When Customers Ask AI Instead of Searching: What GEO Is and How to Prepare

A business owner wants to pick a knowledge-base tool for his company. Two years ago he would have typed "enterprise knowledge base recommendations" and opened the top ten links one by one. Today he is increasingly likely to just ask an AI: "Trading company, 50 people, we need our product documents and quotations under control — what should we use?" The AI returns one answer synthesised from multiple sources — and at no point does he see a search results page.

For companies that win customers through content, this raises a new question: when customers form their first impressions inside an AI answer, what earns your content a citation in it?

Search is turning into direct answers

The shift has a clear timeline. Google introduced AI Overviews at its I/O conference in May 2024, generating answer summaries above traditional results; by May 2025 the company reported the feature had passed 1.5 billion monthly users across more than 200 countries and territories. That same month, on May 20, 2025, Google fully launched AI Mode in the United States — a standalone conversational search tab whose queries, according to Google, run two to three times longer than traditional searches: people are asking full questions instead of typing keywords.

China is moving on a similar clock. At the Baidu World Conference on November 13, 2025, Baidu disclosed that the great majority of its search results are now AI-generated, with rich-media coverage reaching 70% of first results. Both threads point the same way: the answer layer is intercepting attention that used to flow to lists of web pages. Users take the answer and leave; clicks through to individual sites are shrinking — unless your content becomes part of the answer.

GEO: optimising content to be cited by AI

The problem already has a name and a method. In November 2023, researchers from Princeton University, Georgia Tech, the Allen Institute for AI and IIT Delhi released the paper GEO: Generative Engine Optimization, formally published at KDD 2024, the international data mining conference, in August 2024. GEO — generative engine optimization — takes its name from that work: a set of practices that make content easier for generative engines to retrieve, understand and cite in their answers.

The study ran systematic tests across nine domains and ten thousand queries, and its finding is worth remembering: adding statistics, citing sources and including authoritative quotations raised a page's visibility in AI answers by up to about 40%. In other words, the same information, organised differently, gets cited at meaningfully different rates — and that difference is exactly the space where companies can act.

GEO stacks on top of SEO — it does not replace it

One common misreading: GEO is not an "SEO is dead" story. Generative engines still build their answers on retrieval — fetch pages first, then compose the answer from what was fetched, the same retrieve-then-generate mechanism as RAG (retrieval-augmented generation). Your content must be crawlable, indexed and understood before it can ever be cited; crawlability, structured data and page speed — the SEO fundamentals — remain the entry ticket.

What changes is the unit of competition. SEO fights for a ranking slot on a results page; GEO fights for a citation inside the answer. A results page holds ten slots — an answer typically cites two or three sources. The competition concentrates, and the bar for content quality rises with it.

Five things companies can do now

Combining what the paper validated with plain content-operations sense, the worthwhile actions cluster into five kinds:

• Answer first: open every piece with two or three sentences that directly answer the question in the title, so both humans and machines catch the conclusion immediately;
• Clear structure: subheadings that each settle one matter — machines understand and excerpt by section, so clean structure is the precondition for being cited correctly;
• Data with sources: statements backed by figures and references get cited more readily than vague adjectives — one of the strongest effects in the paper;
• Attribution: name the author, the organisation, the publish and update dates — signals an engine can use to judge credibility;
• Freshness: generative engines favour content that is current and maintained; stale information gets squeezed out of answers over time.

None of the five is a gaming trick; each simply makes content more verifiable and easier to cite. The Key Takeaway box at the top of every article on this site is the same idea in practice — letting a person or a machine grasp within three seconds what question the piece answers.

The uncertainty, stated plainly

Two caveats belong on the record. First, AI citation mechanics are a black box: which content gets selected and how it is weighted is not fully disclosed, and the rules change far faster than traditional search ever did. Second, the paper's "about 40%" is a research result under specific test conditions — not a forecast for any particular company.

So treat services promising "guaranteed AI recommendations" or "first place in AI search" with deep suspicion — in a system this opaque and this fast-moving, no one can guarantee outcomes. GEO deserves investment, but approach it as raising your odds, not purchasing a certainty.

The answer changed; the homework did not

Back to the business owner. His first-round shortlist now happens inside an AI answer, and who gets cited there depends on whose content is direct, credible, sourced and current. That homework was already on the list in AI content marketing — GEO invents no new virtue; it simply gives companies that do content honestly one more doorway to being seen. Rather than wait for the mechanics to clarify, start writing every piece to be worth citing: however the rules shift, that effort keeps its value.