AI Didn’t Kill The Dark Funnel - Here’s How It Rewired It
Why the B2B Buyer Journey Is Now a Signal Stack, Not a Funnel
There’s been a rising tide of LinkedIn posts, conference chatter, and newsletter takes around the so-called collapse of the “dark funnel.” People are noticing site traffic is down. First-party intent signals are noisier. Analytics dashboards aren’t lighting up like they used to.
LLMs as a replacement for search hit something of an inflection point as we entered 2025. Back in January they amounted to 1.5% of search yet by June they had crossed 10% - a near order of magnitude explosion. The clickless search journey has been Google’s path for years has accelerated with their AI overviews. The net result - a year on year decrease of 25% in search referee traffic. With it narratives like the training of the content web reservoir and doomsday for the open internet. Our interest is B2B though, so what’s the impact on B2B marketers?
So despite doomsday, in our own platform data here at FunnelFuel, specifically our B2B web analytics which tracks 100s of vendor sites globally, and in conversations with enterprise clients, a different truth emerges:
Meetings haven’t stopped. Pipeline is still building. Buyers are still moving. 97% of all companies which will buy from a vendor will still visit the vendors site before they buy - even if that is potentially now more representative of the last mile then then full buying cycle
They’re just not doing it in ways we used to recognize.
So where did the signals go?
The answer isn’t that the funnel died. It’s that it’s been restructured, re-segmented, and in many cases, re-routed through LLMs and private research environments that aren’t emitting traditional marketing signals at all.
The buyer journey hasn’t gotten shorter. If anything committees and buying groups are bigger. But it has gotten quieter. And that changes everything for B2B marketers, product leaders, and signal designers.
What’s Actually Happening With Traffic?
Across the industry, many analytics teams are reporting a meaningful dip in top-of-funnel website activity. We see this across several publishers and Big vendors that we work with, particularly in software and high-consideration B2B categories. But here’s the tension: while visits are down, meeting volumes and sales velocity haven’t dropped proportionally according to the clients themselves - in fact the marketing we’re doing has never worked better for them and investments in it are significantly up. In fact, some clients are seeing higher-quality meetings from what looks like less traffic. Does the AI era breed efficiency by chopping out semi-random click noise? Is LLM funnelling a more informed and ready to buy audience and if so are we going to see a collapse in sales cycle length over the coming years?
The instinct is to celebrate this as a sharpening of signal. “Less noise, more action.” But that conclusion might be premature.
Because just as likely — and arguably more concerning — is the scenario that buyers are doing their research somewhere else entirely, often without ever hitting your website until very late in the cycle. The signal isn’t sharper. It’s simply gone dark again, but through different mechanisms.
For me the clear outcome is that brands need to rapidly escalate their brand activity and upper funnel activity and accept that the funnel will stay dark for longer, accept a lower level of company level research attribution and except the ABM “go signal” to come later in the cycle. There’s a whole post of ramifications in that set of sentences - the need to find buyers everywhere for brand, omnichannel and omnitactics and a need for ABM to have real and solid data pipelines that take the go signal and quite figuratively and literally - go!
The LLM Is Not the Buyer. It’s Their Research Co-Pilot.
Let’s clear something up. No serious B2B buyer is asking an LLM like ChatGPT, Claude, or Perplexity one question and making a seven-figure decision based on the output. That’s not how enterprise buying works — not now, not ever. It would also make a mockery of an 11 month buying cycle - surely if this was the reality it would be nearer 11 days. ChatGPT said buy X, go and get the budget signed off. Clearly that’s not the case
But what is happening is this:
Buyers are using LLMs to summarize, compare, and pre-filter before they engage - they’re short cutting the research hard yards with some prompt stacking
Teams are plugging RFPs, product one-pagers, and comparison sheets into GPTs for internal digestion - they’re outsourcing the comparison and thinking in the same way a junior employee may have been tasked with crunching it 5 years ago
Procurement functions are deploying LLMs to standardize evaluation, prioritize risk, and even vet pricing logic
Teams are using LLM to validate their thinking and find gaps in their logic. A first pass before the boss sees it. What did I miss kind of narrative
LLMs aren’t decision-makers. But they are now very often the first gate your brand needs to pass through. That means a lot of traditional signal behaviors — repeat site visits, whitepaper downloads, high time-on-site — are simply bypassed by research workflows that feel more efficient to the buyer. Clearly as I’ve been arguing for months - we simply have to be cerebral in our ability to collect, ingest, infer and pipeline signal in >2025
These tools have become private, safe, buyer-controlled spaces to explore and validate. And in those spaces, most companies are invisible unless their content is accessible, clear, structured, and optimized for LLM input - or gets crawled by Google (ouch)
Why Funnels Are Failing as Mental Models
Funnels were always a simplification — a model for mapping awareness to conversion in a neat progression. But in a world where buyers conduct asynchronous, multi-source, self-directed research, the funnel flattens. A gross oversimplification is now fully out of date
Let’s reframe the concept: it’s not a funnel.
