The AI Dark Funnel: When 93% of AI Search Sessions Never Click, What Do You Measure?
About 93% of AI search sessions end with no click and 70–73% of the B2B buying journey now happens in the dark funnel. Why last-click attribution actively misleads, what teams are replacing it with, and three cheap instruments for small businesses.
- About 93% of AI search sessions end with no website click (Conductor, 2026), so the research phase leaves no analytics trace.
- 70-73% of the B2B buying journey now occurs in the dark funnel, which averages 38% of B2B pipeline.
- 94% of B2B buyers use LLMs during procurement research — standard behaviour rather than early adoption.
- Multi-touch attribution adoption reached 47% and marketing mix modelling 26%, both roughly tripling since 2023.
Roughly 93% of AI search sessions end without a single website click, according to Conductor's 2026 benchmarks, and 70% to 73% of the B2B buying journey now happens in the dark funnel (Mental Momentum Research, 2026). Your analytics can see the last 7%.
This is not a tracking bug you can fix with better UTMs. Conversations inside an AI assistant carry no referrer, set no cookie and fire no pixel. The research happened; you simply were not in the room.
What is the AI dark funnel?
It is every stage of the buying journey that unfolds inside an AI conversation, where tracking pixels, UTM parameters and referral data do not exist. A buyer asks an assistant to compare three vendors, gets a synthesised answer, forms a shortlist, and arrives at your site typing your brand name directly.
Your analytics records that as direct traffic. The comparison that actually decided the shortlist is invisible, and so is every competitor who was in it. Direct traffic used to mean someone who already knew you. Increasingly it means someone an AI told about you.
How widespread is the behaviour?
94% of B2B buyers now use LLMs during procurement and research (Mental Momentum Research, 2026). That is not an emerging segment to plan for. It is effectively the whole market.
Unsurprisingly, 65% of marketers cite AI-driven search changes as their single biggest challenge (Omnibound, 2026). The difficulty is less the loss of traffic than the loss of evidence — it is hard to defend a budget for influence you cannot demonstrate.
Why does last-click attribution now actively mislead?
Last-click was always a simplification. In an AI-mediated journey it becomes a distortion, because the model systematically credits the channel closest to the purchase and the dark funnel sits entirely upstream of that.
The predictable consequence: brand and content budgets look inefficient while branded paid search looks extraordinary. Teams then cut the upstream work that generated the demand and increase spend on capturing it, which works until the demand stops arriving.
If your branded search volume is growing while your content attribution is flat, that is not proof content is failing. It is close to a signature of a working dark funnel.
What are teams replacing it with?
Multi-touch attribution adoption has reached 47%, up from 31% in 2023, and marketing mix modelling 26%, up from 9% (Digital Applied, 2026). MMM is the more interesting jump: it is a statistical approach that never needed user-level tracking, which is exactly why it is returning.
For a small business, full MMM is overkill. Three cheaper instruments do most of the job. Self-reported attribution — a required "how did you hear about us" field on your form — captures what tracking cannot. Branded search volume works as a demand proxy. And geographic or temporal holdout tests give you causal evidence without any tracking at all.
Self-reported attribution is the one most often dismissed as unreliable. It is unreliable at the individual level and quite good in aggregate, which is the level you make budget decisions at.
How do you get visibility inside the AI conversation?
You cannot instrument it, so you sample it. Run your top 20 buying-intent questions through the assistants your customers use, on a schedule, and record whether you appear, how you are described, and who appears beside you. That is a monthly report a junior marketer can produce.
It is qualitative and it is imperfect. It is also the only direct read on a surface that decides most of your shortlist placement. Being described inaccurately by an assistant is a fixable content problem, but only once you know it is happening.
Our post on why ranking no longer predicts AI citation covers the visibility side, and what LLM referral traffic is worth covers the small slice that does click through. Our marketing dashboards service builds the self-reported attribution, branded-demand and citation-tracking views into one report, and our SEO service handles the content changes those reports point to.
What to verify before acting on The AI Dark Funnel
Rules and platform behaviour change after an article is published. Confirm campaign policy, billing settings, attribution windows, conversion tracking, and platform changes against the Google Ads Help before you act on anything below, because the right answer depends on your entity, state, turnover, and current setup.
| Checkpoint | Why it matters | Where to confirm |
|---|---|---|
| Current rule or platform status | Limits, forms, policies, and APIs can change after a blog update. | Google Ads Help |
| Your exact business case | A local shop, freelancer, D2C store, agency, and SaaS team rarely need the same next step. | Documents, invoices, campaign data, analytics setup, or workflow logs |
| Implementation evidence | The safest campaign decision is backed by proof, not memory or screenshots from an old setup. | Portal acknowledgement, dashboard export, invoice sample, test lead, or error log |
Going deeper: Marketing Dashboards, and SEO Services.
Frequently asked questions
What is the AI dark funnel?
It is every stage of the buying journey that happens inside an AI conversation, where tracking pixels, UTM parameters and referral data do not exist. A buyer asks an assistant to compare vendors, forms a shortlist, then arrives at your site typing your brand name directly. Analytics records direct traffic; the comparison that decided the shortlist is invisible, as is every competitor in it.
Why does last-click attribution mislead in 2026?
Last-click credits the channel closest to purchase, and the dark funnel sits entirely upstream of that. The result is that brand and content budgets look inefficient while branded paid search looks extraordinary. Teams then cut the upstream work that generated the demand and increase spend capturing it, which works until the demand stops arriving.
How can a small business measure the dark funnel?
Three cheap instruments do most of the job. Self-reported attribution — a required "how did you hear about us" field on your forms — captures what tracking cannot. Branded search volume works as a demand proxy. Geographic or temporal holdout tests give causal evidence with no tracking at all. Self-reported data is unreliable individually and quite good in aggregate.
How do you get visibility inside AI conversations?
You cannot instrument it, so you sample it. Run your top 20 buying-intent questions through the assistants your customers use on a schedule, and record whether you appear, how you are described and who appears alongside you. It is qualitative and imperfect, but it is the only direct read on a surface that decides most shortlist placement.
Let's talk about your business.
Tell us what you're working on and where you want to go. We'll put together a plan. No obligation, no sales pitch.
- Free 30-minute call
- A plan built around your goals
- No obligation, no pressure
- Your own account manager