AI-Powered Commerce

AI-powered commerce refers to the use of AI chatbots and large language model (LLM) platforms as channels for product discovery, advertising, and purchasing. By early 2026, approximately 2% of all ChatGPT conversations — roughly 50 million queries per day — involve shopping, according to a joint study by OpenAI's Economic Research team and Harvard economist David Deming.¹³ The category spans in-chat product recommendations, native advertising inside chatbot responses, and checkout systems that allow users to complete purchases without leaving the conversation.

How AI Shopping Works

Traditional web search returns a list of links ranked by relevance. The user clicks through, opens multiple tabs, compares specifications, and navigates back and forth. AI shopping replaces this with a conversational loop: the user states a need in natural language — "quietest cordless vacuum under $300 for a small apartment with a dog" — and the model asks clarifying questions about budget, floor type, and storage constraints before returning a curated set of 4–6 product recommendations with structured data, pricing, reviews, and tradeoff explanations.¹

OpenAI reports that its shopping-specific GPT-5 mini model achieves 64% accuracy in matching products to user requirements, compared to 37% for standard ChatGPT queries.¹ The system pulls from structured product data, third-party review sites, and community discussions on platforms like Reddit, rather than relying on keyword-matched advertising or retailer self-descriptions.¹ Users can refine results by marking items "More like this" or "Not interested."

The feature performs best in specification-heavy categories — electronics, kitchen appliances, beauty, and outdoor gear — where products have objective, comparable attributes. Fashion and subjective categories perform worse.¹

The Shopify Integration

In January 2026, Shopify informed merchants that products sold through ChatGPT's checkout system would carry an additional 4% transaction fee on top of standard Shopify charges.⁶¹⁵ The integration allows users to browse, select, and pay for products without leaving the chat window.

ChatGPT's product ranking considers price, availability, review scores, and merchant reputation.² Google's AI Mode and Gemini, along with Microsoft's Copilot, announced competing checkout integrations with no additional merchant fees at launch.⁶

By March 2026, OpenAI pulled back from its instant checkout ambitions. Only about a dozen Shopify merchants had integrated the feature, and the company shifted its strategy toward directing users to complete purchases inside retailer apps rather than handling payments natively.¹¹¹² The retreat left open questions about returns, refunds, and customer service responsibility that OpenAI had not resolved.¹²

Referral Traffic and the Amazon Vacuum

Similarweb data from August 2025 showed that one in five of Walmart's referral clicks came from ChatGPT, up 15% from July.³ ChatGPT drove more than 20% of referral traffic to Etsy, nearly 15% to Target, and 10% to eBay.³

These numbers come with context: referral clicks account for less than 5% of total site visits for retailers like Walmart, with direct traffic, paid channels, and search engines making up the rest.³

Amazon blocked all OpenAI crawlers — ChatGPT-User, OAI-SearchBot, and GPTBot — from accessing its site, making approximately 600 million Amazon product listings invisible to ChatGPT.¹³ Amazon's $56 billion annual advertising business depends on shoppers browsing Amazon.com directly; allowing an outside AI to intercept that traffic would undermine both ad revenue and marketplace control.³ Instead, Amazon invested in its own chatbot, Rufus, which has fielded over 500 million customer questions and began displaying video ads in its interface.³

With Amazon absent, Walmart's 420 million SKUs filled more of the surface area inside AI chat results.³

Cloudflare and the Crawler Wars

Amazon's decision to block AI crawlers was individual, but Cloudflare — which sits in front of roughly 20% of all websites — turned anti-bot infrastructure into an industry-wide default. In September 2024, Cloudflare added a one-click option for site owners to block AI crawlers. More than one million customers activated it.²⁰

In March 2025, Cloudflare launched AI Labyrinth, a tool that detects unauthorized AI crawlers and, instead of blocking them outright, redirects them into a maze of AI-generated pages filled with plausible but irrelevant scientific content.¹⁹ The pages are invisible to human visitors but look like real content to bots parsing HTML. Any visitor that follows four or more links deep into the labyrinth is flagged as a bot, and its fingerprint is fed into Cloudflare's machine learning models to improve future detection across all customers.¹⁹ Cloudflare described the approach as using "AI-generated content as a defensive weapon" — turning the scraping tool against the scrapers.¹⁹

Cloudflare processes more than 20% of all global internet request traffic.²² AI crawlers alone generate more than 50 billion requests per day to Cloudflare's network, approximately 1% of the company's total volume.¹⁹ Since July 2024, Cloudflare customers have collectively blocked 416 billion AI bot requests.²¹

In July 2025, Cloudflare went further: all new domains registered through the platform began blocking AI crawlers by default, making it the first major internet infrastructure company to take that step.²⁰ Alongside the default block, Cloudflare introduced a pay-per-crawl marketplace allowing site owners to sell access to AI companies rather than giving it away for free.²⁰

