of more than 1,000 B2B software buyers found 82 percent had sourced a software recommendation from an AI chatbot in the past 24 months, and half of those buyers said AI’s biggest value was narrowing and comparing options, not just spitting out a first list. Fifty-one percent are using agents specifically to build shortlists or run cost comparisons.
IDC found nearly the same pattern in August 2026: roughly 80 percent of B2B technology buyers now use AI agents somewhere in the purchasing process, per coverage from MarketScale. Two firms, two surveys, the same range. That’s not noise.
Here’s the part that surprised me: review sites aren’t losing ground to AI chatbots the way you’d expect. G2 says review sites influence 38 percent of shortlists against 37 percent for AI chatbots, basically a tie. Agents seem to be reading the review sites, not replacing them.
If agents already narrow the list, why won’t anyone let them buy?
Because evaluating and committing aren’t the same risk, and buyers know it. G2 found just 9 percent of buyers are comfortable letting an agent execute a purchase, and only 2 percent allow a fully autonomous buy with no review at all. Finance has only gotten more involved, not less: CFO or finance involvement in software deals jumped from 31 percent to 46 percent year over year, and 46 percent of buyers reported an outright CFO veto on a deal already in motion.
So you get a strange split. Research is already agentic. Signing isn’t, and it’s getting more cautious, not less. I wouldn’t bet on that gap closing fast.
Gartner’s still willing to put a date on it. Don Scheibenreif, a Gartner analyst, called the shift a move from copilots to autopilots, and Gartner’s forecast, reported in November 2025, puts one in four enterprise software purchases in the hands of an agent, no human sign off, by 2028. That’s a prediction, not a measured fact today, and it’s worth treating it that way.
What gets a software vendor cut before a human ever sees the shortlist?
An agentic shortlist is really a data matching exercise. The agent pulls from review profiles, documentation, pricing pages, and whatever structured data it can parse, then ranks what’s complete and consistent higher than what isn’t. A vendor with a thin or outdated review profile, vague pricing, or documentation that only makes sense to a human reading it slowly is easy to drop silently. Nobody sends a rejection email. The vendor just doesn’t show up on the list a person eventually sees.
That’s the real shift, and it’s not what most martech coverage leads with. The exclusion already happened by the time a human buyer feels ready to say yes. If your product page, review presence, and pricing page weren’t written with a parsing agent in mind, you’re not losing deals at the demo. You’re losing them weeks earlier, invisibly. We looked at a related piece of this in AI visibility tools versus earned media: the same logic applies on the buying side, not just discovery.
| Buying stage | Share of buyers using AI there |
|---|---|
| Sourcing initial recommendations | 82 percent |
| Narrowing and comparing options | 50 percent |
| Building shortlists or cost analysis | 51 percent |
| Executing the purchase itself | 9 percent |
| Fully autonomous purchase, no review | 2 percent |
Source: G2 2026 Buyer Behavior Report.
What should a martech or SaaS vendor do about it?
Structure your public data for a machine reader, not just a human one.
- Audit your G2, Capterra, and comparable review profiles for completeness, not just star rating. An agent reading a thin profile treats missing fields as missing capability.
- Put pricing logic somewhere a machine can parse it, even in general terms. A vague “contact us” page is a worse signal to an agent than it ever was to a person.
- Keep documentation and integration specs current. We’ve covered why free AI tiers without the budget behind them hurt vendors on revenue targets; thin documentation is the quieter version of the same problem.
- Track where qualified pipeline actually comes from. If chatbot and review site referrals aren’t showing up in your attribution, you’re probably undercounting how many deals you already lost before the first call.
FAQ
What is an AI agent shortlist? It’s a list of vendors an AI tool builds on a buyer’s behalf by comparing product data, reviews, and pricing, before a human researcher looks at any of it directly.
Are AI agents actually buying B2B software without people involved? Not yet in any real volume. G2 found only 2 percent of buyers allow a fully autonomous purchase today. Gartner’s one-in-four-by-2028 figure is a forecast, not a current measurement.
Why do review sites still matter if buyers use AI chatbots? Because the agents read the review sites too. G2 found the two are nearly tied as shortlist influencers, 38 percent versus 37 percent, so your review profile is now an input to the agent, not a separate channel.
What’s the single biggest risk for a software vendor here? Getting excluded from a shortlist silently, before a salesperson ever gets a call, because product data was too thin or inconsistent for an agent to rank it fairly.
Sources: G2’s 2026 Buyer Behavior Report (July 2026, over 1,000 B2B software buyers); IDC’s buyer research as reported by MarketScale (August 2026); and Gartner’s B2B agentic spending forecast as reported by Digital Commerce 360 (November 2025).