Generative engine optimisation (GEO) is the new label for an old commercial problem: when a shopper asks an AI assistant which product to buy, does it name yours — and does it get the facts right?
The term is not marketing folklore. It was coined in a peer-reviewed paper, GEO: Generative Engine Optimization, presented at ACM SIGKDD 2024. The authors tested nine ways of rewriting a page and measured whether a generative engine used it in the synthesised answer. Citing sources, adding statistics and adding quotations lifted visibility by up to 40% in their benchmark. Keyword stuffing — the classic SEO reflex — did not.
What changed since then is the stake. Google now answers many searches on the results page, and the link underneath goes largely unclicked. Pew Research Center tracked 68,879 real Google searches: where an AI summary appeared, people clicked a normal search result on 8% of visits, against 15% where no summary appeared — roughly half the clicks, gone. And of the sources the summary itself cited, just 1% were clicked. You get named, not visited.
For product brands, GEO is a data problem, not a copywriting problem
General GEO advice is written for publishers: earn mentions, structure your prose, get quoted. For a brand that sells physical things, the questions an AI is asked are far more specific:
- Which model fits my machine? — the variant, the year, the part number.
- Is this still available, and at what price? — live offer data, not last season's page.
- What refill, filter or spare does it take? — accessory relationships.
- Is it still supported? — including the SKU you delisted three years ago.
- What is it made of, and can it be repaired? — the same fields ESPR asks for.
Those answers do not live in a blog post. They live in product data — and if a brand does not publish it in a form machines can read, the engine fills the gap from a retailer listing, a marketplace clone or a four-year-old forum thread. That is not a visibility problem. It is a brands-losing-control-of-their-own-spec-sheet problem.
There is no secret AI markup
Google's own guidance on its AI features is blunt: you do not need to create new machine-readable files or special schema.org structured data to appear in them. That line gets misread as "do nothing". The honest reading is different: there is no magic tag, so the ordinary plumbing has to be genuinely complete.
In practice, for a product catalogue, that plumbing means:
- A canonical page per product, including discontinued lines, that stays at a stable URL.
- Complete Product and Offer structured data — name, brand, images, description, SKU, GTIN, price, currency, availability — matching what the page visibly says.
- Standard identifiers. A GTIN, and ideally a GS1 Digital Link URL, lets every system agree that two records describe the same item.
- Documents machines can read. A manual locked inside a scanned PDF is invisible; the same content as structured text is quotable.
- A retrieval surface. Grounded answers come from a real corpus — RAG over the brand's own records, and increasingly an MCP endpoint an agent can query directly.
- Feeds where they are read. Google's Merchant Center specification, which Perplexity also accepts, and OpenAI's own commerce feed specification for ChatGPT.
Every item on that list is something a brand can do this quarter. None of it depends on guessing a ranking algorithm.
The commercial upside comes before the compliance one
Brands that publish structured product truth get three things that show up in revenue before they show up in a compliance file. AI answers name the right product instead of a competitor's near-equivalent. Repeat purchases — refills, spares, upgrades, consumables — become findable at the moment of need rather than buried in a catalogue. And the brand keeps the relationship with the person holding the product, instead of handing it to whoever ranked better.
The compliance dividend is real but secondary: the same structured record that answers an AI's question about materials and repairability is most of what a Digital Product Passport requires under ESPR.
Where SmartLinks fits
SmartLinks Hub gives every product — current or discontinued — a connected record: a canonical page, structured data, GS1 Digital Link identity, manuals converted into machine-readable text, and an AI-readable view that RAG and MCP clients can query and cite. It is generative engine optimisation applied at the level that matters for a product brand: the product itself, not the homepage.
GEO is a new word. Publishing the truth about your own products, in a form both people and machines can read, is not.
See how SmartLinks Hub builds AI-ready product records
Read the SmartLinks knowledge base
Talk to us about your catalogue
Sources: Pranjal Aggarwal et al., "GEO: Generative Engine Optimization", KDD '24, August 2024. Pew Research Center, "Google users are less likely to click on links when an AI summary appears in the results", 22 July 2025. Google Search Central, "Top ways to ensure your content performs well in Google's AI experiences". OpenAI, Product Feed Specification.