AI Chatbots Capturing Consumer Intent Signals Invisible to Ad Platforms

Full analysis

Debra Aho Williamson of Sonata Insights, on stage at Beet.TV's AI Summit during Cannes Lions 2026, made a plain argument: the purchase-intent journey that used to spray across a dozen Google queries and a trail of retargeting cookies now happens inside one ChatGPT thread. She asked ChatGPT about a Goop skincare product, got steered to a competitor, bought on Amazon. Amazon got the sale. Nobody but OpenAI saw the intent. That's the whole story in one anecdote.

What's actually being decided (briefing frame): whether ad-tech operators should reprice the value of open-web behavioral signal now, on the bet that intent is migrating into walled chatbot threads they can't see. Reversibility: the strategic read is Type 1 for anyone whose product is built on open-web signal arbitrage; the tactical audit is Type 2 and cheap. Forcing function: none hard, but the attribution gaps show up on their own timeline whether or not you look.

The Skeptic. The anecdote is one analyst buying moisturizer. That is not a demand-side signal collapse, that is a Tuesday. The 1,641% chatbot ad-growth number is a magician's flourish. It's 1,641% of almost nothing, landing at ~$960 million against a global digital ad market north of $300 billion. Rounding error. For this to bite, three things all have to be true: OpenAI actually monetizes the intent instead of just hoarding it, consumers trust chatbot product picks at scale, and advertisers move budget into an unauditable black box. None of those is reliably true in 2026. Locked-up signal generates revenue for no one, including OpenAI. Plain version: the intent may be trapped inside ChatGPT, but trapped is not the same as sold.

The Market Analyst. Here's the part the Skeptic underweights. OpenAI's advantage compounds every quarter it doesn't monetize, because it keeps accumulating the richest purchase-consideration data since Google indexed the web while owing nobody a rev share yet. The wrong trade is short Google. Search revenue has outlived "search is dead" for fifteen years, and Google monetizes the query on-page regardless. The exposed names are the independent data layer: DSPs, DMPs, and identity players whose entire pitch is turning open-web behavior into targetable audiences. If intent drains into walled threads, that addressable market shrinks structurally, not cyclically. Plain version: the middlemen who resell signal the giants are about to stop leaking are the companies that take the hit.

The Operator. Forget 2030. What breaks first is the paid-search-to-conversion attribution chain, and it breaks quietly. As query volume fragments into chatbot threads that never hit Google or Bing, upper-funnel search signal thins out. By Q3 2026 the attribution models tying search behavior to conversion start showing unexplained gaps. Not because performance dropped. Because the signal chain snapped upstream and nobody got an alert. The move is unglamorous and immediate: audit what share of your current intent data comes from search-derived cookies versus owned first-party signal, and do it before the next planning cycle bakes the broken assumption into a budget. Plain version: your dashboards will look fine right up until they can't explain why conversions happen.

The Customer / End User (the brand advertiser). From the brand's chair this is worse than losing a channel. Williamson's own line: Amazon got the sale, ChatGPT got the intent, the brand got nothing it can act on. The retailer captures the transaction but not the consideration, so retail media, the thing everyone spent three years building, sees the bottom of the funnel and goes blind on the top. And the chatbot didn't just fail to inform the brand, it actively recommended a competitor mid-conversation. That's not a measurement gap, that's a distribution channel that can reroute your customer before you know they exist. Plain version: the machine your customer trusts for advice can send them to a rival, and you never even get to bid.

The CFO. The audit costs a few analyst-weeks and buys real information, so do it. What I won't do is reprice the whole media plan on a $960 million market and a forecast that lands in 2030. The near-term money is still where the budget sits: AI search-adjacent advertising at a projected $26.42 billion in 2026, over 80% of AI ad spend running next to AI content rather than inside the chatbot. That's the line item that moves this year. Chasing chatbot-native inventory now means paying to be a beta tester in an unauditable environment with no clearing price. Plain version: fund the audit, watch the trend, don't rebuild the P&L around a market that's still a rounding error.

Where they split. Three real disagreements. First, the Skeptic and the Market Analyst are arguing about time, not direction. Both agree signal is migrating; they disagree on whether "compounds slowly and monetizes never-yet" is a threat you act on in 2026 or a story you revisit in 2028. Second, the CFO and the Operator disagree on urgency of spend versus urgency of measurement. The CFO says don't move money; the Operator says the money's already leaking through a broken attribution model whether you approve it or not. Those aren't in conflict once you separate the audit (cheap, now) from the budget shift (premature). Third, the Customer lens raises something the investor frames miss entirely: the chatbot is an active recommender that can hand your customer to a competitor, not a passive signal vault sitting quietly on purchase intent. That's a channel-control problem, not a data problem.

What it hinges on. One belief: does OpenAI turn intent into an ad product buyers can actually target and audit, and when. Everything downstream, the multiple compression on independent identity players, the retail-media blind spot, the attribution gaps, follows from that one event. Until it happens, the Skeptic is right on revenue and the Market Analyst is right on trajectory, and both can stay right for a while. The council leans this way: the strategic threat is real and directional, the near-term revenue is not, and the only mistake that's cheap to avoid is letting a broken signal chain quietly corrupt your attribution while you wait for the big shift.

What to de-risk now: audit search-derived versus first-party signal share, and pressure-test retail-media attribution for upper-funnel blind spots. Both are Type 2 moves. Neither requires believing the 2030 number.

Prediction: OpenAI will launch a paid product that lets advertisers or brands buy placement, sponsorship, or targeting against ChatGPT purchase-intent conversations, generally available to US advertisers, before the 2027 upfront/NewFront season concludes in May 2027.

Confidence: Medium. The incentive is overwhelming, but OpenAI's timing is its own to control.

Why: OpenAI is sitting on the richest purchase-consideration signal built since Google indexed the web, and eMarketer already projects chatbot-native ad spend growing 1,641% in 2026 off a near-zero base, which tells you the demand is forming faster than the supply exists. A company burning cash on compute does not leave a monetizable intent stream unpriced indefinitely once rivals (Google's AI answers, Perplexity's ad tests) are visibly racing for the same surface. The opposite outcome, OpenAI keeping the thread ad-free to protect user trust, is the less likely one because the pressure to show ad revenue against its spend only compounds, and a controlled ad product lets it monetize without breaking the conversation. What would have to break for this to fail: OpenAI decides subscription and enterprise revenue alone justify the valuation, and holds the line through the buying season.

Revisit by 2027-05-31: We're right if OpenAI has made a paid intent-targeting, sponsorship, or ad-placement product against ChatGPT conversations generally available to US advertisers by then. We're wrong if ChatGPT's consumer chat surface still carries no buyable advertiser placement of any kind at that date.

That's the call worth grading. Not whether the signal is migrating, everyone at Cannes already agreed on that. The contested question is whether OpenAI is willing to be the one who turns the conversation into inventory, and I think the cash-burn math answers it for them.