The B2B SaaS Dark Funnel in 2026: Measured, Not Estimated

TL;DR: The statistic the whole category quotes, that 73% of the buying journey happens before you see the buyer, is a 2024 reading from one region. The same researchers published 60% for 2025, and trade coverage of that release reported buyers reaching out roughly 12 weeks earlier. Stop quoting it and measure your own funnel. LoudFace measured its own: in the 90 days to 26 August 2026, 34% of conversion events arrived with no traceable source.

Short answer

The B2B SaaS dark funnel is the part of the buying journey that happens where your analytics cannot follow: private Slack and Discord groups, Reddit threads, review sites, podcasts, forwarded links, and AI answer engines. It can be partly measured. In LoudFace's own funnel, 34% of conversion events carried no traceable first touch, across 41 booked calls, audits and applications in the 90 days to 26 August 2026.

Where does the 73% figure actually come from?

Search "dark funnel" and page one repeats one number. Column Five titled a piece "The Dark Funnel: Why You're Missing 73% of Your Buyer's Journey", last updated February 2025. MarketBetter published "B2B Dark Funnel 2026: How to Find the 73% of Buyers Hiding From You" in February 2026. Others put the range at 70 to 80 percent.

To their credit, they attribute it. Column Five names the APAC B2B Buyer Journey Research Report by Green Hat and 6sense, and links both organisations. MarketBetter credits "6sense/Green Hat APAC Research", though that link points to Column Five's article and not to the research. One publisher cites the study; the next cites the publisher. This is not a fabricated statistic and nobody should say it is.

The problem is what happened to it afterwards.

That research runs in editions, and the two most recent measure the same thing: how far through the buying process a buyer has travelled at the moment they first contact a vendor.

Green Hat's own report page states the 2025 figure plainly: 60% of the buying journey is over by the time the buyer engages the vendor. Trade coverage of that release reported buyers reaching out roughly 12 weeks earlier than the year before, a shift that reflects both the earlier contact point and a shorter overall cycle.

So the number these 2026 articles quote to prove buyers are hiding longer is the previous edition's figure, from a study whose current edition shows movement in the opposite direction. It is also APAC, and it was never global.

Why does the same statistic keep changing shape?

Because it gets restated, and each restatement moves the thing being counted. The primary source measures journey position at first vendor contact. Downstream it appears as a share of buyers, as a share of the evaluation process, and as a share of activity in channels analytics cannot capture.

That last one is the leap that matters. Anonymous and untrackable are not the same condition. A buyer reading your pricing page without identifying themselves is anonymous and completely tracked: you have the session, the pages, the referrer. The dark funnel is the narrower case where you have none of that. Conflating the two inflates the problem and then sells you software for it.

The incentive is worth naming plainly. 6sense co-published the research this number comes from, and 6sense sells software for the problem the number describes. That is not an accusation. The study discloses its method and its sample, and its own 2025 edition revised the figure downward, which is not the behaviour of a vendor massaging a statistic. But a number that travels this far, this loosely, in a direction that happens to grow a market, has earned a reader who checks which edition they are quoting.

Where does dark funnel activity actually happen?

The channels are well established:

• Private communities: Slack groups, Discord servers, Reddit threads
• Dark social: WhatsApp, Signal, iMessage, LinkedIn DMs, forwarded email
• Peer recommendation and word of mouth
• Third-party review platforms such as G2 and Gartner Peer Insights
• Answer engines: ChatGPT, Perplexity, Claude, Google AI Overviews
• Podcasts, webinars, and ungated content

The last two matter more each quarter, and they behave differently from the rest. A Slack conversation leaves no artifact you can query. An AI answer does, because the engine fetched a page to build it. LoudFace has written before about why an AI visitor does not behave like a Google visitor.

How much of your pipeline is actually dark?

LoudFace ran the measurement on its own funnel. Every conversion event in the 90 days to 26 August 2026 was checked against the first source recorded for that person.

Those 14 people booked directly on the calendar link without a browsing session under that cookie. What they have in common is that nothing in the analytics explains how they got there. Some will be referrals or repeat contacts. What was measured is the absence of a traceable source, and not the presence of a dark funnel touch.

The figure is also a floor. Anyone who researched in a community and later searched the brand name is counted inside the visible 66%.

Three caveats belong with the number. The sample is 41 events from one agency funnel, so treat it as a method to copy and not a benchmark to quote. Two of those 41 events carry a localhost:3005 first touch, which is local development traffic; excluding both lifts the figure to 36%, and the unexcluded 34% is the published number. And the denominator rolls daily, so the figure only means anything with its date attached.

Is AI search part of the dark funnel?

Yes, and it is the part you can still quantify. LoudFace recorded 42 AI-referred visitors across 28 days and one booked call traceable to an AI engine over 90 days. Referrer handling varies by engine, so those counts understate the real number.

