Search used to return ten links, and you competed for a click. An assistant returns one answer and a few products. If you are not in it, the shopper never sees you.

Nobody can promise placement in that answer, and anyone who does is selling something. There is a growing body of evidence on what improves your odds, and most of it is less mysterious than the acronym suggests.

What the research says

A team behind the GEO benchmark tested nine ways of rewriting web content to see which raised its visibility in generative engine responses. The headline result is that well-chosen changes can boost visibility by up to 40%, and that the effect varies by domain. (arXiv)

The detail is more useful than the headline. In their results, the methods that helped most were adding quotations from relevant sources, adding statistics, and citing sources. Making text more fluent helped too. Stuffing keywords, the old search habit, did no good and scored slightly below the baseline. Sites that ranked lower in conventional results gained the most, which suggests the method rewards substance and levels the field.

Treat these as a reason to test, not as a promise. The benchmark isn't your category, and the assistants change.

What assistants can use

An assistant answering "waterproof hiking jacket under $300" looks for facts it can check: a waterproof rating, a material, a price, a delivery promise, a return policy. A page that says "great for the outdoors" gives it nothing to quote. A page that says "20,000 mm waterproof, taped seams, ships tomorrow" does.

That points to a few habits.

• Put specific facts in plain text. Ratings, dimensions, materials, certifications and prices, written as numbers with units.
• Cite your sources. If you claim a result or a standard, say where it comes from. The research suggests assistants reward pages that do.
• Quote real people. Customer reviews, expert comments and test results give an assistant something concrete to attribute.
• Mark up what a machine should read. Structured product data, ratings and reviews, and policies in consistent fields.
• Keep your facts consistent. The same price, stock and policy on your site, in your feeds and on marketplaces. A mismatch makes an assistant less likely to trust any version.

None of this is a trick. It is the same discipline as writing good product pages, taken seriously.

Decide who is welcome

Your robots rules say which AI crawlers and agents may read your site. Many stores have never chosen, and some are blocking assistants by accident through rules written for scrapers. Someone should decide on purpose, with a view on what you want assistants to be able to quote and which uses you want to allow.

The major assistants publish the names of their crawlers, and most respect your robots file. Read the rules each one documents, set your policy, and revisit it as new agents appear.

Third-party voices count

Assistants lean on what other sources say about you. Reviews, comparison sites, forums, industry write-ups and retailer pages all feed the answer. You can't edit those, but you can make sure they exist and are accurate. Ask satisfied customers for reviews, give reviewers the details they need, and correct wrong information where it appears.

Measure it monthly

Write down the ten questions your customers ask before they buy. Put them to each major assistant once a month and record who gets named, what is said about you, and which sources the assistant cites. It takes an hour.

After a few months you have a trend line. You can see whether a catalog fix moved anything, whether a competitor is gaining, and which claims about you are wrong. Keep the log. It is the only way to tell whether your effort is working, because there is no ranking report to look at.

Where to start

Pick one product category. Rewrite twenty product pages with specific facts, sources and a real customer quote each. Check that your structured data matches the page. Run your ten questions before and after, and compare.

For the wider picture, start with why your next customer may be an AI agent, and use the agent-readiness checklist to score the rest of the store.