For founders, CMOs, and ecommerce leaders deciding how much of their growth plan to build around AI assistants.
In April 2025, OpenAI added shopping to ChatGPT. Product results arrived as cards with images, reviews, and links to buy, and OpenAI said at the time that the results would carry no advertisements and that it would take no commission on purchases. When we wrote about that launch, we argued that ad-free recommendations could build a new kind of trust between shoppers and AI platforms, and we warned that commercial pressure could erode that neutrality over time.
Eighteen months later, three things have happened. ChatGPT tried to become a place where people complete purchases, then stepped back. The no-ads model ended, and ChatGPT now runs one of the fastest-growing ad businesses on the internet. And shoppers who arrive at retail sites from AI tools have gone from converting worse than other visitors to converting much better.
The short version for a leadership team is this. AI discovery is now a real commerce channel with both an earned side and a paid side, much like search became twenty years ago. The brands that treat it that way, rather than as an experiment, are likely to have an advantage over the next 18 months.
ChatGPT tried to become the checkout, then handed it back to merchants
Discovery moved into AI. The purchase, for now, still happens on the merchant's site.
In September 2025, OpenAI launched Instant Checkout, which let shoppers buy certain products without leaving the ChatGPT conversation. It was built on the Agentic Commerce Protocol (an open standard, developed with Stripe, that lets an AI assistant pass an order and payment to a merchant). Many observers, including us, read this as the start of AI platforms owning the full purchase.
That is not how it played out. On March 24, 2026, OpenAI published an update saying the first version of Instant Checkout did not offer the flexibility it wanted, and that it would let merchants use their own checkout experiences while it focused on product discovery. In practice, ChatGPT now helps people browse, compare, and decide, then sends them to the retailer's site to buy.
The lesson is useful, and it is not that AI commerce failed. Recommending a product turned out to be far easier than handling everything that comes after a sale, such as payments, returns, inventory, and customer service. For brands, this means the moment of decision has moved into AI tools, but the moment of purchase, and the customer relationship that follows it, is still yours to win or lose on your own site.
AI-driven shopping
18 months of ChatGPT shopping: no ads, then a $1 billion ad run rate.
-
Apr 2025
Shopping arrives in ChatGPT. OpenAI says no ads and no commissions.
-
Sep 29, 2025
Instant Checkout lets shoppers buy without leaving the chat.
-
Jan 16, 2026
OpenAI announces it will test ads.
-
Feb 9, 2026
Ads go live in the US on the Free and Go plans.
-
Mar 24, 2026
Checkout goes back to merchants. OpenAI steps back from Instant Checkout and focuses on discovery.
-
May 5, 2026
Self-serve Ads Manager opens in beta to US businesses.
-
Aug 31, 2026
$1 billion annualized run rate for ChatGPT Ads, in under 200 days.
-
Sep 16, 2026
Sponsored Agents test: shoppers can chat with a brand’s own assistant.
-
Oct 21, 2026
More than 100,000 advertisers have run campaigns since launch. Ads expand to 26 more countries.
Discovery moved into AI. The purchase, and the customer, still land with you.
Sources: Reuters, April 2025; OpenAI announcements, September 29, 2025, and January 16, February 9, March 24, May 5, August 31, September 16 and October 21, 2026. The run rate and advertiser count are company-reported and unaudited; the advertiser count is cumulative since launch.
The no-ads model ended, and that changes the trust question
ChatGPT now has an organic answer and a paid placement, the same two tracks brands know from Google.
The neutrality we described in 2025 lasted less than a year. OpenAI announced on January 16, 2026 that it would test ads, and the test began in the United States on February 9 for logged-in adults on the Free and Go plans. Paid subscribers on Plus, Pro, and business plans do not see ads. A self-serve Ads Manager opened to U.S. businesses on May 5. On August 31, OpenAI said ChatGPT Ads had reached a $1 billion annualized revenue run rate in under 200 days. (A run rate projects a recent month's revenue across a full year, so it describes pace, not money already collected.) On September 16, OpenAI began testing Sponsored Agents, a format that lets a shopper chat with a brand's own assistant inside ChatGPT, along with integrations for Shopify and HubSpot. On October 21, OpenAI said the ads platform had been used by more than 100,000 advertisers and was expanding to 26 more countries.
