AI-referred traffic to U.S. retail sites grew 693% year over year during the 2025 holiday season, according to Adobe Analytics, which tracked more than a trillion visits across the period. That surge landed on top of a structural shift: since September 2025, ChatGPT users can complete a purchase without leaving the chat window, through OpenAI's Instant Checkout and the agentic commerce protocol behind it. Perplexity, Google, and a growing list of retailers are building competing versions of the same idea. The demand curve is real. What most brands haven't checked yet is whether the AI agents doing the buying can actually read their product pages well enough to recommend them in the first place.
What agentic commerce actually means right now
Agentic commerce means an AI agent, not a person clicking through a browser, initiates or finishes a purchase on a shopper's behalf. Three live systems define the category as of September 2026: OpenAI's Instant Checkout, Perplexity's Buy with Pro, and Google's AP2 payment protocol, each designed to close a sale inside the conversation itself.
OpenAI announced Instant Checkout and open-sourced the underlying Agentic Commerce Protocol on September 29, 2025. Etsy went live the same day, and OpenAI said more than a million Shopify merchants were coming soon, naming Glossier, SKIMS, Spanx, and Vuori as early partners. It's the same discipline this blog covers under generative engine optimization, extended from earning a citation to earning a sale.
Google answered five weeks later with AP2, the Agent Payments Protocol. TechCrunch described it in October 2025 as arriving with broader technical ambitions but noticeably less merchant momentum than OpenAI's version. Perplexity had already moved first: its Shop Like a Pro assistant launched in November 2024, and the one-click Buy with Pro checkout, originally limited to paying subscribers, opened to free users in November 2025, according to CNBC.
All three platforms now run a merchant program on top of the checkout layer. Perplexity's gives participating retailers free API access and a dashboard tracking search trends. OpenAI's routes orders through the merchant's own backend systems, so the retailer keeps its existing fulfillment and customer data. None of the three currently charge for placement, though that is the part most likely to change once usage scales.
How ChatGPT and Perplexity actually decide which product to show
Neither system sells placement. OpenAI states directly that product results are organic and unsponsored, ranked purely on relevance to the user, and only weighs availability, price, quality, and primary-seller status when multiple merchants list the same item. Turning on Instant Checkout does not move a product up the list.
That logic mirrors how ChatGPT already picks citations for ordinary questions, where relevance and source quality outrank paid placement. Perplexity runs a comparable process for its shopping cards: recommendations are unsponsored, and the assistant compares specs and prices in plain language rather than ranking by advertiser spend, close to how it weighs source authority when it selects citations for a research query.
The mechanism underneath both is real-time retrieval: the agent queries a live product feed at the moment someone asks, instead of relying on a cached index from weeks earlier. That's why Shopify integration, accurate stock counts, and current pricing carry more weight here than they do for a typical informational query, where a slightly stale page rarely changes the answer an agent gives.
The one page type that scores worst on AI readability
Adobe's AI Content Visibility Checker grades retail pages from 0 to 100% on how much of their content a large language model can actually extract, and product pages score the worst of any page type Adobe measured: an average of 66%, against 80% for FAQ pages and 82% for returns and exchange pages.
Homepages average 75%, category pages 74%, and customer-service pages 79%. Adobe's sample covers more than a trillion visits to U.S. retail sites, and the spread between the best-performing sites, averaging 82.5%, and the worst, averaging 54.2%, is wide enough that two competitors selling an identical product can have very different odds of being read correctly by the same shopping agent.
The page built to close the sale is, on average across the retail sector, the page an AI shopping agent finds hardest to read.
That gap sits inside familiar GEO territory. Structured facts, not persuasive copy, are what an agent extracts and repeats, and product pages are usually written for the opposite job: to persuade a person scrolling images, badges, and cross-sells rather than to state price, size, and stock in a form a model can lift cleanly.
Why adoption is still catching up to the infrastructure
The growth numbers look dramatic against a small base. Adobe reports AI-referred traffic to U.S. retail sites rose 693% year over year during the 2025 holiday season and 393% in the first quarter of 2026, but the company's own report cautions that the base of users remains modest.
