Optimizing commerce for ChatGPT and AI discovery

When La Vanguardia ran the story in November, several operators forwarded it to me the same morning, asking whether the premise was real. The paper had reported on a Barcelona-based startup building software to help Shopify storefronts win sales not on Google, but inside ChatGPT. The framing was unusual enough that it stopped people mid-scroll.

I had been making the same argument to clients at Shugert Marketing in private for the previous eighteen months. The story, in my view, only confirmed something already underway: the storefronts being optimized for conversational surfaces — for ChatGPT, for Perplexity, for Gemini, for whatever conversational layer wins next — are compounding an advantage the ones ignoring it will not be able to catch up to.

It is already happening. A meaningful share of high-intent product research has migrated out of Google and into ChatGPT in the last year. Buyers no longer type "best running shoes for flat feet." They describe their problem in a paragraph and ask for a recommendation. The model responds with brands, comparisons, and increasingly direct purchase paths. If your storefront is not legible to the model when it goes looking, you do not exist in that conversation.

The instinct most operators have when they hear this is to ask which keywords to target. That is the wrong question. It carries the entire mental model of search engine optimization into a surface that does not work like search. There are no keywords. There is no results page. There is a model with a context window, a set of retrieval tools, and a goal: produce an answer that satisfies the user. The operator's job is to make the storefront the easiest, most credible, most structured source of truth for the questions the model is being asked about the category.

That breaks down into a handful of concrete things, none of them mysterious.

The first is structured data, taken seriously. Schema.org markup for products, organizations, articles, FAQs, and reviews is no longer a checkbox for rich snippets in Google. It is the substrate models use to extract structured facts about what a brand sells, who it is, and what other people say about it. A Shopify storefront that ships clean Product, Offer, AggregateRating, and Organization schema is dramatically more legible to a model than one that ships a generic theme and hopes for the best. The fix is cheap. Almost nobody does it well.

The second is content that answers questions, not content that ranks for queries. The old playbook found a high-volume keyword, wrote a 2,000-word article designed to rank for it, and captured the click. The new playbook identifies the actual questions buyers in a category ask, answers them with specific, verifiable, well-structured content, and lets the model cite the page. Models do not care about word count. They care whether the page contains a clear, accurate, attributable answer. Long thin content padded to hit a length target actively hurts the brand here, because models down-rank pages that read as filler.

The third is canonical, crawlable infrastructure. The agents now indexing the web for AI systems — GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, Google-Extended, CCBot, among others — need to reach a site, render it, and extract content the same way a search crawler does. That means server-rendered HTML, not JavaScript that hydrates the actual content client-side. It means a real robots.txt that explicitly permits the agents the brand wants indexing it. It means a sitemap. It means canonical URLs and clean Open Graph metadata on every page. The Shopify storefronts that fail this test most often, in my experience, are headless builds where the team optimized for developer experience and forgot that a model crawling the page sees an empty shell.

The fourth is reviews, comparisons, and third-party signal. Models do not just read a brand's site. They read everything else that talks about it. The brands that win conversational discovery are the ones that have invested seriously in earning credible third-party content — real reviews on real platforms, comparison articles on category-leading publications, mentions in roundups by sources the model already trusts. The work is slow. It is also the most defensible, because it cannot be faked at scale.

The fifth, and the one most operators are not yet thinking about, is what happens after the model sends the buyer. A visitor arriving from a ChatGPT recommendation is not in a generic search-driven mindset. They came pre-qualified, pre-educated, often pre-decided. The page they land on should reflect that. It should not re-pitch the brand. It should accelerate the decision: clear pricing, clear differentiation, clear path to checkout, clear answers to the residual questions the model could not resolve for them. This is where conversion rate optimization for the AI era actually lives, and almost nobody is designing for it yet.

What this adds up to is that the work of optimizing a Shopify storefront for AI discovery is not exotic. It is the work of building a clean, fast, structured, well-indexed, well-reviewed commerce site — done with intention, instead of as an afterthought. The discipline is the same discipline I have argued operators should bring to the platform underneath them: a posture in which Shopify is treated as infrastructure, not a website builder. Conversational discovery is the surface that finally rewards operators for taking that posture seriously.

The operators who do this in the next twelve months will start showing up in conversational answers their competitors will never appear in. The compounding starts there.

That is the bet my team is making across every market we operate in, and the reason La Vanguardia framed the story the way they did. The next chapter of commerce will be conversational. The storefronts ready for it will not be the ones that shouted loudest in 2015. They will be the ones quietly built right.