OpenAI states that ads are clearly labeled, appear separately from the answer, and do not influence what ChatGPT recommends. That is an important commitment, and brands should take it at face value while watching it closely. The practical effect is already clear, though. A shopping question in ChatGPT can now produce an organic answer that a brand earns and a paid placement that a brand buys. Most marketing teams already know how to manage that split in Google. Very few have applied it to AI assistants yet.
AI shoppers now arrive ready to buy
A year ago, AI referrals were a curiosity. The conversion data says they are now some of the best traffic a retailer gets.
Adobe Analytics, which measures more than a trillion visits to U.S. retail sites, has tracked AI referrals since October 2024. In July 2026, traffic from AI tools to U.S. retail sites was up 62% from a year earlier and up 1,219% since Adobe began tracking it. Those visitors converted to purchases at a rate 60% higher than visitors from other sources. That is a full reversal from March 2025, when Adobe found AI traffic converted 38% worse than non-AI traffic.
The likely reason is that AI tools compress the research stage. A shopper who has already compared options in a conversation arrives on the product page closer to a decision. Growth rates this steep usually start from a small base, so AI referrals are still a smaller channel than search for most retailers. But a smaller channel that converts 60% better deserves a dedicated plan, not a line item.
The numbers behind the shift
The OpenAI figures are company-reported and unaudited. The Adobe figures cover U.S. retail and may differ for other sectors.
AI and retail, 2026
AI shoppers went from converting 38% worse to 60% better. Brands still supply 1% of what AI cites.
ChatGPT’s ad business reached a $1 billion run rate in under 200 days, and more than 100,000 advertisers have tried it. Paid placement in AI is arriving fast; earned citations are still scarce.
| Finding | Figure | Source |
|---|---|---|
| AI-referred retail traffic · US | ||
| AI-referral traffic to US retail sites, July 2026 versus July 2025 | +62% | Adobe Analytics, via Digital Commerce 360, August 19, 2026 |
| Conversion rate of AI-referred retail visits versus other visits, March 2025 | −38% | Adobe, via TechCrunch, April 16, 2026 |
| Conversion rate of AI-referred retail visits versus other visits, July 2026 | +60% | Adobe Analytics, August 19, 2026 |
| ChatGPT Ads | ||
| Annualized revenue run rate, reached in under 200 days from launch | $1 billion | OpenAI, August 31, 2026 |
| Advertisers that have run campaigns since launch | 100,000+ | OpenAI, October 21, 2026 |
| Brand citations | ||
| Share of LLM-cited sources that are brand-owned websites, consumer goods | 1% | McKinsey, State of the Consumer 2026, June 22, 2026 |
OpenAI’s figures are company-reported and unaudited; the advertiser count is cumulative since launch, not active advertisers. A run rate annualizes recent revenue and is not booked annual revenue. Adobe’s figures cover visits to US retail sites measured in Adobe Analytics.
Revisiting the three risks we named in 2025
Each one turned out to be real, and each one has a practical response.
In 2025 we named three risks for brands: dependence on platforms, a discoverability gap for smaller brands, and data privacy. Each one is worth a second look.
Platform dependency is now concrete, because there is a paid track. If a competitor buys the sponsored placement on your category's most common questions, you need a strong organic presence to stay visible next to it. This is the same dynamic brands have faced in Google for two decades, and the same answer applies: invest in both tracks, and do not let either one carry the whole plan.
The discoverability gap is also real, though it is less about company size than about information quality. McKinsey's June 2026 State of the Consumer report found that only 1% of the sources AI models cite for consumer goods questions are brand-owned websites. AI tools rely on product data feeds, retailer listings, reviews, forums, and publishers. A small brand with clean product data and steady reviews can compete. A large brand with inconsistent listings can disappear.
