The Yorkshire Potter ChatGPT Started Recommending

Mara Whitfield makes mugs and bowls in a converted garage in Otley, West Yorkshire. Her business, Whitfield & Clay, sells mostly through Etsy, the odd craft fair, and word of mouth. For years, her marketing routine looked like most small makers': decent product photos, a few keywords in her listing titles, and hoping the Etsy search algorithm liked her that week.

Then, in fewer words than it takes to describe a glaze recipe, the ground shifted under her. On 29 September 2025, OpenAI announced Instant Checkout, a feature that lets people buy products directly inside a ChatGPT conversation, starting with merchants on Etsy and Shopify. The mechanism behind it is an open standard called the Agentic Commerce Protocol, built with Stripe, which lets a retailer's product catalogue be read, recommended, and purchased by the chatbot itself rather than merely linked to.

For a maker like Mara, that is not a small technical footnote. It means the first "page" a customer sees might not be a search results page at all. It might be a single paragraph an AI model generates in response to someone typing "a handmade stoneware mug for my dad who likes strong coffee," with her shop named, described, and priced inside the answer, or quietly left out of it.

How generative engines choose who to name

This is the core shift behind generative engine optimization, or GEO: the discipline of making a business legible to AI systems, not just to search engines. It is worth being precise about what changed and when, because the mechanics matter more than the buzzword.

Google moved first at real scale. In May 2024, Liz Reid, Google's Vice President of Search, wrote on the company's blog that AI Overviews, AI-generated summaries that sit above traditional blue links, would roll out to everyone in the United States. Then, in October 2024, OpenAI announced that ChatGPT could search the live web and show its sources, turning a chatbot that used to work from a static training cutoff into something closer to an answer engine with citations. Instant Checkout, a year later, was the next logical step: once an AI can find and cite a business, the obvious next move is letting it complete the purchase too.

What all three developments share is a change in how a business earns visibility. A generative engine is not counting backlinks or matching exact keyword phrases the way traditional search once did. It is retrieving information from crawlable pages, structured product data, review platforms, and business listings, then deciding, in the moment, which two or three businesses out of thousands are worth naming in its answer. If your product feed is messy, your business details are inconsistent across the web, or your reviews live only on one obscure platform, the model has less to work with and less reason to trust what it finds.

What actually changed for Mara

Mara did not do anything dramatic. She cleaned up her Etsy listing titles so they matched how people actually ask questions ("mug for someone who likes strong coffee" rather than just "stoneware mug 350ml"). She made sure her studio's name, location, and materials description read the same way on her Etsy shop, her Instagram bio, and the one local press mention she had from a Yorkshire craft fair years ago. Small, unglamorous consistency work, the kind that used to matter for SEO and now matters even more, because a language model is quietly cross-checking those same details when it decides whether to trust and name a business.

None of this guarantees a mention. Generative engines are still opaque about exactly how they weigh sources, and no legitimate tool can promise a guaranteed spot inside an AI-generated answer. But the businesses showing up in these answers tend to share the same traits: clear, structured information about what they sell and where, consistent details across the web, and real reviews that a model can point to as evidence.

That is the practical heart of GEO. It is less about chasing a new algorithm and more about making sure that when ChatGPT, Perplexity, Gemini, or Google's AI Overviews go looking for evidence of who does something well, your business is easy to find, easy to verify, and easy to trust.

If you are curious how AI tools currently describe your own business, and where the gaps are, that is exactly the kind of thing Ozvor is built to show you.