Free report: “5 Things $39.6M in B2B Programmatic Spend Taught Us.” Three years of managed spend and a billion tracked page loads, turned into five findings platforms won’t put in a pitch deck. Subscribe free to unlock it, plus every new issue of The B2B Stack.
Join 1,600+ B2B marketers, including readers at Dentsu, HubSpot, 6sense, Informa and DirecTV.
A deal lands in your CRM.
No prior form fill. No campaign touch. No attributed source. The SDR who logged it has written “inbound / unknown” in the notes field because that’s the only honest thing to write. Leadership asks where it came from. Marketing checks the attribution model and finds nothing. The contact, when asked how they heard of you, says something vague like “I’d seen you around” or “a colleague mentioned you.”
In most revenue teams, this gets filed under “lucky.” A random inbound. Unexplained pipeline. The kind of deal you’re grateful for but can’t learn from, can’t replicate, and definitely can’t put in a board slide.
But here’s what I’ve come to believe after a long time watching these deals accumulate: they are not random.
They are the visible output of an invisible system — a months-long process of brand impression, peer validation, and autonomous research that happened entirely outside your tracked channels. The buyer didn’t appear from nowhere. They arrived pre-educated, pre-validated, and in many cases, pre-decided. You just didn’t have visibility into any of the journey that led them there.
This is the dark funnel. And the instinct to treat it as a measurement problem — to chase attribution, to install tracking tools, to add “how did you hear about us?” to every form — fundamentally misunderstands what it is and what to do about it.
The dark funnel isn’t a gap in your analytics. It’s where modern B2B buying decisions are actually made. And the organisations that understand this aren’t trying to illuminate it. They’re building strategic presence inside it.
Where Buying Decisions Are Actually Being Formed in 2026
To understand why the dark funnel has become so consequential, you need to start with a clear-eyed account of how B2B buyers actually conduct research today — which looks almost nothing like the linear content journey that most marketing funnels are designed around. The systems are grossly too simple for how complex the B2B journey is in reality
The modern B2B buyer is, first and foremost, deeply peer-networked. They have professional communities — some formal, most informal — where they ask questions, share opinions, and get vendor recommendations from people they trust far more than they trust any vendor. These conversations happen in private LinkedIn groups, in industry Slack communities (many of which have thousands of members and active daily discussion), in WhatsApp threads between former colleagues or groups like the Slack ones mentioned above, in Discord servers built around specific disciplines, and in the DMs of people who’ve become trusted advisors through years of LinkedIn connection. Revenue Collective, Pavilion, demand generation communities like Exit Five, category-specific forums — these are not niche edge cases. They are the primary research environment for a significant proportion of B2B buyers in technology and marketing.
The defining characteristic of all of these channels is that they are invisible to your analytics stack. Invisible today, invisible tomorrow and they will remain invisible forever. A buyer posting in a private community asking “has anyone used FunnelFuel?” will generate opinions, recommendations, warnings, and first-hand accounts from ten different people — none of which will appear anywhere in your attribution data. The entire conversation, which may be the single most influential input into that buyer’s shortlisting decision, will register as zero.
Review platforms represent another dark channel that is simultaneously more measurable and less understood than it should be. G2, Gartner Peer Insights, Capterra, and TrustRadius are heavily consulted during active evaluation phases — but buyers also browse them passively, long before an active buying process begins, forming impressions about vendor categories and the competitive landscape. The buyer who reads twelve reviews on G2 and forms a strong negative impression of your product has had a meaningful interaction with your brand that will shape their entire subsequent evaluation. You will almost certainly never know it happened. Not to mention the impact these platforms have on the sentiment presented back to the user on their favourite chatbot. These are a foundational part of the funnel, and are under estimated in their ability to forge narrative
Podcasts occupy a similar position. The B2B marketing and revenue ecosystem has a remarkably active podcast landscape — The Marketing Millennials, The Revenue Formula, Exit Five, The B2B Playbook, Demand Gen U, and dozens of category-specific shows with engaged professional audiences. A decision-maker who listens to a forty-five minute conversation featuring one of your leaders — or, equally, featuring your competitor — will be influenced by that content in ways that far exceed what you’d expect from a comparable amount of time on your website. The intimacy of the medium, the apparent authenticity of the conversation, and the passive consumption context all make podcast impressions stickier than most owned content. Attribution: zero.
