E-commerce GEO: the guide

What GEO is, in one sentence

GEO (Generative Engine Optimization) is the work of getting a brand or a product to appear, with accurate information, in the answers of AI assistants such as ChatGPT, Claude, Gemini or Perplexity.

SEO aims for a position in a list of links. GEO aims for a place in a written answer: the assistant reads several sources, then recommends two or three products itself. There is no page 2. A product is in the answer, or it is not.

The two disciplines share most of their foundations: a site robots can read, fast pages, clear content. GEO adds two questions SEO never asks: is what the assistant says about you true? and out of a hundred answers, how many think of you?

Why e-commerce is a case apart

Most GEO tools and guides reason at brand level: is the brand mentioned, and by whom. For a store, that is the wrong unit.

• The shopper asks for a product, not a brand. "Which electric bike for commuting, under €2,500?" The answer is a list of models, each with a price and specifications.
• An answer contains checkable facts. A price, a range, a weight, an availability. Each one can be right, outdated or made up.
• The catalog changes every day. A price changed on Monday can stay wrong in the answers for weeks if the sources the assistant reads do not follow.
• The sale can be measured. A visit coming from an assistant leaves a trace in the store, and so does an order.

GEO that is useful to a store therefore works product by product, and checks the facts against the catalog.

How an assistant picks the products it recommends

An assistant relies on two things: what its model learned during training, and what it reads on the web when it answers, if web search is on. For a shopping question the second matters a lot: last year's prices and stock are useless.

What it reads comes from four families of sources:

1. Your product pages, if its robots are allowed to read them and the content is in the HTML.
2. Product feeds that some assistants accept directly from merchants.
3. Third-party sites: comparisons, reviews, specialist media, forums, marketplaces.
4. Structured data (schema.org Product, Offer) that states price, currency and availability without ambiguity.

The same question asked twice does not always get the same answer. This is the point most often forgotten: a screenshot of one answer proves nothing, either way.

The four workstreams

1. Be readable

An assistant does not recommend what it cannot read.

• robots.txt: check that the assistants' search robots are not blocked. Each vendor runs several, with different roles: at OpenAI, OAI-SearchBot serves search and GPTBot serves training. Blocking the second does not keep you out of the answers; blocking the first does.
• Server rendering: the title, the price and the description must be in the page's HTML. Most assistant robots do not run JavaScript.
• Structured data: a complete Product block, with Offer, price, currency and availability, identical to what the page shows.
• Product feed: an up-to-date feed, with identifiers (GTIN), variants and stock.

2. Be accurate

The most common error is not absence, it is contradiction. The page says €2,490, the markup €1,999, a comparison from last year €2,290. The assistant picks one of the three.

For every product that matters: one price, one availability and the same specifications everywhere, on the page, in the structured data, in the feed and on marketplaces. Write numeric specifications in plain text in the description ("120 km range", "21 kg") rather than in an image or a table loaded afterwards.

3. Be cited elsewhere

Assistants cite their sources. On your shopping queries, note the sites that keep coming back: they are the ones feeding the answer. If a comparison cited ten times does not mention you, that is where you need to be, with up-to-date information.

4. Measure

Without measurement there is no way to know whether a change had an effect. A serious measurement follows four rules:

• neutral queries that do not name your brand;
• several answers per query and per assistant, not one;
• a proportion (the shelf share) shown with its confidence interval;
• the same method from one week to the next.

Our methodology describes how skoup applies these rules, and what the measurement does not say.

Where to start

1. List ten to twenty questions a customer would ask before buying in your category: a use, a budget, a constraint.
2. Open your robots.txt and check the assistants' search robots.
3. View the source of a product page: are the price and the description in it?
4. Validate the structured data of your ten best sellers.
5. Compare price and stock between the page, the markup and the feed.
6. Ask the assistants your questions, several times each, and note the products and the sources cited.
7. Fix, then measure again with the same questions.

Steps 6 and 7 are the ones a tool automates. The glossary defines the terms used in this guide.

What GEO does not promise

Nobody controls an assistant's answer. Models change, so do their sources, and no setting guarantees a place. What a store does control: being readable, being accurate, being present where assistants read, and knowing, with numbers, whether its share is growing or shrinking.