AI Didn't Break the Marketing Funnel. It Exposed That Most Teams Never Built One.

Ahrefs studied 300,000 keywords last year and found that pages ranking first on Google lost 58% of their clicks when an AI Overview appeared above them. Seer Interactive put a wider frame on it. Across 3,119 informational queries at 42 companies, organic click-through rate on those queries fell 61% between mid-2024 and September 2025. Every marketing team with a content funnel watched the top of it get answered by a machine before anyone clicked.

The reaction in most marketing Slacks was panic about a broken funnel.

The funnel did not break. AI walked into the room and turned the lights on.

For a decade, the trackable funnel measured the one slice of buying that vendors could actually see. Gartner’s research on the B2B buying journey found that buyers spend just 17% of their time meeting with potential suppliers, and when they weigh three or four options, any single vendor gets 5% to 6%. The MQL dashboards, the last-touch attribution, the lead-scoring models: all of it instrumented that 17%. The other 83% happened in places no CRM could reach, and most teams told themselves it did not matter because they could not chart it.

That 83% has a name now. SparkToro’s zero-click research, built on clickstream data from Datos, found that most Google searches end without a click to the open web, and that by 2026 fewer than one in three sent a click at all. Buyers ask an AI, read the answer, and never arrive, which is why winning the AI answer itself is becoming the real top of funnel. The rest of the self-education runs through dark social: a Slack community, a peer text, a founder’s LinkedIn post forwarded into a private channel. None of it fills out a form.

6sense measured what that does to the deal. In its 2025 Buyer Experience Report, drawn from nearly 4,000 buyers, the winning vendor was already on the buyer’s Day One shortlist 95% of the time, and the favorite chosen before any vendor contact won roughly 80% of deals. First contact now lands at 61% of the journey. By the time a lead turns trackable, the decision is mostly made.

Read those numbers together and the funnel most teams defended was a scoreboard for the last quarter of a game that got decided in the first three.

The mechanism is unglamorous. A funnel counts what it can tag, so teams chased what they could tag: gated whitepapers, form fills, last-click conversions, a demo request credited to whichever ad the buyer touched on the way to a decision already made. It held up while competent content was scarce and a Google ranking delivered the click. AI removed both conditions. When every competitor can generate a passable comparison page in an afternoon, content stops differentiating anyone, and when the AI answer absorbs the click, the tag never fires. The measurement did not lie before. It tracked a proxy that AI just made worthless.

This is why the ROI gap on AI marketing spend reads as a paradox and is not one. Teams poured budget into more content and more touches inside the trackable 17%, then could not tie any of it to pipeline, because the pipeline was forming in the 83% they never staffed. Most buyers sit out of market in any given quarter, and brand is what survives that wait. A funnel that only captures in-market demand was always going to miss most of the work. AI just made the miss impossible to explain away.

A real demand engine now looks different from the funnel it replaces. Four moves define it.

Create demand where buyers self-educate. Be quotable in AI answers, present in the communities and feeds where shortlists form, and worth forwarding into a private channel. The job is to be the Day One favorite, not to catch a form fill on day sixty.

Measure with the buyer, not around them. Self-reported attribution, the “how did you hear about us” field on the demo form, now beats a last-touch model that only sees the final click. Ask the humans, because the software cannot see the Slack thread that sold them.

Separate demand creation from demand capture, and fund the first. Capture harvests intent that already exists. Creation builds the preference that settles the deal before a rep is ever contacted.

Judge the engine on pipeline and won deals, not MQL volume. The vanity funnel counted leads because leads were countable. The number that matters is whether you made the shortlist that 80% of deals never leave.

The teams calling this a broken funnel are mourning a measurement, not a market. Their buyers still research, still shortlist, still buy. What changed is that the part of the journey they never instrumented grew until it swallowed the part they did. AI did not take the funnel away. It stopped letting anyone pretend the dashboard was the demand.