Privacy remains an open question. AI assistants that remember past conversations can personalize recommendations, and the trade-off between relevance and data use has not been settled by regulators or by consumers. Brands should be careful about any tactic that depends on how a platform uses personal memory, because those rules are likely to change.
McKinsey adds a fourth point we did not anticipate. It describes a "dual front door," where shoppers can complete purchases either through a retailer's own AI tools or inside general AI platforms. Loyalty may follow whichever door a customer uses most, which makes a brand's own site and app more important, not less.
The three jobs of AI commerce
A simple way to organize the work across teams.
I group the work into three jobs, because each one tends to belong to a different team, and naming them makes ownership clear.
Be findable. This is about product data. Your catalog, descriptions, prices, and availability need to be accurate and complete wherever AI tools read them, including your site, your retail partners' listings, and any product feeds you submit. This job usually belongs to ecommerce or merchandising.
Be recommended. This is the earned track. Reviews, press coverage, expert roundups, and consistent descriptions across the web are what AI tools draw on when they decide what to suggest. This job usually belongs to marketing and communications.
Be bought on your terms. Since checkout has returned to the merchant, the page an AI referral lands on matters a great deal. It should load quickly, match what the AI tool described, and make buying simple. This job usually belongs to the web and digital product teams.
The paid track, if you choose to test it, sits on top of all three. Ads can buy visibility, but they cannot fix poor product data or a weak landing page.
What this likely means over the next 6 to 18 months
Narrower and more practical than the predictions we made in 2025.
Our 2025 post listed predictions for the next decade, including fully autonomous shopping, emotion-reading retail, immersive virtual storefronts, and biometric checkout. We are narrowing that view. The Instant Checkout reversal showed that AI platforms move faster on discovery than on transactions, and we see little public evidence yet for emotion-responsive or biometric retail at scale.
What seems most likely in the near term is more of what is already happening. OpenAI has said it plans more ad formats and more native ways for businesses to reach people inside ChatGPT, so expect the paid track to grow and become more interactive. AI-referred traffic is likely to keep rising as a share of retail visits. And McKinsey expects AI agents to take on more of the purchase itself over time, which would raise the value of being the brand an agent already trusts. Nobody outside these companies knows exactly how their systems choose what to recommend, so we would plan around these directions rather than any single feature.
If you want an outside read on how your products currently appear in AI shopping results, Camino5 can run that review with your team.
How to start today
Ask ChatGPT, Google's AI Mode, and Perplexity the shopping questions your customers ask most, and record which products appear, how yours are described, and whether a sponsored placement shows up. This gives you a baseline for both the earned and paid tracks.
Set up AI referral sources as their own segment in your analytics and compare their conversion rate to your other channels. If your numbers look anything like Adobe's, this will justify a dedicated budget.
Audit your product data on your own site and your top three retail partners for accuracy and consistency in names, descriptions, prices, and availability. Mismatched information is one of the most common reasons a product is described poorly or left out.
Review the landing pages AI referrals arrive on and confirm each one matches the product the AI tool described and offers a short path to purchase. Checkout is back in your hands, so this is where the sale is won or lost.
If you decide to test ChatGPT Ads, start with a small budget you are comfortable treating as a learning cost, run it for about 30 days, and measure it against your organic AI traffic. As a practical rule, 30 days is usually enough to see a pattern without committing a full quarter's spend; it is not an official benchmark.
The brands that do best in AI shopping over the next year will most likely be the ones with accurate product data, a steady reputation, and a site that makes the last step easy, because those are the things every version of these platforms has rewarded so far.
Sources
Reuters, "OpenAI rolls out new shopping features with ChatGPT search update," April 28, 2025
OpenAI, "Powering product discovery in ChatGPT," March 24, 2026
OpenAI, "Expanding access to AI with ChatGPT Ads," August 31, 2026
OpenAI announcement on ChatGPT Ads expansion to 26 countries, as reported October 21, 2026
Adobe, "AI traffic grows but retail sites lag in AI search visibility," 2026
McKinsey & Company, "State of the Consumer 2026," June 22, 2026