How Shoppers Use ChatGPT to Research Products Before Buying
The AI shopping research tool that's already beating Google at converting browsers into buyers.
ChatGPT now runs a full shopping research session before a customer ever lands on a brand's site. It asks follow-up questions, narrows a field of options down to a handful of finalists, and hands the shopper a comparison built from structured data and outside reviews. That's a different job than a search engine does, and it's already changing who gets found, who gets bought, and who gets left off the list entirely. Brands treating this as just another SEO channel are going to lose to brands that understand the rules have changed, and most of the market hasn't figured that out yet. Anyone still optimizing for keyword density is optimizing for a system that no longer decides who wins.
The scale of shopping queries ChatGPT is now handling
ChatGPT has 700 million weekly users, and shopping questions now make up roughly 50 million of its daily queries. Stackline puts U.S. shopping questions at over 84 million a week, a volume that reached more than 8% of Amazon's own weekly search traffic in under a year. That's not a novelty sitting off to the side of retail. That's a real, fast-growing rival parked next to the biggest product search engine in the country, and it got there in months, not years.
Adoption skews young, and that's the detail worth sitting with. Roughly 17% of consumers, an estimated 45 to 50 million people in the U.S., now use AI regularly to shop. Break that number apart: 37% of Gen Z shoppers use AI to shop online versus 29% of Millennials, while half of Baby Boomers say they simply don't trust it yet. The shoppers driving the fastest growth in direct-to-consumer revenue are the same ones adopting AI shopping first, which means this isn't a bet on some future buyer. It's already happening, among the exact customers brands spend the most money trying to reach.
How ChatGPT's research process differs from a keyword search session
A normal search session looks like this: type a keyword, open a dozen tabs, compare specs by hand, and still feel unsure at the end. Decision fatigue isn't a bug in that process. It's baked into it.
ChatGPT Shopping works differently, and the difference sits in the interrogation up front. A shopper describes a need in plain language, something like "quietest cordless vacuum for a small apartment," and the model asks back: what's the budget, what floor type, how much storage space, what's it actually for. Only then does it spend a few minutes researching and return four to six finalists, laid out with images, pricing, availability, specs, and the trade-offs between them. The shopper can mark an option "not interested" or ask for "more like this," and the model adjusts on the spot. The deliberation that used to happen across browser tabs now happens inside the conversation, before a single product page gets opened.
What the model reads to build that list has nothing in common with what ranks a page on Google. Traditional search rewards keyword density, backlinks, and domain authority. ChatGPT Shopping pulls instead from structured product data, third-party reviews, benefit-driven descriptions, and mentions on places like Reddit and independent review sites. That distinction cuts hard in one direction, and it should worry any brand that's built its whole strategy around Amazon reviews: a product buried in thousands of them is invisible to this system, because Amazon blocks OpenAI's crawlers outright. A product with a fraction of the reviews, but spread across Reddit threads and independent sites, beats it anyway.
The three-to-five-minute research window is a signal in itself. Nobody spends that long asking about a phone charger. This is built for considered purchases, the kind where getting it wrong costs real money or real regret. Checkout still happens somewhere else: the shopper lands on a merchant's site already filtered down to a short list, but still checks the actual page and confirms the seller is legitimate before paying. A shopper clicking through from ChatGPT has already been vetted against their own stated needs. They arrive in verification mode, not discovery mode, and a brand's product page has to be built for that mode specifically, not for the browsing mode Google traffic still shows up in.
Why shoppers who arrive from ChatGPT convert at higher rates and spend more
Shopify's platform-wide numbers make the case plainly. Referral sessions from AI chatbots, ChatGPT, Perplexity, Gemini, Copilot, Claude, Grok, grew more than eightfold year-over-year on Shopify stores in the first quarter of 2026, and AI-referred orders grew almost thirteenfold in that same stretch. Visitors who arrive from AI convert at nearly 50% higher rates on product pages than visitors from organic search, a pattern holding in 23 of 25 merchant categories. More than half of AI-referred sessions land directly on a product page, against just 20% for organic search. These shoppers skip the browsing entirely and go straight to the thing they came for.
The mechanism isn't complicated, so it's worth stating plainly instead of hedging around it: a shopper who's already run through ChatGPT's clarifying questions and eliminated the alternatives doesn't need to be persuaded on the product page anymore. That page's job has shifted from selling to confirming. If the facts on the page don't match what ChatGPT told the shopper, the sale disappears in seconds, no second chance, no browsing around to reconsider. Accuracy on a product page isn't a trust signal anymore. It's the entire conversion mechanism, full stop.
Black Friday data sharpens the point. Shoppers referred from ChatGPT converted on Amazon at 1.7 times the rate of shoppers referred from Google. Retailers with AI agent integrations saw roughly seven times better sales growth than those without, a gap too wide to write off as noise or seasonal luck.
