On July 14, 2026, OpenAI quietly expanded its ChatGPT ad pilot into the UK, Mexico, Brazil, Japan and South Korea, and started testing something new: ad units that bundle several competing brands into one placement chosen by the model, not by whoever bid highest. The same week, Google pushed its AI Mode shopping formats out of testing and into US rollout, and released Agentic Chrome, a browser built around Autobrowse, a feature that clicks, compares and checks out on a person's behalf. Search trackers now put the share of US Google queries ending with zero clicks at 68%, up from roughly 48% just five weeks earlier. The click didn't fade slowly. It fell off a shelf, and something else picked up the job of deciding what you buy.
Who is your ad actually talking to now, a person or a ranking function?
Increasingly, it's the ranking function. OpenAI's multi-advertiser unit means a handful of brands get grouped into one answer and the rest simply don't appear, regardless of budget. Google's Business Agent for Leads and its Conversational Discovery ads sit inside the AI Mode reply itself, not below it. Perplexity Comet and ChatGPT Atlas go further still: they execute multi-step shopping tasks, comparing prices and specs across sites, without a person clicking a single link along the way. Five weeks ago the fight was about getting cited in an AI Overview while a human still decided whether to click through. Now the click has been delegated too, and the decision that used to belong to a browsing person increasingly belongs to the software doing the browsing.
How do you sell to a reader that doesn't feel anything?
You mostly don't, and that's the uncomfortable update to an argument this newsletter made in June: that creative was the last lever left once targeting became a commodity. Creative still wins the human half of the funnel. It has nothing to grip in the agent half. Behavioral economists Logg, Minson and Moore found people show algorithm appreciation, not aversion, on quantifiable comparison tasks like price and spec shopping, the exact opposite of the algorithm aversion that governs trust in creative or strategic judgment. An agent built to shop is built to exploit that appreciation. It reads structured product feeds, review-score distributions, return-policy text and price history. It does not register a hook in the first frame, because it was never built to feel one. The funnel didn't shrink. It split into a human lane that still responds to story, and an agent lane that only responds to data it can parse.
When the gate narrows to three brands, what does a bid actually buy?
Less than it used to, and the shape of the loss is different from a normal price increase. An open auction with ten organic results and several paid slots gave most bidders some odds, worse odds at a higher price, but odds. A model-curated bundle of three or four brands is a binary cutoff: you're in the bundle or your budget buys nothing at all, no matter how much of it you spend. Amazon is reportedly defending its own $56 billion advertising business against exactly this bypass, agents that shop across the open web without ever loading an Amazon results page. When the company that invented pay-to-rank retail media treats agentic bypass as a threat to its own ad line, the shift is structural, not a trend piece. The AI shopping assistant market is already tracking toward $12.8 billion by the end of 2026. If a category's paid inventory consolidates from eight visible placements down to three agent-selected ones, competition for each surviving slot can plausibly triple, and the cost per qualified result for whoever wins it rises well past what the shrinking unit count would suggest on its own.
The ad used to compete for a person's attention. Now half of it competes for a data structure that doesn't have any.
What should change in this quarter's budget?
Start with the feed, not the campaign. Audit your product data, review scores, return policy and price history for the same clarity you'd demand from a landing page, because that's what an agent actually reads before it ever reaches your creative. Pilot OpenAI's multi-advertiser unit and Google's Business Agent for Leads now, while the inventory is new and the field bidding for it is thin; that window closes as soon as every competitor notices the same numbers you just read. Tag agent-referred sessions separately from human ones before the September review, so a real shift in demand doesn't get mistaken for a phantom lift, or missed entirely. And keep the creative budget. Just stop asking it to do a job half your audience is no longer built to feel.
Want this kind of thinking on your brand?
We build brand strategy, AI content and performance for the AI era.
START A PROJECT →