The New B2B Funnel: Half of Buyers Now Start in a Chatbot

Every B2B funnel you have ever drawn assumed the buyer starts with a search and a click. That assumption just expired. Here is what the funnel looks like now, stage by stage, and how to become the name it returns.

Do B2B buyers really start research in a chatbot?

Yes, and the speed of the shift is the story. In March 2026, a G2 buyer-behavior survey, reported in Profound's coverage, found that 51 percent of B2B buyers now begin their research in an AI chatbot more often than on Google. Eleven months earlier, in April 2025, that figure was 29 percent. In under a year, chatbot-first research went from a minority behavior to the majority one. The same survey found 71 percent of B2B buyers use AI chatbots somewhere in their research, even if they do not start there.

Sit with what that means practically. The buyer who used to type "CRM implementation consultants" into a search box and skim results now types a paragraph into ChatGPT: their industry, their team size, their budget constraints, their timeline, and a request for a shortlist. The machine answers with names and reasons. No impressions were served. No snippet was clicked. The entire awareness and consideration phase, the part your marketing budget was built to influence, happened inside a single synthesized reply.

The old funnel is missing its top

The chatbot shift did not happen in isolation. It landed on top of a search experience that was already keeping users away from websites. In the twelve months after Google launched AI Overviews in May 2024, zero-click searches grew from 56 percent to 69 percent, and AI Overviews now appear on roughly a quarter of searches. Even the buyers who still start on Google increasingly get an assembled answer instead of a list of doors to knock on.

So the classic funnel is being squeezed from both ends. Chatbots absorb the buyers who no longer start with search, and AI Overviews absorb the clicks of many who do. The result is a funnel whose top two stages, awareness and consideration, now happen largely out of your sight and beyond your ad budget's reach. We wrote about the shape of this transition in from blue links to spoken names: the unit of visibility is no longer a ranked page, it is a name the machine is willing to say.

Small stream, serious buyers

Skeptics point at traffic reports and note that AI referrals remain a sliver of total visits, and they are right about the sliver. Conductor's 2026 benchmarks put AI referral traffic at around 1.08 percent of all website traffic across ten industries. But the same benchmarks show why that sliver deserves outsized attention: ChatGPT referrals convert at 14.2 to 15.9 percent, Perplexity at roughly 10.5 percent, and Claude at up to 16.8 percent, against about 1.76 percent for Google organic.

The explanation is simple once you picture the buyer. Someone arriving from an AI answer already asked their question, already received a recommendation, and already read the reasons. They are not browsing, they are confirming. And the volume behind this behavior keeps growing: ChatGPT reached roughly 900 million weekly active users by February 2026, up from around 400 million a year before, and it drives about 87.4 percent of all AI referral traffic. The economics of a mention are the subject of the AI referral economy, but the short version is this: one named recommendation is worth a pile of impressions.

The funnel, then and now

What the machine does with your name at each stage

The stage that deserves your obsession is the shortlist. When a buyer asks for specific recommendations, the engine reaches for entities it can defend: people and firms with consistent identities, attributable expertise, and independent references. It is composing an answer it must justify, which is why it favors names that credible third parties have already vouched for. We documented a clean example of the mechanism in the day a machine recommended a stranger: the person who got named was not the biggest brand in the category, but the most legible one.

Validation is the sleeper stage. Buyers increasingly run a second prompt on each shortlisted name: what is their reputation, what do clients say, what are the risks. Whatever the machine finds becomes your reference call. If it finds specific testimonials, interviews and a consistent record, you pass. If it finds thin marketing pages and a two-year-old bio, you quietly drop off a shortlist you never knew you were on. This is also how hiring committees now vet executives, and the defense is identical: give the machine something true and substantial to find.

How do you rebuild a funnel for chatbot-first buyers?

1. Map the prompts, not the keywords. Write out the fifteen to twenty questions your buyers would actually type into a chatbot at shortlist stage, in full sentences with context. These are your money queries.
2. Run the audit. Put every query through ChatGPT, Gemini and Perplexity monthly. Log who gets named and why. The gap between those answers and your name is your real pipeline problem, stated precisely.
3. Become citable at the shortlist stage. Publish deep, opinionated, ungated work on the exact problems in your query map, attributed to a real named person, so the engine has something specific to quote when it justifies naming you.
4. Stock the validation stage. Earn reviews, journalist quotes and podcast interviews that live on crawlable third-party pages. The machine treats these as votes, and it counts them at the exact moment the buyer asks whether to trust you.
5. Instrument the last mile. Segment AI referrals in your analytics, add "how did you hear about us" to every intake form, and train whoever answers the phone to log mentions of ChatGPT or Perplexity. You cannot manage a channel you refuse to measure.

What does this mean for your sales team?

Buyers now arrive later and warmer. When the first human conversation happens after the machine has already framed the category, assembled the shortlist and summarized your reputation, the job of sales shifts from persuasion to confirmation. The teams adapting fastest do three things: they ask every inbound lead what the AI told them, they fix the gaps between that answer and reality, and they stop spending to interrupt buyers at a funnel stage that no longer exists. The full engineering of that position, identity, knowledge and regard, is what our services are built around.

The funnel did not die. It moved indoors, into a window you cannot buy your way into but can absolutely earn your way into. Half your market is already asking. The only question is whose name comes back.