TechCrunch's October 2025 analysis of the OpenAI-Google commerce race called consumer appetite the industry's biggest unanswered question, noting that shoppers have so far shown more interest in using AI for simple product search than in letting an agent finish the purchase end to end. There is a control question sitting underneath the growth curve, too: because ChatGPT's recommendations draw on what the platform already knows about the user, TechCrunch pointed out that this shifts negotiating leverage away from the retailer and toward whichever platform is running the agent.
Adoption is also uneven by category. Adobe's holiday data shows AI shopping tools got the most use in video games, toys, appliances, electronics, and personal care, categories with clear specs and easy model-to-model comparisons. A retailer selling something harder to reduce to a spec sheet, like apparel or furniture, may see slower agent-driven traffic even as the platforms themselves keep expanding.
What to fix first if you sell through AI agents
Structured product data is the highest-leverage fix, because it gives an agent facts to retrieve instead of marketing copy to interpret. Complete Product schema markup, including price, availability, GTIN or SKU, and aggregate review data, turns a page a model might paraphrase inaccurately into one it can quote with confidence.
Consistency matters almost as much as completeness. An agent pulling a live feed will cross-check price and stock against what it already knows from your reviews, your listings on marketplaces like Etsy or Amazon, and any cached version of the page it has seen before, so a number that is wrong in one place undermines trust in all of them. A structured audit of where those numbers disagree is usually a faster starting point than rewriting product copy from scratch.
Most analytics platforms still cannot tell you whether your product pages are actually surfacing inside ChatGPT or Perplexity's shopping flows. vizibleai.com tracks brand and product mentions across those engines specifically, which is the only reliable way to confirm a fix actually worked instead of guessing from traffic numbers alone.
Frequently Asked Questions
What is agentic commerce?
Agentic commerce is a purchase that an AI agent initiates or completes on a shopper's behalf, without the person navigating a traditional checkout flow themselves. The clearest examples live inside chat interfaces today: OpenAI's Instant Checkout lets ChatGPT users buy a product without leaving the conversation, and Perplexity's Buy with Pro does the same for its subscribers. Google's AP2 protocol extends the same idea to agent-to-agent payments.
Does enabling ChatGPT's Instant Checkout improve my product's ranking in ChatGPT?
No. OpenAI states that product results are organic and unsponsored, ranked purely on relevance to the user. When several merchants sell the same item, ChatGPT weighs availability, price, quality, and whether a merchant is the primary seller, but having Instant Checkout enabled does not move a listing up the list on its own.
How is Google's AP2 different from OpenAI's Agentic Commerce Protocol?
Both let an AI agent complete a payment without the shopper switching to a separate checkout page, but they launched from different starting points. OpenAI's protocol shipped with Etsy live and Shopify's merchant base coming on board. TechCrunch reported in October 2025 that Google's AP2 arrived with broader technical ambitions but noticeably less merchant adoption at launch.
Why do product pages score worse on AI readability than other page types?
Adobe's AI Content Visibility Checker found product pages average 66% machine-readability, the lowest of any retail page type it measured, versus 80% for FAQ pages. Product pages are usually written to persuade a person scanning images and layout, not to state facts an algorithm can extract cleanly, which is exactly the gap structured data is meant to close.
How can I check whether AI shopping agents can read my product pages correctly?
Start with a structured-data validator to confirm your Product schema parses correctly, then ask ChatGPT, Perplexity, and Gemini directly what they know about a handful of your bestselling products. A mismatch between what the agent says and what your page actually states is usually a schema or feed problem, not a content problem, and it is worth fixing before spending on new product copy.
See whether your product pages are actually reaching AI shopping agents
Most brands only discover an agentic commerce gap when a competitor's product shows up in a ChatGPT answer and theirs does not. Measuring it directly, across the same engines now handling checkout, catches the problem before it costs a sale.
Start your free 7-day trial at VizibleAI — no credit card required.