Ads inside AI assistants like ChatGPT work differently from ads in search or social for one structural reason: they do not only compete with other ads; they compete with the assistant’s own answer. That is the argument Rokt, the New York-based e-commerce technology company, lays out in a recent analysis of how AI assistants are changing advertising. When a shopper asks ChatGPT for a recommendation, the model answers first, and a sponsored placement sits below it. OpenAI says the ad does not shape the answer, and the two systems are kept separate. The shopper still sees both on one screen, and that proximity changes what the ad can do.
The reason sits in how people read machine confidence. A May 2026 study from the University of Waterloo and University College London, published in Communications Psychology, found that people consistently judge AI systems as more confident than humans even when the content is identical. The researchers call it the illusion of confidence, and Rokt’s contribution is to trace what it does to a paid ad. If the model’s organic answer covers a category without naming a brand, the shopper has already anchored to a high-confidence opinion before reaching the ad, and the paid placement is left arguing against a trust signal the shopper formed in seconds and over-weighted. When the answer already favors the brand, the same psychology runs in reverse, and the ad reinforces authority the model has effectively granted.
A different kind of competition for the ad
In search, an ad competes with other ads for the same query. In an AI assistant, Rokt argues, the ad competes with the model’s own response, which most users treat as neutral and authoritative. Rokt’s framing treats the assistant as its own kind of environment, one where the platform’s output is the advertiser’s first and most persuasive competitor. The ad still runs regardless of what the model knows about the advertiser, but its ceiling is set partly by the answer above it. For the advertiser, that means the media buy is only half the task. The other half is influencing the answer itself, which no budget controls directly.
Buying the ad is not the same as being known
A brand can buy placements in ChatGPT without investing in how the model understands it, and the ads will still function. OpenAI matches ads to conversations using context hints from advertisers, landing page content, ad copy, and titles, so a well-targeted campaign earns impressions and clicks whether or not the model has a strong sense of the brand. Rokt’s point is that the ceiling is lower. Thin or poorly structured web content weakens the matching signals, so the brand appears in fewer relevant conversations, or in loosely relevant ones where shoppers do not engage.
The gap widens at a feature OpenAI built into the format. Users can tap “Ask ChatGPT” on any ad to pull the model’s organic take on the product. If the model holds strong, well-sourced information about the brand, the follow-up builds on the ad. If it does not, the follow-up can surface a competitor or return something generic, and that can undercut the placement the advertiser just paid for.
Structured data becomes an ad-performance lever
The work that decides what the model knows used to sit under brand marketing or public relations. Rokt’s analysis argues it is now operational and tied directly to paid performance. Google’s Universal Commerce Protocol, launched in January 2026 and backed by Amazon, Meta, Microsoft, Salesforce, Stripe, Shopify, Target, and Wayfair, is building the standard for how AI agents read commerce, and agents weigh brands on structured data, pricing signals, fulfillment reliability, and the consistency of product information across the web. PwC labels the discipline generative engine optimization. The blunt version, which Rokt draws from industry discussion around Google Cloud Next, is that the largest ad budget in the world cannot make up for missing catalog attributes. If the data is not structured for an agent to parse, the brand never enters the consideration set.
That reorders the funnel. The familiar sequence of awareness, consideration, and conversion gives way, in Rokt’s telling, to legitimacy, eligibility, recommendation, and conversion. Paid spend can accelerate any stage, yet it cannot manufacture legitimacy from nothing. Product titles, attributes, descriptions, and taxonomy become the foundation a paid strategy sits on, and a catalog that once looked like a housekeeping task turns into a performance lever.
Why measurement breaks in these environments
The channel also strains the tools marketers use to judge it. Shoppers in AI assistants ask several questions across sessions and devices, and the assistant often summarizes sources without sending a click, so last-touch attribution falls apart. The industry is moving toward incrementality as the baseline measure, though most measurement stacks were not built for a channel where the user never leaves the assistant. Rokt has pressed the same point across its wider commerce media work, arguing that incrementality, rather than attribution, is the standard that will decide which spend actually created a sale.
Why Rokt’s view carries weight
Rokt’s read on AI advertising rests on more than a decade spent at the point of purchase. Its AI engine, the Rokt Brain, weighs each transaction in real time against more than 1.95 trillion data points a year, and per Rokt by the Numbers, the network will power more than 10 billion transactions in 2026, reach 165 million monthly active users, and serve over 33,000 clients, including more than half of the largest e-commerce companies globally. Rokt Ads posts a 4.03 percent click-through rate, which the company reports is roughly ten times that of Google Display. In April 2026, Rokt was named for the first time in Gartner’s 2026 market guide for retail and commerce media networks, where chief revenue officer Craig Galvin called the transaction “one of the most valuable and underleveraged” interactions in commerce.
Rokt’s executives have been making a version of the AI-advertising case in public. In its recap of an appearance on eMarketer’s Behind the Numbers podcast, Rokt described the view of Ashley Firmstone, senior vice president of Rokt Ads, that brands should keep investing where they hold real customer data and understand lifetime value, rather than trade direct customer relationships for short-term reach on AI platforms. At an eMarketer summit, Rokt’s Sophie Donoghue made the adjacent point that the clearest sign a shopper is open to an offer is “when they’re literally buying something,” which is why Rokt treats the transaction, not the feed, as the highest-intent moment in commerce.
That focus connects to a discipline Rokt built into its own products: the willingness to show nothing when nothing would help. As The Silicon Review noted in its coverage of the company’s 2026 outlook, Rokt’s systems withhold an offer if it cannot clear a quality threshold. Chief commercial officer Elizabeth Buchanan summed up the philosophy in that 2026 outlook by predicting that the smartest brands will win by “doing less” with far more relevance. The same logic underpins the illusion-of-confidence argument. In an environment where trust is fragile and over-attributed, precision protects a brand better than volume.
The record has kept pace with the rhetoric. Rokt’s revenue grew past $800 million in 2025, according to the Gartner announcement, and the company landed on the 2025 Deloitte Technology Fast 500, as Dataconomy noted in its analysis of Rokt’s renewed partnership with Fanatics. Its Gartner debut arrived as commerce media accelerated, with US retail media spend projected to reach $69.33 billion in 2026, up from $58.79 billion in 2025 on eMarketer figures. As TechAnnouncer framed it, Rokt’s advantage is owning the transaction experience, the point where intent, attention, and trust converge at once.
Still early, and moving fast
The stakes are rising quickly. ChatGPT ads went live in February 2026 and crossed $100 million in annualized revenue within six weeks, with more than 600 advertisers enrolled, and OpenAI is already shifting from roughly $60 CPM pricing toward $3 to $5 per click. The broader pull is bigger than one platform: a December 2025 eMarketer forecast put AI platforms at about $20.6 billion of US retail e-commerce sales in 2026, rising to $144 billion by 2029. Google’s commerce protocol is only months old, and self-serve tools are still rolling out.
Rokt’s view is that the brands entering now have a short window to shape how they appear in these environments before the playbook hardens, but only if they pair the media buy with the upstream work of being legible to the model. Rokt has said it will keep watching the mechanics closely and expects the specifics to look different in six months. What it does not expect to change is the underlying dynamic. Inside an AI assistant, what the model already knows about a brand shapes the return on every dollar spent there, and the ad below the answer inherits whatever confidence the answer has already earned.
The post Why AI Assistant Ads Play by Different Rules: Rokt and the Illusion of Confidence appeared first on IoT Business News.