Why most brands are undercounting the revenue ChatGPT is already sending them
The trap is simple, and it's already draining marketing budgets without anyone noticing. A shopper gets a recommendation from ChatGPT, opens a new tab, types the brand name into Google, and buys. Every analytics tool in common use credits that sale to Google Search, or to direct traffic, never to ChatGPT. The recommendation did the work. The credit lands somewhere else entirely, and most marketing teams have no idea it's happening.
Some DTC brands have run into this exact pattern: organic traffic dropped while revenue held flat, and it took months of digging to find out why. The answer traced back to ChatGPT links landing shoppers directly on product pages, skipping the homepage and every category page along the way. On a dashboard, an 18% organic drop reads as decline. In the actual business, it was growth the whole time, invisible to the tool measuring it.
The same behavior that lifts conversion, going straight to the product, is exactly what makes the channel hard to see. A direct-to-checkout session doesn't leave the referral trail attribution tools are built to catch. Post-purchase surveys stop being a nice-to-have here; they become the only real way to see what's going on, since the dashboard won't tell you. Brands making budget decisions without accounting for this end up pouring money into channels that get credit on a report, while starving the channel actually closing the considered purchases.
What ChatGPT actually reads when it evaluates a product, and what makes a brand invisible
ChatGPT Shopping doesn't see a homepage's design, its navigation, or its brand polish. It reads structured data, review text, and whatever descriptive content sits out in the open for a crawler to find.
What consistently gets surfaced: complete structured data (brand name, model number, variant-level identifiers like GTIN or MPN, live pricing, stock status), schema rendered server-side rather than loaded in late by JavaScript, product descriptions written around a use case ("built for small apartments with pets") instead of a bare spec sheet, and reviews pulled from more than one outside source, not just whatever sits on the brand's own site.
Invisibility comes from a short, recognizable list of failures, and the same few mistakes show up over and over. Variant data that's poorly structured can return the wrong product variant entirely, mismatching what the shopper asked for. Pricing that's gone stale conflicts with the live merchant page and can undermine a product's standing in the comparison. Descriptions that list specs with no context about who the product is for lose out to a competitor's page that answers the question directly. And there's the Amazon blind spot again: a product with a deep well of Amazon reviews and nothing anywhere else is unreachable, while a product with a smaller but more scattered footprint across independent sites takes the spot instead.
There's a training data problem underneath this too. Newer products and younger brands end up under-represented compared to established products sitting on years of accumulated reviews, a real structural disadvantage for any DTC brand that hasn't actively built up third-party coverage. Clean, structured product data isn't a technical chore handled once and forgotten. It's the raw material the model works from to decide whether a brand gets mentioned at all, or passed over without anyone ever finding out why. Platforms such as Kinect, an AI sales and catalog intelligence layer for DTC stores, exist specifically to keep that data structured and surfaced across AI shopping channels.
Where agentic commerce is taking this, and why the window to act is narrow
Everything described above still ends with a human clicking "buy." Agentic commerce is the step after that: an AI agent that searches, compares, and completes the purchase with little to no human input at each stage. That step is closer than most brands are planning for, and waiting for it to arrive before building toward it is a mistake.
The infrastructure is already being built, by more than one company at once, and none of them are waiting for the others. Stripe, OpenAI, and Meta released the Agentic Commerce Protocol under an open Apache 2.0 license. ChatGPT's Instant Checkout launched in 2025 with Etsy as its first live merchant partner, with additional Shopify merchants and major retailers joining soon after. Google's Universal Commerce Protocol launched in January 2026 with Shopify, Wayfair, Target, Etsy, and Walmart among its founding merchant partners. Anthropic's Model Context Protocol, donated to the Linux Foundation's Agentic AI Foundation in December 2025, standardizes how an AI agent connects to outside tools and data sources in the first place. Three protocols, three companies, all converging on the same idea within months of each other. That's not a coincidence. That's a sign the direction is settled even if the pace isn't, and any brand still debating whether this is real has already missed the part of the argument that mattered.
Large retailers have already picked their model, and it's worth naming: Target, Lowe's, and Home Depot have connected their product feeds through the Agentic Commerce Protocol for discovery, while keeping the actual transaction on their own platform. AI handles the finding; the retailer keeps the closing. Smaller DTC brands can copy that split without building any of the underlying infrastructure themselves.
What none of them can copy is time. The brands showing up correctly in ChatGPT's shopping results right now are building a data and review footprint that compounds, quarter over quarter, in a way that gets harder to catch up to the longer it runs. The ones sitting this out aren't standing still. They're starting further behind every single quarter that passes.
