How B2B Buyers Use AI to Research & Shortlist Vendors

Half of B2B buyers now start vendor research in AI chatbots. See how AI-mediated shortlisting works — and how to update your win-loss, VoC and buyer research.

How the AI-mediated buying journey actually works
The numbers reshaping vendor shortlists
What buyers actually ask AI — and what the model tells them
What this means for your win-loss analysis
Updating your buyer research and VoC methods
Using AI in your own research — without importing its blind spots
A research-first playbook to stay on the AI shortlist
FAQ

How the AI-mediated buying journey actually works

Most B2B journey maps still draw a straight line — awareness, research, evaluation, decision — with your website and sales team touching every stage. That picture is out of date. Discovery, comparison and evaluation now happen inside a chatbot, often before a buyer speaks to a single vendor. By the time someone fills in a contact form, the shortlist is usually already fixed.

Understanding how B2B buyers use AI to research vendors starts with that shift. Forrester’s 2026 buyers’ journey survey, based on responses from 18,000 buyers, found that 94% used AI during their most recent purchase. Fifty-five percent used it to compare vendors, 54% to research products, and 47% to build the internal business case they later presented to their own stakeholders.

Evaluation has now overtaken research as the longest stage in the journey. That reads as counter-intuitive until you see the mechanism: AI compresses weeks of manual research into a single session, so buyers move faster into evaluation and then linger there, testing claims and running scenarios with the model.

The stakes of that early moment are high. In 95% of purchases, the winning vendor was already on the buyer’s Day One shortlist, and the pre-contact favourite wins roughly 80% of deals. Miss that first list and you’re rarely catching up.

This is a research problem before it’s a marketing one. If a model is shaping the shortlist, you need to map the B2B buyer journey with a new, largely invisible stage built in — one that happens off your site, before your team knows a deal exists.

The numbers reshaping vendor shortlists

The shift shows up clearly in buyer-behaviour data from the past year. G2’s research found that 51% of B2B software buyers now start vendor research in an AI chatbot more often than in Google, up from 29% in April 2025 — the first time AI has overtaken search as the default starting point for B2B research.

Seventy-two percent of buyers use ChatGPT specifically to evaluate vendors, and it holds roughly 63% share of AI-assisted research sessions, with Perplexity, Gemini and Google AI Overviews gaining ground. Shortlists have got shorter too: the average has shrunk from around 3.2 vendor names to about 2.5, so each remaining slot carries more weight.

The consequences for vendor selection are significant. Sixty-nine percent of buyers chose a different vendor than originally planned because a chatbot pointed them elsewhere, and roughly one in three bought from a vendor they’d never heard of before that conversation. Despite this — or because of it — 83% say they feel more confident in their final choice with AI in the loop. Confidence, not just convenience, is driving adoption.

These are 2025–2026 figures. Model shares and buyer habits are moving quickly, so treat any snapshot as a moving target, not a fixed baseline.

What buyers actually ask AI — and what the model tells them

Buyer prompts follow a rough sequence tied to where they are in the journey. At discovery, it’s broad: “what tools do I need for [problem]?” At comparison, it narrows: “Vendor A vs Vendor B for an enterprise team.” At procurement, buyers ask the model to research a specific supplier — size, funding, red flags. At purchase, prompts get practical: ROI math, pricing justification, a narrative to take to their own stakeholders.

Whatever the model returns at that first prompt becomes the buyer’s initial vendor list. Being named at all has gone from a nice-to-have to a gating condition — you can’t be shortlisted for a comparison you’re never mentioned in.

There’s no sponsored slot to buy your way in. Over 85% of non-paid AI citations come from earned media and third-party sources, not vendor websites. Analyst reports, review sites, independent comparisons and press coverage feed the model far more than your own marketing pages do.

Models have preferences you can study, as well. They reward specific numeric claims over vague ones, comparisons structured in a way they can parse, and corroboration that sits outside your domain. Apply the rigour you’d bring to how to test AI agents — treat the chatbot as a fallible research subject whose reasoning, and occasional hallucination, you can observe and account for.

What this means for your win-loss analysis

Most win-loss frameworks were designed for a world with a traceable trail: first contact, demo, proposal, close. They weren’t built to capture a stage that happens off-record, inside a chatbot, before your team knows a buyer exists. Run your interview guide as-is and you’ll under-count both why you made it into the room and why you didn’t.

The fix is a small set of direct questions added to every win-loss interview: did you use an AI tool during your research? Which one? What did you ask it? Were we named, and how? What did it say about us relative to competitors?

Give particular attention to buyers who say they’d never heard of you, or had only heard of you through an AI conversation. That cohort explains a category of loss no sales rep ever touched, and standard win-loss guides miss it because they assume the buyer’s awareness came from a channel your team can see.

We’ve seen this play out directly in win-loss work. Once these questions go into the guide, a recurring pattern surfaces across engagements: losses to unfamiliar competitors that trace back to a single chatbot comparison run weeks before any vendor contact, built around criteria such as implementation speed, pricing tier and fit for company size that the losing vendor’s own materials never addressed in a form the model could cite.

