Generative engine optimisation for ecommerce: how to get recommended by AI shopping agents

This is a deep dive from my agentic commerce architecture blueprint, where being chosen is one of five layers.

The term comes from a 2023 research paper by Aggarwal and colleagues at Princeton and other institutions, which defined generative engine optimisation (GEO) as improving how often content appears in answers produced by generative engines, and reported visibility gains of up to 40% on its benchmark. The industry has since added near-synonyms: answer engine optimisation, AI search optimisation, LLM SEO.

Most of that work is about getting articles cited. Ecommerce is different in one important way. When someone asks an assistant for “the quietest 9kg washing machine under £600 that can be delivered this week”, the answer is not built mainly from blog posts. It is built from structured product data: attributes, prices, availability, delivery promises, reviews and policies, pulled from feeds, merchant platforms and page markup, then checked against what the assistant can verify.

So for retailers, GEO is less about writing and more about whether your product data is complete, consistent and current everywhere an assistant looks.

A lot of GEO advice is speculation. Before following any of it, it is worth reading what the two largest platforms have published.

Google’s guidance for site owners is direct: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.” A page has to be indexed and eligible to show with a snippet. Google recommends allowing crawling, keeping important content in text, maintaining good page experience, and making sure structured data matches the visible text. It also says there is no special schema to add, and that Merchant Center and Business Profile information should be kept current because commerce and local results draw on them.

For retailers, Google announced in May 2026 a Merchant Center tool that compares a brand’s share of voice in AI surfaces with similar brands, initially in Australia, Canada, India, New Zealand and the US, and the ability to update product descriptions with conversational attributes.

OpenAI says product results in ChatGPT “are selected independently” and “are not ads”. Products are chosen for relevance to the shopper’s request and context; merchants are ranked by factors such as availability, price, quality and whether they are the maker or primary seller. Data comes from structured metadata supplied by first-party and third-party providers and other third-party content. Shopify merchants are included through Shopify Catalog; other merchants can apply to provide a direct product feed. ChatGPT may rewrite titles and summarise reviews, and prices can lag.

Read together, these statements point in one direction. There is no secret format. The retailers that get recommended are the ones whose information is complete, accurate and easy to verify.

This is the order I would work in, from highest to lowest impact for a typical UK retailer.

Point 8 deserves a warning. Allowing a crawler is a commercial decision as well as a technical one: content used for search answers and content used to train models are often governed by different crawlers and settings. Decide per crawler, and see Web Bot Auth and the Trusted Agent Protocol for how to tell real shopping agents from scrapers.

You cannot manage what you cannot see, and AI answers are harder to see than search rankings. The method I use is a fixed prompt panel.

Share of answer is the proportion of prompts in a category where the assistant recommends you, among prompts where you sell a matching product. I would track three numbers alongside it: price accuracy (how often the price shown matches your live price), position (first, second, third), and competitor presence (who appears when you do not).

The most useful output is not the score but the diagnosis. When you lose a prompt, look at the winning product: which attribute did it have that yours did not show, was its delivery promise sooner, was your price stale in the answer? Then fix the cause in your product information system or offer data, not just in a feed or a page. That loop is described in the main agentic commerce guide.

Once you are being recommended, the next questions are how agents check out and pay. Those are covered in ACP vs UCP and agentic payments explained.