Agentic Commerce and Your Paid Budget: What AI Overviews and AI Checkout Actually Change

The technical groundwork arrived in September 2025, when OpenAI launched Instant Checkout in ChatGPT alongside the Agentic Commerce Protocol, an open standard co-developed with Stripe. Etsy sellers went first, with over a million Shopify merchants including Glossier, SKIMS, Spanx, and Vuori slated to follow. The design point was that merchants keep control: orders, payments, and fulfillment stay in the merchant’s existing systems, and ChatGPT passes information between the two sides. OpenAI was explicit that product results in ChatGPT are organic and unsponsored, ranked on relevance, and that enabling Instant Checkout does not buy preference in those results.

That last detail matters more than anything else in this article for how you plan spend. The assistant’s own recommendation list, at least on ChatGPT, is not an auction. You cannot buy your way onto it. That is an earned-visibility problem, and it is the reason we treat AI recommendation coverage as a separate workstream from paid media: the mechanics of which we broke down in our guide to getting products into ChatGPT shopping recommendations.

How is agentic commerce different from traditional e-commerce?

The difference is where the comparison happens. In traditional e-commerce, a shopper compares options across tabs, retailer sites, and review pages, and your ads compete for attention at each of those stops. In agentic commerce, the comparison collapses into a single conversation, and the shopper arrives at your site already decided, or never arrives at all.

Adobe’s Q1 2026 data, drawn from more than a trillion visits to US retail sites, shows what that pre-qualification does to behavior. Visitors arriving from AI sources in March 2026 spent 48% longer on site, viewed 13% more pages per visit, and showed a 12% higher engagement rate than visitors from other channels. They are not browsing. They are confirming.

The consumer trust that drives this has moved fast: Adobe’s companion survey of over 5,000 US respondents found 39% have used AI for online shopping, 85% of those said it improved the experience, and 66% believe AI tools give accurate results. Separately, Google reported at Marketing Live in May 2026 that 75% of people say they make faster, more confident decisions using AI Mode in Search: a figure from an Ipsos Global Consumer Journeys study Google commissioned in December 2025, surveying 13,189 online shoppers across 25 markets who use AI Overviews or AI Mode for shopping.

Worth reading as directionally useful rather than independent, given who paid for it.

Is AI-referred traffic actually worth paying for?

On conversion quality, the case is now strong, and it flipped in a single year. In March 2025, Adobe measured AI-referred traffic to US retail sites converting 38% worse than non-AI traffic. In March 2026, the same measurement showed it converting 42% better, a record high. That is an 80-point swing in twelve months, and it is the single most important number for anyone arguing about whether this channel deserves budget.

Two cautions before you move money on that. First, this is retail data, so a B2B lead-gen advertiser should treat it as a signal about buyer behavior, not a benchmark to forecast against. Second, and more practically: a large share of AI-referred traffic is earned, arriving from an assistant’s organic recommendation. You cannot buy it directly. The budget question is not “how much do I bid for AI traffic” but “where does paid actually have a lever.”

What ad formats now run inside AI Overviews and AI Mode?

Four, announced by Google at Marketing Live on May 20, 2026, all built on Gemini and all still described by Google as tests or coming in the following months. They are worth knowing by name because they change what a good ad asset looks like.

• Conversational Discovery ads. Your ad answers the person’s specific question. Gemini builds creative tailored to that individual search, pulling out the features relevant to what was asked.
• Highlighted Answers. When AI Mode produces a list of recommendations, high-quality ads become eligible to appear within that list.
• AI-powered Shopping ads. On a product search, Gemini surfaces your most relevant items and writes a custom explainer for why that product may fit the shopper’s situation.
• Business Agent for Leads. Replaces the static lead form inside the ad with a Gemini-powered chat that answers questions from your website content.

Two structural details deserve attention. Google says these formats stay clearly labeled “Sponsored,” and that both AI Mode formats carry an independent AI explainer: Gemini’s own synthesis about the product or service, displayed next to the advertiser’s creative. In other words, the model will describe your product in its own words, right beside the words you paid for, drawing on what it can find about you. If your site is thin, contradictory, or hard for a machine to parse, that explainer is where it shows up.

Google also expanded its Direct Offers pilot, launched in January 2026 with brands including Chewy, Gap, and L’Oreal, adding promotion bundling, native checkout for merchants on Shopify’s Universal Commerce Protocol, and travel offers through partners like Booking and Expedia. Google’s stated prerequisite for all of it is a foundation of AI Max for Search, AI Max for Shopping, and Performance Max, which is to say the automated campaign types are becoming the entry ticket, not an option. We covered what that trade costs advertisers in our breakdown of Google AI Max replacing Dynamic Search Ads.

What infrastructure does agentic commerce actually require?

Machine-readable product and business data, everywhere an agent might look, and most sites do not have it yet. This is where paid and organic stop being separate budgets, because the same data feeds both the ad explainer and the assistant’s shortlist.

Adobe put numbers on the gap in April 2026 using its AI Content Visibility Checker, which scores how much of a page a large language model can actually read. Across the US retail sector, homepages averaged 75% and category pages 74%. Individual product pages (the ones that decide whether you get recommended for a specific item) scored 66%, meaning roughly a third of that content is invisible to the models. The spread between retailers was wide: 82.5% for the best-performing homepages against 54.2% for the weakest.

The practical checklist that follows from this is unglamorous. Structured product data with real attributes, not just images and marketing copy. Consistent pricing and availability across your site, your feed, and any protocol integration. Specifications, sizing, materials, compatibility, and returns written as text a parser can lift. Answers to the questions buyers actually ask before purchase, on the page rather than in a PDF. It is the same work that earns organic citations, which is why it belongs in one plan rather than two.

How should you reallocate paid budget this quarter?

Shift a controlled slice, instrument it properly, and fund the data work out of the same pot. Concretely, five moves:

1. Separate what you can buy from what you must earn. Ads in AI Overviews and AI Mode are purchasable inventory. Placement in an assistant’s own shortlist is not — OpenAI states plainly that Instant Checkout does not influence product results. Budget for the first; resource the second as visibility work.
2. Get your feed and product data audited before you raise bids. With retail product pages averaging 66% machine-readable, the constraint on most accounts is data quality, not spend. Gemini’s independent explainer will read whatever you have.
3. Treat the automated campaign types as an access requirement. Google has tied the new formats to AI Max and Performance Max foundations. If you are holding out on principle, you are opting out of the inventory — go in with tight negative lists, clean conversion signals, and honest reporting on what you can no longer see.
4. Measure AI-referred sessions as their own segment. Split assistant referrers out in analytics now and watch conversion rate and revenue per visit against your other channels. Adobe’s +42% is retail-wide; your number is the one that decides the budget argument.
5. Don’t build anything that assumes in-chat checkout. The industry moved to discovery in the assistant, purchase on the merchant’s property. Make the landing experience good enough to close a shopper who has already made up their mind somewhere else.

None of this is a reason to tear up the account. It is a reason to stop treating AI surfaces as an experiment line item and start treating them as where a growing share of consideration now happens: a shift we work through with clients as part of a broader paid search strategy rather than a bolt-on test.