Discovery

Jeremy Howard proposed LLMs.txt in September 2024 as a way for websites to communicate with AI crawlers. By October 2025, 844,000 sites had implemented it. The major AI crawlers are still largely ignoring it. So where does that leave us?

Google AI Overviews now appear on 48% of searches, with an average zero-click rate of 83% when they do. Between 58% and 68% of all Google searches now end without a click. The organic traffic model that UK retailers have relied on for fifteen years is structurally broken. Here's how to think about that.

AI-generated product content is now standard for large catalogues. The tooling works. The quality variance is the problem nobody planned for, and on a 50,000-SKU catalogue, even a 1% error rate is 500 wrong product descriptions.

Adobe's Q1 2026 data shows AI-referred traffic to US retail sites grew 393% year-on-year. A separate Adobe finding: product pages across the retail sector score an average 66% on machine readability. The traffic is arriving. The infrastructure to capture it is not.

Between AI Mode, agentic checkout, virtual try-on expansion, and Project Mariner, Google confirmed more commerce-relevant technology at I/O 2025 than at any developer conference in recent memory. Most coverage led with the chatbot.

Schema.org Product markup and JSON-LD have been around for over a decade. In 2025, they became the infrastructure that determines whether AI systems can accurately understand and recommend your products. The SEO conversation became a GEO conversation. The stakes got real.

In September 2024, Answer.AI's Jeremy Howard proposed llms.txt — a standard for structuring web content so AI systems can read it more effectively. The technical case is interesting. Whether it matters for ecommerce is a more honest question.

The term GEO (Generative Engine Optimisation) is starting to circulate. Ignore the jargon, but pay attention to the underlying shift: AI-powered search is changing what good product content looks like.