Your best deals keep arriving from somewhere your analytics cannot see. So what do you publish into that research phase instead? The usual answer is a tool, and nothing in that gap changes.
There is a deal in your pipeline that closed with nothing recorded against it. Your team called it direct and moved on. Before that first call, your buyer read your pricing page, compared you to two rivals, and asked an assistant which vendor was cheaper. None of it became an event your analytics could store. The default fix is better tracking. I think the fix is somewhere else.
What the Dark Funnel Actually Describes
The dark funnel is the part of a B2B purchase that happens where your analytics cannot observe it. That covers private communities, peer calls, review sites, podcasts and AI assistants.
The term is vendor originated and contested. More than one origin story circulates, and I will not settle it for you. What matters is that the behavior is older than the phrase. Buyers asked peers for advice long before anyone named the phase.
Two things follow from that. The activity is real, and you can confirm it in a deal review this week. The individual session is not recoverable, so anyone promising a dashboard that replays it is selling you a model, not an observation.
The phrase earns its keep in a budget conversation, because it names spend that a channel report never explains. It stops being useful when it becomes an excuse. A deal attributed to direct can mean an invisible research phase. It can also mean a rep forgot to log a call.
Why Attribution Cannot Solve It
It is tempting to treat this as an attribution problem. Most explainers promise to bring the dark funnel into the light, then recommend multi-touch modeling. Those models tell you which events got credit. They cannot tell you what the people behind them read or concluded.
Intent data deserves a direct answer, because it is where this argument turns practical. A platform of that kind identifies which company visited your site, which helps a rep decide who to call. It tells you nothing about the peer call your buyer made.
If you are still choosing that category, my team keeps a separate guide to marketing attribution software. This article is the piece those guides skip: what to do when the data is not there.
The uncomfortable consequence is the one you will not hear from a vendor. If the phase is genuinely untrackable, attribution cannot be the goal. Attribution scores the events you captured, and the dark funnel is the set you did not.
You will read that the answer is to attribute it better. I think that advice is backwards. Better attribution improves the accounting of a journey you already recorded. It does nothing for the part you never saw.
So the goal becomes presence. Publish into the places where the research happens, then measure what that presence leaves behind.
The Channels Where Dark Research Actually Happens
Sit in on a lost deal review and you will hear the list. Your buyer called a peer at another company, read reviews, listened to a podcast on a commute, and asked an assistant to compare two vendors.
Not all of those are yours to influence. A peer call is your customer's voice. Reviews, communities, publications and AI answers are where you can put something durable this month.
Use one filter before you add a channel. Could a stranger find your contribution six months from now without knowing your name? A sponsored slot fails that test. A review response, a community answer and a page an assistant can quote all pass it.
The arrival is earlier than you think. Forrester's 2024 Buyers' Journey Survey found that 92% of B2B buyers began with at least one vendor in mind. Digital Commerce 360 reported that figure in July 2025, alongside a harder number: 41% already had a single preferred vendor before evaluation. You are defending a position that was built somewhere else.
What to Publish Into Each Channel
The middle column is where the work happens. Naming the artifact is the hard part, because the artifact is the only part of this you control. The right column keeps you honest. If you cannot name a signal that would move, you are doing brand work and should call it that.
Where the research happens | What to publish into it | Proxy signal that moves |
|---|---|---|
Review platforms | A current profile plus public vendor responses to the objections in competitor reviews | Share of new deals where the buyer names a review site on the first call |
Private communities and peer groups | A practitioner teardown a member would forward without a gate, priced in the open | Community referral sessions in server logs; branded mentions inside the group |
Podcasts and video interviews | Your operator, not your logo: one strong opinion and one number per episode | Branded search lift in the four weeks after release |
Independent publications and analysts | One citable data point with a methodology note and a named author | Citations and links from domains you never pitched |
AI assistants | Plain, extractable answers to the comparison and pricing questions buyers actually type | AI referral sessions landing on pricing or comparison pages |
Live events, dinners, roundtables | An agenda people would attend without your booth, and a follow-up artifact worth keeping | Self-reported source on the first call; meeting to opportunity rate |
One to one peer calls | Proof a customer can repeat without you in the room | Customer sourced introductions; win rate on deals with a reference call |
Two rules for the middle column. Publish something a buyer can use without talking to you, because a gate closes the dark funnel rather than opening it. And write at practitioner level, because your reader is comparing you to someone who did.
That last row is where teams underinvest. A reference call is the highest trust channel in B2B and the least instrumented. If your customers cannot explain what you do in their own words, publishing will not fix it. The program that produces that proof is customer advocacy.
A good artifact is specific. Take the AI assistants row. A page promising transparent pricing is too vague to quote. List your bands, what changes between them, and what happens when you outgrow a tier.
The publications row rewards a different instinct. Most teams pitch a story. The version that gets cited is smaller and duller: one number from your own customer base, with the sample size and the period stated next to it. A benchmark of two hundred support tickets is worth more to a journalist than a trend you cannot defend, because it is the only thing on your site they can attribute.
Nothing in that middle column sits behind a form. A gated ebook fails the six-month test, and it is what most B2B teams produce instead.
