The 41% Problem: When AI Chatbots Eat Your Acquisition Funnel

The number that should be sitting on every performance marketing budget review this quarter is 41%. That's the share of people now using AI to research products, per IBM data cited at AdExchanger's Programmatic AI event in Las Vegas. For any team with a 6-to-8-figure paid traffic line item, the question is no longer whether the acquisition funnel is shifting away from classic search, it's whether the current vendor stack was procured for a world that already stopped existing.

The follow-on numbers matter just as much. Trust in AI chatbots to provide reliable information now sits at 62% of US adults, up from 50% a year ago. That's a 12-point jump in a single cycle. Budget owners who signed three-year SEO agency contracts in 2024 are now watching the underlying substrate of that spend erode in real time, and the infrastructure debate about how programmatic itself gets rewired for AI-driven buying is happening at the same moment. Two tectonic shifts, one procurement cycle.

The Numbers

Debra Aho Williamson, founder of Sonata Insights, framed the shift at the keynote as a structural relocation of purchase intent. As AdExchanger reported, Williamson put it bluntly: "wherever decisions happen, monetization follows." The 41% IBM figure is the leading indicator. The 62% trust number, from Sonata Insights and MRI Simmons research, is the lagging validator that the behavior isn't a novelty spike.

A few things to note about that 62%. First, it's specifically about interactions with AI chatbots, not AI-generated content pushed into feeds. That distinction matters because the same audience that trusts a ChatGPT response still, by most read of the market, hates AI slop in social. It's a permissioned trust, granted inside a conversational surface, not a general endorsement of synthetic media. Second, the jump from 50% to 62% in twelve months is the kind of consumer behavior slope that breaks attribution models before it breaks awareness.

The unit economics question sitting behind these numbers is uncomfortable. Product research traffic that used to originate from Google organic or paid search now originates inside a chat session where the brand often has no bidding surface, no keyword strategy, and no direct measurement hook. Williamson's observation that people using AI for product research "often encounter brands they've never seen before" is the tell: the substitution effect is not just moving traffic, it's changing which brands even get considered. Incumbent SEO advantage compresses. New entrant discovery expands. For a category leader that has spent seven figures a year defending branded search terms, that's a defensive moat draining out through a channel with no fence around it.

The 62% trust figure also has a compounding effect on conversion. Traffic sourced from a trusted chatbot recommendation arrives further down the funnel than traffic from a paid search click. That should, in theory, lift downstream conversion rates, but only for brands the model surfaces. For everyone else, the funnel simply gets shorter and narrower.

What's Actually New

Three things are genuinely different this cycle, and platform leads need to separate them cleanly before making any procurement call.

First, the infrastructure fight inside programmatic itself is now openly factional. The IAB-backed Agentic RTB Framework (ARTF) is at odds with the consortium behind the Ad Context Protocol. That's not a minor spec disagreement, that's two competing visions for how AI agents transact against ad inventory. Anyone standardizing on IAB specs as a default assumption needs to re-examine that reflex. The ARTF path and the Ad Context Protocol path imply different vendor ecosystems, different integration surfaces, and different long-term switching costs.

Second, containerization is now contested at the architecture layer. Cognitiv, a custom bidder startup, claims a server-to-server model it says is "900% more powerful than containers." The technical argument is that containerized environments, where an SSP hosts a demand-side model, offer limited compute in real-time moments and no persistent storage. Meanwhile, Chalice AI, one of Cognitiv's early rivals, just announced a partnership with SSP Equativ that puts Chalice's models directly into Equativ's containers. So the market has two custom bidder vendors, one arguing containers are the ceiling, the other productizing containers as the distribution channel. Both cannot be right, and the choice between them is a bet on where compute lives in the bid path.

Third, publisher collectivization has crossed from advocacy into standards work. SPUR (Standards for Publisher Usage Rights), the BBC-led coalition with the Financial Times, The Guardian, Sky News, and The Telegraph, is building shared technical standards and licensing frameworks for AI training and retrieval. A parallel Danish collective exists. RSL, co-founded by Doug Leeds, is working on an open standard for machine-readable licensing terms directed at AI crawlers. Leeds's framing is direct: LLMs "cannot claim that, because it's on the web, it is available for them to retrieve, to train on, to basically substitute their own content for yours." That's a legal posture backed by a technical artifact, which is a stronger position than pure litigation.

