51% of B2B Software Buyers Now Start Research in a Chatbot. Your Analytics Sees None of It.

G2's 2026 survey, "The Answer Economy: How AI Search Is Rewiring B2B Software Buying," puts a number on something most GTM teams have felt but couldn't quantify: 51% of B2B software buyers now start their research with an AI chatbot instead of a search engine. (G2, via Demand Gen Report, 2026)

The follow-on numbers are the part that should worry a GTM leader more than the headline stat. Sixty-nine percent of buyers ended up selecting a different vendor than they had originally planned, based on what the chatbot told them. A third bought from a vendor they hadn't previously heard of. Eighty-three percent said they felt more confident in their final decision because of the chatbot research. Seventy-one percent now use chatbots regularly for this kind of research, up from 60% previously, and 53% think chatbot research is more productive than a traditional search - up from 36%. ChatGPT alone accounts for 63% of that usage. (G2, via Demand Gen Report, 2026)

None of those interactions - the ones that changed a third of buyers' final vendor choice - produced a UTM parameter, a cookie, a website visit, or an account-scoring event. That is the AI dark funnel, and it is not the same problem your dashboard was already ignoring.

Why this "dark funnel" is a different animal

The original dark funnel was buyers browsing anonymously - real interest, just untracked. It was neutral: no one was pushing them toward or away from anyone.

The AI dark funnel is not neutral. When a buyer asks a chatbot "who are the top vendors for revenue marketing consulting," the model produces an answer that actively favors specific companies over others, inside a closed environment that generates no trackable signal at all. Being absent from that answer isn't invisibility - it's functional exclusion from a third of the vendor swaps that just happened in the market. The AI dark funnel behaves like a curated shortlist you don't get to see, let alone contest.

That distinction matters for what you do next. A neutral dark funnel is a measurement problem - track harder and you close the gap. A curated, opinionated one is a positioning problem - no amount of better tracking gets you into an answer you weren't cited in.

Why this isn't the traffic story Nukipa already told

Nukipa covered AI referral traffic in August: it's roughly 1% of B2B sessions but converts 5 to 23 times better than Google organic, and most dashboards hide it because it isn't tagged as a distinct channel. (^[1]) That post is about the tail end of the AI-mediated journey - the click that does happen, arriving from a citation in an AI answer, landing on your site with a trackable session.

The dark funnel is everything upstream of that click. It's the 51% of buyers who start in a chatbot, form a shortlist, narrow to two or three vendors, and only then - if at all - visit a site directly by typing the URL, generating a session with no referrer that your analytics quietly buckets into "direct traffic" alongside everyone who just remembered your name. The AI search traffic measurement problem and the dark funnel problem are sequential, not the same: one measures what you can already partially see; the other explains why so much of the decision happened somewhere you can't see at all.

What still leaks through, and what to do with it

You cannot instrument your way out of a closed model's chat window. But two things do leak signal, and both point at the same response.

Comparison and review-site content wins disproportionate share of AI citations. Nukipa's own search data shows exactly this pattern on comparison queries: position 4 to 6, zero clicks - because the AI answer is quoting the page directly in-chat rather than sending a click. (^[2]) That is dark-funnel behavior you can actually see the edges of: rankings you hold with no corresponding traffic. Comparison and review-site content is disproportionately what AI models cite when a buyer asks "how does X compare to Y" - exactly the moment 41% of buyers say they use chatbots for. (G2, via Demand Gen Report, 2026)

Persona-specific visibility gaps are measurable even when individual sessions aren't. You can't see the buyer who asked ChatGPT about you and left without a click. You can measure, systematically, whether your brand shows up when a model is asked the questions that persona would ask - and where a competitor shows up instead. That's a standing measurement practice, not a one-off audit: run it on a cadence, segmented by persona and buying stage, and track the trend rather than a single snapshot.

The response: publish for the answer, not the click

Since you can't track the dark funnel directly, the only lever left is upstream: increase the odds you're the answer the model gives, on the questions your buyers are actually asking, before they ever reach a trackable moment.

That means three concrete things, none of which are a tracking pixel:

Ungated, plainly-written content addressing buyer questions across every persona in the buying group, not just the champion who eventually fills out a form. Gated content cannot be cited in an AI answer if the model can't read it.

Comparison and category content that names competitors directly, matching the exact framing buyers use with chatbots ("X vs Y," "best tools for Z"). This is the highest-leverage citation surface identified above, and most B2B teams still avoid writing it out of habit, not strategy.

Sustained publishing cadence over one-off campaigns. Models retrain and re-index continuously; a single cornerstone piece from eighteen months ago loses citation share to whoever published something fresher and more specific last month.

The point

The dark funnel used to be a rounding error in your attribution model. At 51% of buyers starting research in a chatbot and 69% changing their vendor pick because of it, it is now the majority of the decision, happening somewhere your stack cannot see, shaped by an answer that is actively choosing winners.

You will not fix this with better UTM discipline. You fix it by being the citation the model reaches for before the buyer ever lands on a trackable page - which means treating AI visibility as a standing measurement and content practice, not a campaign.

Nukipa tracks exactly this: which questions your buyers are asking AI models, whether you're the answer, and where the gap against competitors sits - grounded in your own account and pipeline data rather than a generic visibility score. If you want to see where your dark funnel actually stands, test Nukipa.