When buyers ask AI what to buy: where product recommendations come from

In September 2025, OpenAI and Stripe announced Instant Checkout: buy a product without leaving ChatGPT, starting with Etsy and a million Shopify merchants on the way. It looked like the start of a world where AI assistants close the sale. Then, in March 2026, OpenAI quietly pulled back. According to reporting first surfaced by The Information and confirmed by CNBC, the company found that people were using ChatGPT to research and compare products but completing the purchase on the merchant's own site, where they felt comfortable. The lesson the leader in agentic commerce learned, in public, is the one your strategy should start from: AI is where buyers decide, not where they check out. That makes AI shopping a discovery problem, and discovery is won long before anyone reaches a cart.

AI shopping turned out to be a discovery engine, not a store

The most useful fact about AI commerce in 2026 is a failure, not a launch. OpenAI built in-chat checkout with Stripe, signed up Etsy and Shopify brands like Glossier, Vuori, Spanx, and SKIMS, then walked it back within months. Forrester read the retreat plainly: the hard problem was never payments, it was keeping merchant pricing and inventory perfectly synced in real time, and shoppers simply preferred to finish buying in a familiar storefront. ChatGPT is now positioned as a discovery surface that routes qualified buyers to merchants and earns referral economics, not a checkout.

For a marketing leader, this reframes the entire question. The goal is not "get my checkout button inside ChatGPT." It is "be the product ChatGPT recommends when my buyer is deciding." Soar is a community marketing agency that has run 4,200+ community campaigns across 280+ brands since 2017, and the pattern we keep seeing holds here: the money moves through the storefront, but the decision is made upstream, in the research. If AI owns the research moment, then your visibility inside that moment is the asset worth building. The checkout was a distraction. The recommendation is the prize.

How big is AI shopping traffic in 2026?

This is no longer a trend to monitor; it is a channel with measurable revenue behind it. Adobe Analytics found that AI-referred traffic to US retail sites grew 393% year over year in Q1 2026, after spiking 693% during the 2025 holiday season (Adobe). The traffic is not just larger, it is better: in March 2026, visitors arriving from AI assistants converted 42% better than other traffic, a sharp reversal from a year earlier when AI traffic converted worse.

The engagement profile explains why. Adobe reports AI-referred shoppers stay 48% longer on site, browse 13% more pages, and engage at a 12% higher rate than non-AI visitors, because they arrive pre-qualified by a research conversation. The measured sources are the assistants your buyers already use: ChatGPT, Perplexity, Copilot, Gemini, and Google AI Mode. The takeaway for budget owners is direct. A channel growing triple digits, sending traffic that converts better than your existing mix, is not one to "wait and see" on. The brands captured in those answers now are compounding an advantage while their competitors treat it as experimental.

Where do AI product recommendations actually come from?

When an AI assistant answers "best X for Y," it is summarizing the open web, and the open web's most trusted product voice is community discussion. The clearest evidence yet comes from a Novi analysis of 10.7 million ChatGPT citations across beauty product queries between January and May 2026, spanning more than 98,000 source websites. Reddit ranked first by a wide margin, ahead of Who What Wear, Wikipedia, Sephora, and Allure. Those five sources alone drove roughly a third of all citations (PPC Land).

Read that carefully: a community forum outranked the category's biggest retailer and its most established magazine, combined, as a source for what to buy. The reason is the same one that makes AI shopping work at all. Models weight sources buyers find credible, and buyers trust other buyers describing real experience over marketing copy. This is the mechanism we documented in how Reddit became the biggest LLM citation source, now showing up specifically in purchase recommendations. The implication for your brand is uncomfortable but actionable: the content deciding whether AI recommends you is mostly not on your website. It is in threads other people wrote.

A community forum outranked the category's biggest retailer and its most established magazine, combined, as the source for what to buy.

Can you pay your way into AI shopping results?

No, and this is the part that breaks most marketing playbooks. OpenAI's shopping results are organic and unsponsored, ranked on relevance to the query rather than ad spend (Retail Dive). There is no bid, no promoted slot, no media budget that buys you into the buyer's guide. The merchant programs that ChatGPT, Perplexity, and Google AI Mode have launched let you submit a product feed with prices, specs, reviews, and images, which improves the accuracy of how your product is represented once it is chosen. They do not decide whether it gets chosen.

This is worth saying twice to anyone who controls budget, because the instinct is to treat AI shopping like paid search. The Agentic Commerce Protocol that Stripe and OpenAI built gives merchants "full control over what's sold, how their brand shows up, and how orders are fulfilled" (Stripe). Control over presentation is not the same as control over selection. Selection is earned upstream, by what the web says about you. If your competitor is the consensus pick in the threads and reviews the model reads, a perfect product feed will render your competitor's win beautifully.

ChatGPT vs Perplexity vs Google AI Mode: how each surfaces products

The three major AI shopping surfaces differ in how they show products and where they send buyers, but they reward the same underlying input. ChatGPT runs shopping research on a model tuned for the task, returns organic buyer's guides, and routes purchases to merchant sites. Perplexity offers "Buy with Pro" checkout for Pro users on select merchants and leans on a merchant catalog plus live web research (Perplexity). Google AI Mode supports in-chat checkout powered by Shopify and pulls from Google's Shopping graph and the broader index. The table below sets expectations for your team.

The pattern that matters across all three: a structured feed makes your product eligible and accurate, but the recommendation itself is driven by web-wide signals, especially reviews and community discussion. That is why we tell clients to treat feeds as table stakes and earned community presence as the differentiator. The same threads that win you a Perplexity citation tend to win you a ChatGPT pick, because the credible source about your category is the same regardless of which assistant is reading it.

What this means for your brand's discovery strategy

If AI is the new research layer and recommendations are earned, not bought, then the work is to become the product real buyers recommend in the places AI trusts. That means a deliberate presence in the communities where your category is discussed, Reddit foremost among them, plus the review ecosystems and comparison content that models cite. This is not a campaign you run for 30 days. It is durable presence that accumulates until the consensus answer in your category includes your brand. We lay out the broader mechanism in how community marketing drives AI visibility.

The honest part: this is slow and it is hard to fake, which is exactly why it holds value. Seed praise with fake accounts and you trip spam filters and teach the model your brand is low-credibility. Post promotionally in skeptical subreddits and you get downvoted, which is a negative signal the model reads too. The path that works is unglamorous: real participation, useful contributions, and genuine recommendations earned over months, especially in categories where buyers compare in public. For brands selling to consumers, the trust mechanics are the same ones we cover in community marketing for DTC brands. Before committing budget, find out where you currently stand with a step-by-step AI visibility audit.

Who should prioritize AI shopping visibility now?

This is an urgent priority for brands in categories where buyers actively research and compare before buying: beauty, consumer electronics, home goods, software, supplements, anything with specs, reviews, or trade-offs to weigh. Adobe's data shows AI shopping concentrating exactly in these research-heavy categories, and the Novi findings show community sources dominating them. If your category already has active Reddit threads and review discussion, those sources are being summarized into recommendations today, and the only question is whether your brand is in the answer or absent from it.

The brands that can reasonably wait are those selling low-consideration, impulse, or purely local products where buyers do not research first, or brands without product-market fit, where forcing community presence produces thin signals the model ignores. The conditional we give clients: if buyers compare your category in public before purchasing, AI shopping visibility is a now problem, because the recommendations forming this quarter are built on the content that exists this quarter. If your buyers do not research, your spend belongs elsewhere. Either way, the diagnostic is the same as figuring out why ChatGPT recommends your competitor instead of you: look at what the model is reading.

:::