The Funnel Went Dark: What B2B Marketing Does When the Top of It Turns Invisible

Pull up your organic traffic chart for the last eighteen months. For most B2B companies, the line does something that used to signal a catastrophe. It falls, and it keeps falling, and no amount of new content seems to bend it back up.

Now pull up your pipeline over the same period. For a lot of those same companies, it held. In some cases it grew.

Those two charts are not supposed to coexist. The whole logic of demand generation says traffic feeds leads and leads feed pipeline, so when the top of the funnel drops by a third, the bottom is supposed to follow it down a few months later. When it doesn't, most marketing leaders assume they got lucky, or that sales closed some deals that were already in flight, and they move on.

They didn't get lucky. What actually happened is that the top of their funnel moved somewhere they can't see it.

For CMOs, Demand Generation Leaders, and RevOps Teams.

The traffic really is gone, and it is not coming back

Start with the part that shows up in your own dashboards, because it is not subtle.

Roughly 68 percent of searches in the United States now end without a single click to any website. When Google shows an AI Overview at the top of the results, that figure climbs to around 83 percent. On mobile it sits in the mid-to-high seventies. The search still happens. The answer still gets delivered. It just gets delivered inside the results page, and the user never leaves.

B2B took the worst of it. One analysis of B2B websites found that 73 percent of them lost meaningful organic traffic between 2024 and 2025, with the average year-over-year decline landing near 34 percent. A third of your organic traffic, gone in a year, across three out of four sites in the category.

The reason B2B got hit harder than most is structural. B2B content skews toward exactly the kind of query an AI Overview answers best: definitions, comparisons, "how does X work," "what is the difference between Y and Z." Those informational articles were the workhorses of B2B content marketing for fifteen years. They were also the easiest pages in the world for a language model to summarize and replace.

There is one piece of good news buried in the numbers, and it matters for what comes later. The clicks that survive are better. When someone reads an AI summary and still decides to click through, they are further along and more serious, and analyses put their conversion rate meaningfully higher than pre-AI baseline traffic. You are losing volume at the top while the quality of what remains goes up. Hold that thought.

The part you can't see is the bigger story

The traffic collapse is the visible symptom. The invisible cause is where your buyers went instead.

Forrester's 2026 Buyers' Journey Survey ran across 18,000 business buyers worldwide, and the top-line number reorders how you should think about your funnel. 94 percent of them used AI somewhere in their most recent purchase. More than half compared vendors directly inside AI tools. Roughly the same share used those tools to research products, and nearly half used them to build the internal business case before they ever contacted a vendor.

Read that last part again. Almost half of your future buyers are assembling the case for buying something like your product, naming competitors, weighing tradeoffs, drafting the argument they will take to their boss, and they are doing all of it inside a chat window you will never see.

The direction is not new. It is an old one that finally hit its limit. Gartner has spent years documenting that B2B buyers spend only about 17 percent of their total buying time meeting with any potential supplier, which leaves the other 80-odd percent as self-directed research. What changed in the last two years is where that self-directed research lives. It used to live on your website, your competitors' websites, review sites, and Google. A trail of cookies and UTM parameters followed the buyer around, and you could reconstruct roughly what they looked at. Now a large chunk of it lives inside ChatGPT, Perplexity, Gemini, and Copilot, and those leave no trail on your property at all.

So the buyer still does the work. You just lost the ability to watch them do it.

Why this breaks marketing specifically

Sales has always operated with imperfect visibility into buyer research, so this shift lands on them as a difference of degree. For marketing it is a difference of kind, because the entire operating model was built on being able to see and measure the top of the funnel.

Think about what depends on that visibility. Lead scoring assumes you can observe the behaviors that predict intent. Attribution assumes you can trace a closed deal back to the touchpoints that created it. Content ROI assumes you can connect a piece of content to the traffic it drew and the pipeline that followed. Channel budgets get set by comparing measured cost per lead across channels. Every one of those depends on the buyer moving through instrumented space where you can tag them.

When the research moves into AI tools, the instrumentation goes dark. A buyer can read three of your competitor comparisons through an AI summary, decide you are the frontrunner, watch two of your customer videos, and land on your site by typing your name straight into the address bar. In your analytics that shows up as "direct traffic, converted." No source. No content attribution. No journey. Just a serious buyer who appeared, fully formed, from a process you had no window into.