The more useful measurement is what the engines are reading. Across 96,674 retrievals and 63,983 citations on B2B SaaS buyer questions, over the 30 days to 25 August 2026, covering the top 1,000 source domains:

Five source types are shown; loudface.co's own domain, an unclassified residual bucket and one very small category are excluded, so the rows do not sum to the totals above. Corporate pages are by far the most retrieved, so engines are not ignoring company websites. They are discarding them: roughly four in ten pages pulled from a company site never reach the answer, against fewer than two in ten from a community source. Reddit alone produced 1,667 citations from 1,272 fetches, more than one quote per fetch. The same breakdown appears in the answer engine optimization guide.

Why does the prompt you track miss most of the retrieval?

Because the question a buyer types is usually not the query the engine runs. LoudFace measured 3,718 real AI answers over the 30 days to 25 August 2026 and counted the search queries each engine issued before answering, across 6,000 of the 6,404 sub-query records in the window.

Read the median first. Both engines answer most questions on a single query. An average would hide that, because ChatGPT's long tail drags the mean up to nearly two and makes fan-out look routine. ChatGPT split the prompt on almost half of its answers, running up to fifteen separate searches for one question, while Perplexity barely split at all. A tracked-prompt list watches one query while the engine runs several, and the retrieval that decides the answer happens in sub-queries you never chose to track. The mechanics are covered in query fan-out.

How do you measure the dark funnel?

Six methods, ordered by how quickly they return a real number.

1. Self-reported attribution. Add an open "How did you hear about us?" field to the booking form. Buyers name podcasts, colleagues, AI chats and other things no pixel can capture. This is the highest-yield change on the list and it takes an afternoon.
2. The no-source count. Query your analytics for conversion events whose person carries no first touch. That share is your measured dark funnel. For LoudFace, in the 90 days to 26 August 2026, it came to 34%.
3. AI referral capture. Detect visits from chatgpt.com, perplexity.ai, claude.ai, gemini.google.com and copilot.microsoft.com as a named channel. Engines differ: Perplexity passes a referrer reliably, while ChatGPT drops attribution from its mobile app, so expect a floor.
4. Citation tracking. Monitor which URLs answer engines retrieve and cite for your category. Peec AI and Profound sample the engines from outside and estimate this statistically. Profound also sells agent analytics, which it describes as tracking how your site is crawled by ChatGPT, Gemini, Claude and Perplexity. Evertune and AthenaHQ cover the same category. LoudFace scored the field in its review of AEO tools.
5. Branded search lift. Watch brand-name search volume in Google Search Console against periods with no paid or campaign activity. Rising branded search without a tracked cause is dark funnel influence surfacing. This is distinct from measuring an agency's return, which has its own method.
6. Direct traffic forensics. Segment direct traffic by landing page. Deep pages arriving as direct are a strong signal of shared links from Slack, email or DMs, though the signal is directional and cannot be confirmed per visit.

Methods one and two produce a number this week. The remaining four produce direction.

Which tools actually help?

The category splits into two halves that are rarely discussed together, and most teams buy from only one.

Intent and attribution platforms map the human dark funnel: third-party intent signals, multi-touch attribution across long cycles, community and social capture. 6sense, Dreamdata, HockeyStack and Common Room sit here.

AI-visibility platforms map the machine one: which URLs answer engines retrieve, and which they cite. Peec AI, Profound, Evertune and AthenaHQ sit here. LoudFace scored this half in its review of AEO tools, where Evertune, Profound and AthenaHQ all place above Peec. Disclosure: LoudFace runs Peec daily and sells AI-visibility work, so read that ranking knowing who wrote it.

Vendor positioning in both halves moves fast enough that any capability list is stale within months. Check each one's current product pages before shortlisting.

A programme that buys only the first half will keep reporting that AI search sends little traffic, because AI search mostly produces conviction, and conviction does not arrive as a click.

What this means for a B2B SaaS team

Treat the dark funnel as a measurement you have not run yet.

Start by shipping self-reported attribution on every conversion form, then count the conversion events with no traceable source and publish that number internally, however uncomfortable it reads. Once you have it, find out what answer engines say about your category, because the fastest-growing dark channel is the only one that leaves a readable artifact behind.

This is the shift from rankings to recommendations. The brands that win the next two years will be the ones present where attribution was never going to reach.

The honest limits

LoudFace's 34% comes from 41 conversion events over the 90 days to 26 August 2026, in a single agency funnel. It is one company and a small sample, and two of those events carry local development traffic as their first touch. The measurement identifies conversions with no traceable source, which is not the same as proving each one was influenced in the dark funnel.

The citation figures cover the top 1,000 source domains for one prompt set aimed at B2B SaaS agency buyers, over the 30 days to 25 August 2026, with no prior baseline for trend. The fan-out data covers two engines because the measurement tool does not expose sub-queries for the Google AI Overviews channel. That is a gap in the instrument, and it is not evidence about how Google behaves.

On the 73% figure: the number is not wrong. Other publishers attribute a similar 70 to 80 percent range to Forrester rather than to this study. That Forrester material was not checked here, and it is not in question. The narrow point stands. A 2024 APAC reading is being quoted in 2026 as a current global constant, and its own publisher has since revised it down.