LinkedIn itself is partly visible and partly dark. Company page impressions, paid campaign performance, and post engagement are measurable. But the dark social dimension of LinkedIn — content shared into DMs, screenshots circulated in WhatsApp groups, posts saved and returned to weeks later, organic algorithm reach into second and third-degree networks — is largely untracked. Research from Wynter and others into B2B buying behaviour consistently shows that LinkedIn content influences purchasing decisions that never surface in any marketing attribution model.
And then there’s the new dimension that changes everything.
The AI Research Layer: The Dark Funnel Just Got Pitch Black
Something has shifted in the past eighteen months that most B2B marketing teams have not yet fully reckoned with, even if they are faintly understood conceptually.
Buyers are now using AI-powered research tools — ChatGPT, Perplexity, Claude, Gemini — as active research assistants at every stage of the purchasing process. This is not a marginal behaviour. A 2024 survey by Forrester found that over 60% of B2B technology buyers reported using generative AI tools to assist with vendor research and evaluation, a number that has almost certainly continued to climb. Gartner has projected that by 2026, traditional search engine volume will fall by 25% as AI assistants absorb a significant proportion of informational queries.
The implications for B2B marketing are profound and largely unaddressed.
When a VP of Marketing at a mid-market SaaS company asks Perplexity “what are the best B2B programmatic advertising platforms for a company targeting enterprise accounts in financial services,” they will receive a synthesised answer drawn from whatever sources the AI has indexed, weighted, and deemed authoritative. That answer will influence whether you appear in their consideration set at all — and the buyer will have no awareness of, and no particular interest in, understanding what determined the AI’s response. They will treat it as a relatively neutral research output, in the same way a previous generation treated Google’s first page results.
The critical question this raises — and one that the B2B marketing community is only beginning to grapple with seriously — is: what determines whether your brand, your positioning, your distinct point of view appears in AI-generated research outputs?
The answer, based on the current evidence and the way large language models work, appears to be some combination of: the breadth and authority of your published content across the open web, the frequency with which your brand and specific claims are referenced in third-party sources (analyst coverage, review platforms, trade press, guest articles, podcast transcripts, community discussions), the clarity and consistency of your positioning language across those sources, and the degree to which your perspective is genuinely differentiated rather than category-generic.
This is, in a deep sense, the 2026 version of SEO. The underlying mechanisms are different — you’re not optimising for keyword rankings in a search index, you’re building the kind of distributed content presence that gives an LLM enough signal to include you as a credible answer to a relevant question. But the strategic logic is familiar: sustained investment in high-quality, genuinely useful content, consistently published and distributed across authoritative channels, creates compounding presence that eventually influences how your category gets described by automated systems.
What’s new — and what makes this more urgent than traditional SEO — is that LLM-generated research outputs are presented with an authority and completeness that search results are not. A search results page is obviously incomplete; the buyer knows they’re seeing ten blue links out of millions of possible results. An AI-generated research summary presents itself as a synthesis. The buyer’s instinct is to treat it as more definitive than it actually is. This means the stakes of appearing or not appearing in AI research outputs are higher than the equivalent SEO stakes — and the consequences of absence are more severe. This last point is largely unrecognised, in my opinion.
The industry is, understandably, concerned about this. Revenue teams watching unexplained inbounds arrive referencing “I did some research online” without being able to identify the touchpoint are experiencing a genuine measurement anxiety. The instinct is to demand a solution — a way to track AI-referred traffic, to measure LLM citation share, to get the dark funnel back under the attribution model. Several vendors are beginning to offer tools in this space: Share of Model tools that test how frequently your brand appears in AI responses to relevant category questions, AI SEO platforms that attempt to optimise content for LLM consumption, and attribution approaches that attempt to infer AI-influenced journeys from first-party data patterns. Bluntly I am yet to see an enterprise grade version of this concept, most are clunky and full of gaps, and seem to miss just how personalised results are alongside the depth of prompt versus old school keyword research - the former uses paragraph long unique inputs, the latter uses 1-2-3 keyword inputs. Clearly the latter is more modellable
That said, these tools have real value as directional intelligence. But the underlying anxiety they’re designed to address is worth examining critically, because I think it reflects the wrong frame.
Below is how I think that invisible system works, and how I’d plan media for it.