The scale of Cloudflare's position gives these decisions outsized effect on AI commerce. If a retailer's product pages sit behind Cloudflare and the default crawler block is active, AI shopping systems cannot access pricing, inventory, or review data from that site — effectively removing those products from AI-powered recommendations the same way Amazon's robots.txt changes did, but applied automatically across millions of domains rather than one.²⁰

The Conversion Paradox

Users who arrive at retail websites from LLM platforms convert at approximately 1.5 times the rate of other referral channels, according to Criteo data from a sample of 500 US retailers in February 2026.⁴ But the overall click-through rate from AI chat to external websites is far lower than from Google search results, because many users get enough information from the chatbot's answer to make a decision without clicking out.¹⁴

This mismatch creates a measurement problem. Traditional digital advertising is priced on cost-per-click (CPC), but if the click never happens, CPC pricing undervalues the AI channel's influence. OpenAI addressed this by pricing ChatGPT ads on a view-based CPM model at approximately $60 per thousand impressions — comparable to NFL broadcast inventory and well above Meta's typical $10–20 CPM range.⁵¹³ The minimum advertiser commitment was set at $200,000.⁴¹³

Early confirmed advertisers included Target, Ford, Adobe, Expedia, Qualcomm, and Best Buy. Holding companies WPP, Omnicom, and Dentsu signed on as launch partners. In the third week of February 2026, search intelligence firm Adthena found ads appearing in roughly 0.8% of ChatGPT responses across 500 tested prompts.⁴

The Consumer Experience

AI shopping appeals to users because the model retains context across a conversation. A shopper can specify constraints once — budget, room dimensions, past products that failed — and the AI applies those constraints to every subsequent recommendation without the user re-entering filters on each new product page.¹⁴

The trust question cuts the other way. Ads in ChatGPT's Free and Go tiers appear at the bottom of responses and are labeled as sponsored, but they do not appear near sensitive topics like health or politics.⁴ Plus, Pro, Business, and Enterprise subscribers see no ads.⁴ Whether users perceive a recommended product as an organic suggestion or a paid placement remains unclear, particularly when the same interface delivers both.

Anthropic made the trust argument explicit. In a series of Super Bowl LX commercials in February 2026, the company mocked ChatGPT's ad integration with the tagline "Ads are coming to AI. But not to Claude."⁸¹⁶ Anthropic's head of product stated that the personal nature of chatbot conversations would make ads feel "incongruous" and "in many cases, inappropriate."⁹ The campaign produced an 11% jump in Claude's daily active users.¹⁷

Platform Positions

Each company has taken a different stance on monetizing AI through commerce and advertising.

ChatGPT went furthest. OpenAI launched shopping features in November 2025,² began testing ads in February 2026,⁴ onboarded Criteo as its first ad tech partner in March 2026,⁴ and briefly attempted in-chat checkout via Shopify before retreating to a retailer-app model.¹¹¹²

Perplexity was one of the first AI companies to experiment with ads in 2024, when CEO Aravind Srinivas predicted advertising would be the company's core revenue engine.⁷ By February 2026, Perplexity reversed course and announced it would wind down its advertising program by year's end, citing concerns that ads would erode user trust.⁷ Perplexity had roughly 60 million monthly active users on web and mobile in January 2026 — less than 10% of ChatGPT's or Gemini's user base — making an ad-supported model less viable at that scale.⁷

Anthropic took an explicitly anti-advertising position, publishing a blog post titled "Claude is a space to think" and running the Super Bowl campaign against ChatGPT ads.⁸⁹ The company has not introduced commerce features.

Google has not placed ads inside Gemini's conversational interface as of March 2026. CEO Sundar Pichai told investors in February 2025 that the company has "very good ideas for native ad concepts" for Gemini but would "lead with the user experience."¹⁰ Google's AI Mode in Search, which functions similarly to Perplexity's original product, already displays ads through Google's existing ad stack.¹⁰

Agentic Commerce

The next phase involves AI agents that do not just recommend products but autonomously complete multi-step purchasing tasks — booking travel, comparing insurance quotes, reordering household supplies.¹⁸ Criteo introduced an Agentic Commerce Recommendation Service in February 2026, claiming a 60% improvement in recommendation accuracy over text-description-only approaches by using commerce transaction data across $1 trillion in annual sales.⁴

Walmart's US Chief Technology Officer Hari Vasudev told The Wall Street Journal that the company expects the industry to eventually adopt common standards allowing outside shopping agents to interact directly with retailers' backend systems.³ Whether those standards emerge from an open protocol or from individual platform deals remains unresolved.

The infrastructure question — whether AI commerce creates an open ecosystem or a set of walled gardens controlled by model providers — will determine how the economics divide between platforms, merchants, and advertisers.¹⁸