Online stores are no longer optimizing only for people and Google. They are also starting to optimize for assistants that need to find products, compare specifications and sometimes take the next steps on the user’s behalf. WooCommerce has now dragged that conversation back down to earth: an AI agent does not need another slogan about “premium quality”. It needs data it can read and compare without guessing.

The starting point is hard to ignore. According to Adobe, traffic from AI tools to US retail websites grew 393% year over year in the first quarter of 2026. In March, that traffic converted 42% better than non-AI traffic. A year earlier, it had converted 38% worse. Adobe based the analysis on direct transaction data covering more than one trillion visits to US retail websites, plus a survey of more than 5,000 people in the United States.

Those percentages need one fairly large asterisk. Explosive growth does not automatically mean AI has become the biggest sales channel. WooCommerce itself points out that the absolute numbers are still small. Growth from a small base can look spectacular, so 393% should not be translated into “AI has taken over ecommerce”.

WebsiteGarage / Services

Explore our services.

A third of product-page content is hard for machines to read

Adobe also looked at how machine-readable retail content actually is. In its AI Content Visibility Checker, the average product page in the group of US retailers it analysed scored 66%. Put simply, roughly one third of the content Adobe examined was not considered readily readable by models. Product pages performed worse than homepages and category pages.

That does not prove that a product scoring 66% suddenly disappears from ChatGPT, Gemini or AI Overviews. WooCommerce is quite clear that there is no hard data directly linking a score like this to lost recommendations or revenue. The technical problem is much simpler: if an important specification exists only inside an image, a strange JavaScript component or an inconsistent marketing description, a system trying to compare products automatically has fewer reliable facts to work with.

WooCommerce already provides some of the infrastructure

The good news for WooCommerce stores is that nobody needs to build a separate “store for robots” overnight. WooCommerce notes that core product fields such as title, description, image, SKU, price, currency and stock status are already mapped to Schema.org data in JSON-LD on product pages. That gives search engines and compatible systems a structured layer they can read without having to guess which number is the price and which one is the model number.

The messier part starts with richer catalogs. Size, material, compatibility, variant, capacity, technical specifications and industry-specific attributes often live in custom fields, tables or free-form descriptions. WooCommerce points out that richer fields and variant-level data may require extensions or custom markup.

AI Search will not clean up a messy catalog for you

The most practical part of WooCommerce’s new guidance is barely about language models at all. It is about an old ecommerce problem: one price on the website, another in the feed, a missing attribute on one variant, stock that updates late, and a product description that says something different from the specification table. For a human, that is annoying. For a system comparing offers automatically, it is simply low-confidence data.

A WooCommerce-sponsored IDC InfoBrief published in July uses the term “agent buyer” for a buyer represented by an agent. The idea is that such systems will make decisions using structured catalogs, current pricing, availability and consistent product identification. It is worth remembering who paid for the paper: this is directional market research produced by IDC for WooCommerce, not neutral proof that one specific piece of markup guarantees sales through one specific AI model.

What you can do without buying a magical “GEO” package

The first sensible audit is almost comically unfuturistic. Export the catalog and check which important attributes are empty or buried only in descriptions. Then compare prices, availability, identifiers and variants across the website, feeds and sales channels. Finally, inspect the structured product data and check whether the important information actually exists in the HTML in a form that can be read without clicking through half the interface.

Only after that does it make sense to experiment with additional formats designed specifically for agents. An llms.txt file, an extra “for AI” page or a hundred paragraphs generated around possible user questions will not repair a catalog where price, stock and product specifications all tell different stories.

WebsiteGarage comment

AI Search is already spawning a new flavour of SEO voodoo at impressive speed. The label changes, the invoice stays the same. The most interesting thing about WooCommerce’s latest material is actually its return to the basics: good product data, clear structure, consistent information and up-to-date stock. Those things mattered before ChatGPT, and they will still matter after the next marketing acronym stops being fashionable.

That does not mean AI Search can be ignored. The growth measured by Adobe is real, and user behaviour is genuinely changing. It only means the first question should not be: “How do I become more visible in AI answers?” A better question is: “Does my store provide data good enough for a person, a search engine and an agent to understand the same product without getting three different versions of the truth?”