What's Priced In for Performance Marketing

Most performance marketing teams have already priced in the general direction, chatbots are eating search, publishers want to get paid, programmatic gets more automated. What isn't priced in is the timing and the vendor-selection consequence.

The 62% trust figure being a year-over-year jump from 50% is not priced in. Most 2026 media plans I've seen modelled linear growth in AI-influenced discovery, not a 12-point step function in trust. That means the assumption that "we have another 18 months before this really matters" is probably wrong by roughly 18 months.

The bifurcation between ARTF and the Ad Context Protocol is not priced in on the buy-side. Most agency and in-house programmatic teams are still treating "IAB compliant" as a synonym for "safe default." When the IAB itself is backing one of two competing frameworks, that default assumption needs an explicit review. This is a build-vs-buy conversation for anyone running custom bidding: commit to one framework early and eat the switching cost later, or wait and eat the lost-quarter cost of being non-optimized now.

The Head of Platform at any mid-market ad-tech or performance shop should be asking their VP Eng this week: does our roadmap treat ARTF and the Ad Context Protocol as interchangeable, and if so, who on the team is doing the diligence to prove they still are by Q4? That question surfaces the hidden coupling cost before it hardens into technical debt.

The Cognitiv versus Chalice split is also not priced in. The industry narrative has been "custom bidders are the future," treated as monolithic. It's now two distinct architectural bets: bring your model to the SSP's container, or run a server-to-server model outside the exchange. The compute economics, the data residency posture, and the persistent-state capabilities differ meaningfully across those two paths. Regulatory exposure differs too: server-to-server with persistent storage raises very different GDPR questions than a containerized ephemeral model.

Contrarian View

The consensus read is that AI chatbots are the new search and every traffic acquisition strategy needs to rebuild around that surface. I'd push back on the completeness of that read.

Google unveiled the biggest overhaul to its Search box in 25 years. That's not a company conceding defeat, that's an incumbent with distribution, default placement, and a decade of query data doing what incumbents do. The framing of chatbots replacing search assumes the search box stays static while chatbots evolve. It won't. The more likely 2027 endpoint is a hybrid surface where the distinction between "search" and "chatbot" is a UI detail, not a category.

There's also a Microsoft-shaped warning in the story. Former VP Mat Velloso claims Microsoft has "missed the AI wave," and only 3% of paid Copilot users are actively using the tool. That 3% number is the counter-narrative to the 41% product research figure. Consumers are trying AI. Enterprises are paying for AI. Whether either group is using it durably enough to justify infrastructure rebuilds is a separate question. Traffic teams re-platforming their entire acquisition strategy on the assumption that chatbot-mediated discovery is permanent should look at the Copilot engagement number and ask whether they're modelling the top of an S-curve or the top of a hype curve.

Key Takeaways

• The 41% product-research number is the leading indicator, the 62% trust number is the confirmation. Twelve points of trust growth in one year compresses the timeline for any acquisition strategy still anchored to classic search.
• ARTF versus Ad Context Protocol is a real fork, not a spec footnote. Platform leads should stop treating IAB-endorsed as an automatic safe default and require an explicit framework decision from their programmatic vendors this quarter.
• Custom bidder architecture is now a binary bet. Container-hosted models (Chalice plus Equativ) versus server-to-server with persistent storage (Cognitiv) imply different compute economics, different data residency exposure, and different long-term vendor lock-in.
• Publisher collectivization is a European phenomenon by regulation, not by choice. SPUR and the Danish collective work because CMO-style bargaining is legal in Europe and prohibited in the US without a government exemption. US publishers are structurally weaker negotiators until that changes.
• Teams evaluating chatbot-mediated traffic strategies should be asking themselves whether they are betting on a durable shift or on a Copilot-style adoption curve where trial outpaces sustained use. The 3% Microsoft engagement figure is the discipline the 41% IBM figure needs.