Shopping is moving from "search and scroll" to "ask and buy." Instead of a page of blue links, a shopper tells ChatGPT "find me a sustainable running shoe under $150 for flat feet," and the assistant returns a short, reasoned shortlist — product cards, images, prices, and a sentence explaining why each made the cut. Perplexity does the same with an in-answer Buy flow. This is agentic commerce, and the brands that win are the ones an AI can justify recommending.
This guide shows how ChatGPT Shopping and Perplexity Buy actually work, then walks through the optimization that gets your products into the shortlist: giving the AI a reason to pick you, mining reviews so the model's "vibe" data matches reality, adding multimodal signals, and monitoring your Share of Card so retailers don't intercept the sale.
Key takeaways
- AI assistants don't list products, they justify a shortlist. Optimize for the "reason to recommend" — feature, benefit, and proof tied to a specific shopper need — not for keyword-stuffed titles. - ChatGPT reads product data through `OAI-SearchBot` (search results), distinct from `GPTBot` (model training). If you want to appear in shopping answers, both your structured data and your crawler access have to be right. - Reviews are training data. Models absorb the "vibe" from review text, so mine sentiment, address recurring complaints in your copy, and encourage use-case-specific reviews. - The metric that decides revenue is Share of Card: how often your listing wins the AI product card versus a retailer reselling you at a different price. - GEOly's Share of Card and catalog analysis show which SKUs get cards, where retailer quotes intercept your traffic, and which attributes are missing — so you optimize causes, not guesses.
How ChatGPT Shopping and Perplexity Buy work
ChatGPT turned its chat box into a visual shopping surface: product carousels, images, prices, and direct purchase links, assembled to answer a natural-language request. It doesn't just crawl the web the way a search engine indexes pages — it reads product information to build results. That indexing runs through a dedicated crawler, `OAI-SearchBot`, which is separate from `GPTBot` (the crawler that gathers training data). One feeds live shopping results; the other feeds the model's general knowledge.
Perplexity works similarly, pairing cited answers with a native Buy flow so a shopper moves from question to checkout inside the answer. In both cases the assistant acts as a personal shopper: it interprets constraints ("under $150," "flat feet," "sustainable"), filters candidates, and presents a reasoned pick. Your job is to be a candidate it can confidently justify.
Step 1: Optimize for the reason, not the keyword
An AI agent needs a reason to choose you. Old-school product copy optimized for search phrases — "Buy [product] — best price." That gives an assistant nothing to reason with. The pattern that wins reads like a recommendation: "[Product] is the best choice for [specific shopper] because [specific feature] solves [specific pain point]."
