We Asked Claude, ChatGPT, Gemini, and Perplexity How They Buy: the Interview That Explains Agentic Commerce

If you want to understand where ecommerce is heading, stop looking only at other stores and start listening to the new intermediaries. Claude, ChatGPT, Gemini, and Perplexity are no longer just answer engines. They are becoming filters, advisors, and, increasingly, future operators of buying decisions. We asked them what they expect from an online store, what makes them distrust it, what prevents them from completing a purchase, and how they see the next phase of agentic commerce.

Claude was the most exhaustive. Its response built a complete framework for what a trustworthy store looks like to an agent. It started with something most brands never consider: the operational identity of the store. Not just the brand name — the tax ID, a verifiable physical address, domain ownership, and coherence between what the site claims and what external sources confirm.

For Claude, trust is not subjective. It is structural. A store where product copy is vague, prices do not explicitly include tax, or return policies require reading three pages to understand one condition creates cognitive friction for the agent. And that friction, absent explicit user instruction, translates into abstention.

Claude also highlighted structured data — not as an SEO best practice but as direct-read infrastructure. Schema.org/Product with price, currency, availability, condition, SKU, and GTIN. Schema.org/Organization with legal name, country, and contact. BreadcrumbList. FAQPage where applicable. Not as semantic decoration, but as the only reliable way to transmit information without relying on the agent interpreting dynamic HTML.

ChatGPT organizes the conversation around quality criteria, a technical checklist, MCP, WebMCP, optimized checkout, and a maturity framework. That framing reflects a logic very characteristic of the OpenAI ecosystem: turning a confusing space into a scorable sequence, where each layer of the store can be rated for agent readiness.

Gemini evaluates pages from multiple simultaneous signals. Its analysis did not start with the product or the checkout — it started with coherence between what is shown and what is declared. For Gemini, a store with a high-quality product image but no alt text, a long description but no heading structure, or a visible price but no schema markup is not technically incomplete. It is sending contradictory signals.

Perplexity is perhaps the most relevant case right now because it is already in the transition between search engine and buying agent. Its perspective is that of an entity already indexing, comparing, and recommending products in real time. What Perplexity pointed out is that most stores are optimized to capture a click, not to answer a question. And that difference is fundamental when the intermediary is an agent evaluating answers, not pages.