Multiply that by half your pipeline and you understand why the two charts stopped agreeing. The pipeline is real. The measurement that used to explain it is broken.

Four moves that actually respond to this

The wrong response is to keep optimizing the old machine harder. Publishing more of the informational content that AI Overviews eat, chasing keyword rankings that no longer convert to clicks, and defending an attribution model that has gone blind in one eye. Here is what a serious response looks like instead.

Rebuild your content around what AI can't answer for the buyer

The content that lost its traffic was the content a model can fully answer on its own. Definitions, basic how-tos, generic listicles. That work is now a commodity, and the machine does it for free.

What a model cannot manufacture is anything grounded in something only you have. Proprietary data from your own customer base. Named case studies with real numbers. Original benchmarks. Strong, specific points of view that take a side. Frameworks that carry your fingerprints. This content does two jobs at once: it is the material AI tools cite when they answer buyer questions, and it is the reason a buyer who read the summary still clicks through to the source. Shift your content budget from volume toward the pieces that are worth citing and worth the click.

Measure the funnel you can see, and stop pretending you see the rest

You will not get the old attribution back. Trying to reverse-engineer a full journey out of "direct traffic" is a fiction, and dressing it up in a dashboard just launders a guess into a number your CFO will hold you to.

Track the leading indicators that still work. Branded search volume tells you how many people are looking for you specifically, which rises when your invisible research presence is strong. Direct traffic and its quality, share of voice inside your category, the percentage of new pipeline where the buyer already knew what they wanted on the first call. Run self-reported attribution by simply asking "how did you first hear about us" on demo forms and in win interviews, and treat the answers as real signal, because in an unmeasurable world they often beat the model that claims precision it no longer has.

Get yourself cited inside the tools where the research now happens

If buyers form their shortlist inside AI tools, your job is to be present and accurate when those tools describe your category. Check what ChatGPT, Perplexity, and Google's AI Overviews actually say when someone asks about your space and your competitors. You will usually find a mix of the outdated, the incomplete, and the occasionally wrong.

Fixing that is real work, not a plugin. It means publishing clear, structured, well-sourced material that models can parse and trust, keeping third-party sources like review sites and analyst pages current because those get pulled into answers heavily, and monitoring your AI visibility the way you once monitored keyword rankings. This is its own discipline now, and the teams treating it seriously are the ones showing up in the answer instead of hoping to.

Give sales a map for buyers who arrive already decided

The buyer who did all their research in AI shows up to the first call with a formed opinion, a shortlist, and sometimes a business case already drafted. That is not a threat to sales, it is a different kind of opportunity, but only if the rep knows it is happening.

Brief your sales team on the new pattern. Buyers now consult around seven sources before a conversation, and even after all that AI research a large majority still want a human to validate what they found. That validation moment is where deals are won now. The first call is no longer discovery in the old sense. The buyer already did that part. What they want from a rep now is confirmation, a correction where they got something wrong, and enough trust to move. Reps who open with "tell me about your challenges" waste the goodwill of a buyer who spent five hours getting ready. Reps who open with "here is what most people in your position get wrong about this decision" earn it.

The honest reframe

It is tempting to read all of this as pure loss. A third of the traffic gone, attribution in ruins, the funnel disappearing into a black box owned by a handful of AI companies. On the surface that is a grim story for anyone whose job is to generate and prove demand.

But the demand did not disappear. Buyers are researching more than ever, more thoroughly than ever, and arriving more prepared than ever. What disappeared is a particular way of counting it, one that gave marketing a comforting sense of control that was always partly an illusion. The cookie trail made you feel like you understood the buyer's mind. You never really did. You understood the parts of their behavior that happened to land on instrumented pages.

The teams that will do well from here are the ones that stop mourning the metrics and start playing the actual game: be the most useful, most citable, most trustworthy presence in a category conversation that now happens largely without you in the room. You cannot watch it. You can still shape it. And the buyer who finally types your name into the address bar, having decided in a room you were never invited to, is worth more than a hundred of the anonymous clicks you used to count.

That is the trade. Less to measure, better buyers to close. Most marketing organizations would take that deal if they understood it was the one on the table.