Search used to end in a click. Increasingly, it doesn't. When Google's AI Mode answers a query directly, roughly 93% of those sessions end without a click to an external website at all, compared with 34% for traditional Google Search without AI Overviews - and even a standard AI Overview ends the session without a click 83% of the time (Conductor 2026 benchmarks, via Superlines). Ahrefs' analysis of 300,000 keywords found that AI Overviews now cut the average click-through rate to the top-ranking page by 58%, comparing December 2023 to December 2025 (Ahrefs research, reported via Businesswire, May 2026). And AI Overviews themselves aren't a niche feature anymore - they now appear on roughly 48% of tracked Google search results, up 58% year-over-year as of early 2026 (tracked AI Overview prevalence data).

Marketers have a name for what's disappearing: the "dark funnel" - the growing share of the customer journey that now happens inside an AI conversation, with no pixel, no UTM, and no referral row in your analytics to show it happened at all. This isn't a future risk to plan around. It's already showing up in the data marketers are looking at every day, and it's forcing a real rethink of what measurement can even see.

Why last-click and multi-touch attribution are the first casualties

Attribution - last-click or multi-touch - was built on an assumption that's quietly stopped holding: that a meaningful share of the journey leaves a digital trail you can stitch together. Multi-touch attribution needs touchpoints. When a prospect asks an AI assistant to compare products, gets a synthesized answer, and later converts through a channel with no reference to that conversation, there's nothing left for an MTA model to stitch. The interaction happened. The credit-assignment mechanism just never saw it.

That's part of why marketers are already shifting spend and belief toward aggregate methods: reported adoption of multi-touch attribution has grown to 47% (up from 31% in 2023), but marketing mix modeling adoption has grown faster in relative terms, to 26% from just 9% (Digital Applied, 2026). MMM doesn't need a touchpoint to work - it needs an aggregate outcome series and a plausible set of causal drivers, which is exactly what survives when individual-level tracking doesn't.

The twist most measurement conversations are missing

Most of the AI-dark-funnel commentary treats this as a digital measurement problem: your website analytics, your MTA stack, your paid search reporting are all going dark at the edges. That's true, but it understates the issue for any brand whose growth is still substantially driven by offline media.

Think about the mechanics of an AI-mediated research session: someone asks an assistant to compare mattresses, insurers, or SUVs. Where did the brands they mention, and the brands the person already has an opinion about, come from? For a large share of category purchases, the answer is still TV, OOH, radio, and print - media that built awareness and consideration long before the AI conversation ever happened, with no click generated at either end. The dark funnel doesn't start when someone opens ChatGPT. It often starts weeks earlier, in front of a television, and the AI session is just the invisible middle step between that offline exposure and an eventual sale.

A measurement stack that only watches digital touchpoints was already missing this loop. Now that even the digital surface is going dark, the gap compounds: you can't see the offline trigger, and you increasingly can't see the online research step either.

Why causal MMM is structurally the right tool for this moment

This is the specific reason marketing mix modeling is having a moment beyond the usual privacy-and-signal-loss argument. MMM was never trying to track the individual journey - it estimates the causal relationship between marketing activity and business outcomes using aggregate time series, which means it doesn't care whether a given research session left a click behind. If TV spend goes up in a region and revenue follows a predictable lag later, a properly specified MMM will find that relationship whether or not anyone can point to the specific AI conversation that happened in between.

That's also why MMM handles the dark funnel better than incrementality testing alone. A geo holdout test tells you the causal lift of a specific, isolated campaign; it doesn't reconstruct the whole system of offline and online channels interacting with each other and with an AI-mediated research layer sitting in the middle. You need both - but MMM is the layer that keeps working when the touchpoint-level view degrades further, and it's the layer that can actually account for offline media's role in driving the invisible research that AI search now absorbs.

What digital-only measurement tools structurally can't see

This is where the choice of measurement platform starts to matter as much as the choice of methodology. A digital-only MMM or incrementality tool - even a good one - is built around ad platform spend, web and app conversion data, and digital-channel time series. It can absolutely detect that "something is driving conversions that digital channels alone don't explain." What it can't do is tell you whether that something is TV, OOH, radio, or print, because it was never given that data as an input in the first place. The dark-funnel gap gets relabeled as "brand" or "baseline" and left unexplained, right at the moment when explaining it matters most.

What to actually do about it

A few practical moves for measurement teams grappling with this shift:

  1. Stop treating MTA gaps as a data-quality problem you'll eventually fix. The missing touchpoints are structural, not temporary - AI-mediated research isn't going to start leaving a clean trail.
  2. Feed offline media into the same causal model as digital, rather than running two disconnected systems and trying to explain the gap between them qualitatively.
  3. Watch AI referral and citation trends as a leading indicator, even though they'll never fully substitute for the volume of research happening inside AI conversations that never gets attributed anywhere.
  4. Re-weight how much confidence you place in channel-level digital attribution relative to aggregate, causal reads - the ground truth has moved toward the latter, not because attribution got worse at its job, but because there's structurally less trail left for it to follow.

Conclusion

The AI dark funnel isn't a new privacy scare - it's a new, larger version of a problem measurement teams have been circling for years: individual-level tracking degrades, and the brands still trying to explain performance with touchpoint-based attribution alone are the ones left guessing. The brands that will navigate this well are the ones whose measurement already assumes some of the journey is invisible, and that account for the full mix of media - offline and online - that's shaping it anyway.

That's the specific gap Animo is built to close: one causal MMM that reads TV, OOH, radio, print, and digital together, so the "invisible" part of the funnel gets attributed to the media that's actually driving it, instead of disappearing into an unexplained baseline. If you want to see what your model shows once offline media is treated as a real input rather than an afterthought, Animo's team can walk you through it.

Sources

  • Conductor 2026 AI search benchmarks (93% no-click AI Mode sessions, 83% no-click AI Overview sessions, 34% baseline), aggregated via Superlines: superlines.io
  • Ahrefs research on AI Overviews' impact on click-through rate (58% reduction, 300,000 keywords, Dec 2023 vs Dec 2025), reported via Businesswire, May 2026: businesswire.com
  • AI Overview prevalence data (48% of tracked Google searches, up 58% YoY, early 2026): thestacc.com
  • Digital Applied, "Marketing Attribution in 2026: What Still Works" (MTA adoption 47% vs 31% in 2023; MMM adoption 26% vs 9%): digitalapplied.com