That’s the value of the added questions. They tell you which criteria the model surfaced and whether the buyer trusted or verified them. Feed that into a win-loss analysis for B2B SaaS process and you get a signal your roadmap and positioning teams can act on, distinct from losses driven by a human evaluator’s own judgement.

Updating your buyer research and VoC methods

Buyer research needs the same refresh — and so does Voice of Customer (VoC) work more broadly. Committees don’t experience AI as a single voice: one person usually prompts the model, others act on what it returns, and the group negotiates how much weight to give its recommendation. Understanding who does which of those things, and how disagreement gets resolved, is now core to mapping how a decision actually happens.

Contextual inquiry — sitting with buyers while they work, rather than asking them to recall it afterwards — is the right method for this. Watch the prompts they type, the follow-up questions they ask when the first answer isn’t good enough, and the point at which they trust the model’s answer or go and verify it elsewhere. This is exactly the kind of moment-by-moment observation you’d use to run a contextual inquiry for any other high-stakes decision.

What you learn feeds straight into positioning and messaging. If a claim needs to be parseable and corroborable by a model to reach a buyer at all, testing whether your claims meet that bar is now part of the research remit, not a separate technical exercise.

Pay close attention to the exact language buyers use when prompting the model — their words, not your marketing’s. That raw vocabulary is some of the most useful material you’ll get for category framing, because it reflects how buyers actually think about the problem, not how you’ve chosen to describe the solution.

Build this into your standing B2B buyer research methods rather than a one-off audit. Prompts and model answers will keep shifting as tools change — treat it as a recurring VoC input, refreshed on the same cadence as any other market signal.

It’s a reasonable instinct to run your own win-loss transcripts and VoC data through the same models your buyers use to research you. Do it — but with guardrails, because the failure modes that shape buyer shortlists can just as easily distort your own analysis.

Models flatten nuance. They over-index on vendors and themes already well covered in their training data, and they can fabricate quotes or synthesise themes that sound plausible but don’t trace back to anything a real buyer said. None of that is malicious — it’s a structural property of how these tools generate text — but it means you can’t take model output as findings.

Triangulate anything the model surfaces against the original transcripts. If a theme looks strong in a summary but you can’t find three or four interviews that actually support it, treat it as a hypothesis, not a finding. Keep a human reviewer in the loop for anything decision-critical: roadmap calls, positioning changes, messaging that will go external.

Document which prompts and model versions you used for any AI-assisted synthesis. Skip that step and your findings aren’t reproducible — and you won’t know later whether a shift in results reflects real change in buyer behaviour or just a model update. We cover this in more detail in how we validate AI-generated research insights.

A research-first playbook to stay on the AI shortlist

Start by auditing the prompts that matter for your category — the discovery, comparison and procurement questions a real buyer would ask. Run them across the major models and record, systematically, whether you’re named, how you’re described, and who you’re compared against. This gives you a baseline you can track over time rather than a one-off snapshot.

Turn what you learn from win-loss and VoC work into claims a model can actually cite: specific, numeric, and corroborated by a source outside your own domain. Research isn’t just informing positioning here — it’s feeding the visibility that gets you onto shortlists at all.

Because non-paid AI citations skew so heavily towards earned media, prioritise coverage, analyst mentions and independent comparisons over owned content. That’s where the corroboration models pull from actually lives.

Set a monitoring cadence — quarterly is a reasonable starting point — to re-run key prompts as tools, market share and model behaviour shift. Feed what changes back into positioning research, so messaging updates are grounded in what’s actually happening on the shortlist, not assumption.

Make AI-shortlist evidence a standing item in your research readouts, alongside the rest of your product research hub findings. Product, marketing and sales teams need to work from the same picture of how buyers find and judge you — otherwise each team keeps optimising for a journey stage that no longer exists in isolation.

FAQ

Do B2B buyers really trust AI to choose vendors? Not blindly, but enough to let it shape the outcome. Eighty-three percent of buyers feel more confident in their final choice with AI in the loop, and 69% switched from their originally planned vendor after a chatbot pointed them elsewhere. Most still verify what the model tells them, but that verification happens after the model has already set the shortlist — the stage your research needs to account for.

Can you pay to appear in AI vendor recommendations? No. There’s no sponsored slot in a chatbot’s answer. Over 85% of non-paid AI citations come from earned media and third-party sources rather than vendor websites, so inclusion has to be earned through corroborated, machine-readable evidence — analyst coverage, independent reviews, press mentions. Much of that evidence is exactly what rigorous buyer and market research is built to produce.

How do I update win-loss analysis for an AI-mediated buying journey? Add a small set of direct questions to your existing interview guide: whether the buyer used an AI tool, which one, what they asked it, whether you were named, and what it said about you relative to competitors. Pay particular attention to buyers who say they only heard of you through an AI conversation — that cohort explains losses no sales rep ever touched, and it won’t show up unless you ask for it directly.

About Glasgow Research — Glasgow Research helps B2B SaaS teams turn customer and market research into product decisions. Work with us.