Objections work the same way. Answer the ones that keep appearing in your competitors' one-star reviews, in public. A reader comparing two vendors will find both pages, and only one will have an answer.
How to Tell It Is Working Without a Dashboard
You will not get a clean report, so stop asking for one. A proxy is only useful when it is tied to a decision you already made. Every proxy needs a numerator, a denominator and a fixed window, and none may be added to another.
Start with one question on every first call. How did you first hear about us? Write the answer verbatim into a field your team actually fills. After a quarter you have a distribution instead of a hunch.
Branded search is the second signal, and the noisiest. A lift two to six weeks after a podcast or a publication hit is the nearest thing to a receipt. Compare it against the weeks you published nothing.
Then score direct traffic as a bucket rather than a default. Pull the share of direct sessions that land on pricing or comparison pages. If direct converts better than paid and lands on late-stage pages, it is not direct traffic. It is your buyer, arriving from somewhere the referrer field could not name.
A clickstream study published by Similarweb makes the gap concrete, and the figures are the vendor's own. Its panel covered US desktop traffic from July to December 2025. Users exposed to a ChatGPT brand recommendation were 2.5 times more likely to visit that brand than a direct competitor. Of AI influenced visits, 8.8% arrived as AI referrals and 55.9% through search. Build your report on referral data alone and you capture a tenth of the effect.
For the deals you can trace, our guide to revenue attribution covers which model fits which sales cycle. Use the proxies above for the ones you cannot.
Four quarters of a signal is a pattern you can defend. One quarter is a hypothesis, and you should call it that when the budget conversation happens.
The AI Dark Funnel
The newest version of this is the least observable. When a buyer asks an assistant to compare three vendors, the answer is assembled before any of those vendors knows a question was asked. Nothing lands in your referrer field, and there is no session for you to inspect.
Gartner's May 2026 survey of 645 B2B buyers found that 45% of them used generative AI during a recent purchase. G2, a review marketplace publishing research about its own category, surveyed 1,076 software buyers in March 2026. In that vendor study, 51% said they now start research in a chatbot more often than in a search engine, up from 29% in April 2025. In the same survey, 69% said they picked a different vendor than planned because a chatbot recommended it.
That last number is the one that should stop you. The shortlist can change in a channel you cannot log into, so the content feeding those answers matters more than the ad you run against the category term.
There is a practical test here that costs nothing. Open a private window and ask three assistants to compare you against your two closest rivals. Whatever they say is what a real buyer has already been told.
Gartner's survey carries a counterweight for anyone claiming the rep is obsolete. Roughly two in three buyers still prefer to validate AI generated insights with a sales rep.
The practical move is unglamorous. Write the comparison yourself, with the numbers a buyer would ask for, on a page a machine can read and quote. If you want to know whether it is working, AI visibility tools show how often you appear in those answers and which pages get cited.
One detail decides whether that page gets used. State your weaknesses. An assistant asked to compare two vendors will quote whichever source admits a limitation, because that reads as evidence rather than marketing. A page that only lists strengths gives the model nothing to work with.
Where to Start This Quarter
Three moves you can make this quarter, in this order.
First, run the source question on your last twenty closed deals. You will get a distribution you did not expect.
Second, pick two channels from the table where you already have something credible, a person or a data point. Publish into both once a month for a quarter, attach one proxy to each, and decide in advance what would change your mind. Two is deliberate: a team that starts with five produces five half-finished artifacts and learns nothing from any of them.
Third, make your best customers repeatable. You cannot script the conversation they have about you. You can equip the people willing to have it, with a proof point and a page they can send.
None of this will appear as a clean line in a report. It will show up as a shorter sales cycle and a buyer who already knows what you cost. Give the work a full quarter before you judge the channel, and judge the artifact before you judge the channel.
So what does your own last twenty deals say? If most of them started somewhere you cannot log into, the budget conversation this quarter is not about attribution. It is about what you are willing to publish without a form in front of it.
Frequently Asked Questions
Is the dark funnel the same thing as dark social?
No. Dark social is one channel, the private sharing in messaging apps and email. The dark funnel is the wider set: peer calls, communities, review sites, podcasts and AI assistants. Dark social is a slice of it, not a synonym.
Can you measure the dark funnel at all?
Not directly, and anyone claiming otherwise is selling you a proxy. You can watch self-reported attribution, branded search lift, the quality of direct traffic and AI referral patterns. Each is an indicator with a denominator, never a total, so never add them together.
How much of the B2B buying journey happens before a buyer contacts sales?
Published estimates use incompatible definitions, so I will not quote one. Some count time, some count touchpoints, some count buyers who never contact sales. Use your own self-reported data instead. It is narrower and far more actionable.
Should we buy an ABM platform to reach dark funnel buyers?
Only if you need account identification for a specific sales motion. That is a different purchase from this one. A platform can tell you which company visited. It cannot tell you what they concluded, and it will not publish anything for you.
Start with the research phase, not the dashboard
The dark funnel is a publishing problem before it is a measurement problem. Allable connects your search data, your content and your CMS, so the answer to a buyer's comparison question can exist before they ask it.