It’s a signal stack.
At any point in time, a buyer may:
Read an analyst report
Ask a question to their LLM
Visit your website briefly
Consult internal benchmarks
Revisit a Slack thread from six months ago
Attend a partner webinar
And only then, contact sales
Each of those steps emits different kinds of signals, many of which aren’t visible in traditional attribution models. And many of which are now being processed — or initiated — by AI.
Your challenge isn’t to pull people “down” a funnel. It’s to design systems that listen across the stack, that can pattern real interest from anonymized or fragmented behaviors.
Economic Incentive Always Drives Signal Design
There’s another structural force at play. Buyers, when given the tools to avoid emitting signals, will use them. Because emitting signals (like filling out forms or triggering demo flows) gives power back to the seller. Who wants their phone lighting up with calls, or inbox clogged with crap?
And if there’s one pattern that defines how buyers are using AI today, it’s this:
They are using it to stay in control.
To do more research with less exposure.
To make fewer visible moves until they’re ready.
This aligns with findings from TrustRadius’ 2024 B2B Disconnect Report, which noted that buyers increasingly prefer vendor-agnostic content, peer validation, and non-attributable research touchpoints. Who’d have thought?
If you assume buyers will behave like they did in 2017 — triggering scoring workflows, hitting pages five times a week — you’ll miss the shift. The pendulum has swung. AI is more effective at defending the buying process than it is at automating sales. That’s a competitive reality.
Why We’re Still Betting on On-Site Signals — With a Twist
Here’s the paradox: buyers still end up on your site — often right before a commercial interaction. That’s consistent. That’s trackable. That’s useful. 97% of buyers have been on a vendors site pre purchase
But how they get there is now messier, more distributed, and less predictable than ever. The journey is long and wide. Your site is the final checkpoint, not the starting point.
This is why we invest so heavily in:
B2B signal analytics for anonymous-to-known signal lift - B2B always needed to be good at lifting signal and that’s accelerated now
IP-based firmographic enrichment linked to a signal graph, linked to media buying to de-anonymise web traffic in and out of the office using 50+ signals not just IP
AI-driven segmentation of top, mid, and bottom funnel site behaviors
Log-level data matching from programmatic platforms like The Trade Desk to tie post-view impressions to on-site return
Because we believe the buyer will hit your site eventually. But the signals leading there are increasingly buried in private environments — and without a measurement stack that reflects that complexity, marketers are flying blind.
From Funnels to Feedback Loops
Buyers still want to buy. That hasn’t changed. But the way they discover, compare, and validate has — and fast.
Search is no longer the dominant discovery mechanism in B2B. It’s the comparison and evaluation engine. Discovery has shifted upstream — to social, to private forums, to team chats, to AI co-pilots. And that means the way we structure marketing — and measurement — needs to follow.
LLMs haven’t killed the dark funnel. They’ve made it more sophisticated.
To win in this new era, B2B teams need to build signal-driven systems that:
Acknowledge the multi-source nature of discovery
Track what content informs decisions, not just what content collects email addresses
Recognize that less traffic doesn’t mean less interest — it may mean your brand is being skipped over upstream
Marketing in 2025 isn’t just about reach or clicks.
It’s about resolution.
And that means the best teams will think like systems architects — not just campaign managers.
Suggested Reading and References
Thanks for shareing this for us