Half of B2B Software Buyers Now Start in a Chatbot. Your Funnel Did Not Notice.
Your CRM says the buyer arrived from a branded search, downloaded a comparison guide, and booked a demo. It does not say that two weeks earlier they asked ChatGPT which three vendors to consider, were told about two you had never heard of, and dropped you from the shortlist before your first touch.
That gap is now the default case, not the edge case.
What the data says
- 51% of B2B software buyers begin their purchasing research inside an AI chatbot rather than a search engine — up from 29% in April 2025 (G2, survey of 1,076 buyers).
- 63% of those name ChatGPT as their primary research tool.
- 69% say the chatbot’s guidance led them to choose a vendor different from the one they had in mind.
- AI chatbots are the single most influential source for shortlists at 54%, ahead of review sites (43%) and vendor websites (36%) in one survey; a later G2 report puts chatbots at 37% versus review sites at 38% — a statistical tie, which is itself the point.
- 47% use AI to build the internal business case before contacting a vendor at all.
- Forrester’s buyer survey of ~18,000 business buyers found 94% used generative AI somewhere in their most recent purchase.
Two readings are worth keeping in tension. The surveys disagree on whether chatbots have overtaken review sites. What they agree on is that a chatbot is now in the room for essentially every shortlist decision — and that it can change the list.
Why your attribution cannot see it
Four reasons, and none of them are fixable by adding a UTM.
1. The first contact moved. When research starts in a chat interface, the “first touch” your CRM records is already the second or third touch of the actual journey.
2. AI visits look like direct. ChatGPT often sends traffic with an unclear or stripped referrer. The session lands in Direct, and the last-click model attributes the deal to whatever came after.
3. The shortlist is set before you are measurable. If the chatbot shaped a three-vendor shortlist and you were not on it, there is no session to record — the loss is invisible rather than negative.
4. Sales hears about it late. By the demo, the buyer has already been told what your product is and what it costs. Reps then spend the call correcting the model’s version instead of selling.
What actually gets a vendor onto an AI shortlist
Three properties, based on how these systems select:
| Property | What it looks like in practice |
|---|---|
| Attribute completeness | Pricing model, deployment, integrations, language support, use cases all published in text, not locked in a PDF |
| Comparison eligibility | Named, structured comparison pages that a model can cite for “A vs B” and “alternatives to A” |
| Independent corroboration | Third-party mentions, reviews, and coverage that confirm your own claims |
| Entity clarity | One consistent description of what you are, repeated across sources |
Note what is not on the list: ad spend. There is no bid in these systems. The selection is closer to editorial than to auction, which is uncomfortable for teams whose whole model is bidding.
The content that survives a chatbot shortlist
Buyers ask sequence-shaped questions. Content should map to that sequence, not to a keyword list:
- “What tools exist for X?” — a category page that names real competitors honestly.
- “Which of these fits a company like mine?” — segmentation by size, budget, and use case.
- “A vs B?” — comparison pages with a table and a verdict, not a both-sides waffle.
- “What does it cost?” — actual pricing bands, published.
- “What are the risks?” — limitations stated plainly. This is the section that builds trust and the one most brands omit.
The last item matters more than it sounds. A model summarizing “what are the downsides of X” needs a source that admits downsides. If only complaints on Reddit are available, that becomes the answer.
Fixing the measurement, roughly
Perfect attribution does not exist here. A working approximation:
- Ask the buyer. One field on the demo form: “Where did you first hear about us?” Add “AI assistant” as an option. Expect 15–25% to select it.
- Track AI-referred sessions as a segment, and compare their demo-request rate to organic.
- Audit the shortlist directly. Monthly, run your five most important buying prompts in ChatGPT, Gemini, and Perplexity and record which vendors are named. That list is your real competitive set.
The takeaway
The B2B funnel did not get shorter; it got less observable. Buyers now do the first two-thirds of the work in a chat window, and arrive pre-informed, pre-biased, and pre-priced.
Content that answers the sequence — honestly, with published pricing and stated limitations — is no longer a marketing nice-to-have. It is the input to the recommendation that decides whether a rep ever gets a call.
Building that library at the depth a shortlist requires, across markets and languages, is the operational half. SEONIB publishes structured comparison, pricing, and product-detail pages in grouped multilingual sets, so the version a buyer’s assistant reads in their own language is the same version your sales team is quoting.
Share Article