One Question Becomes Twenty: How AI Search Splits Up What Your Customers Ask

When someone types "best ordering system for a small cafe" into an AI-powered search, the system does not run that one query. Behind the scenes it typically rewrites the question into a set of smaller, sharper ones — what do ordering systems cost, which work in this country, what do cafe owners complain about, how hard is setup — researches each separately, and assembles a single answer from the pieces.

The industry calls this pattern query fan-out, and it is documented in Google patents and a string of AI research papers. You do not need the theory — but the consequence is worth understanding, because it quietly breaks the old content playbook.

Why the old playbook stops working

Traditional SEO trained everyone to pick a keyword and build a page around it. That made sense when one query matched against one page. But when a question is split into twenty sub-questions, a page that repeats its target phrase while answering nothing specific gives the system almost nothing to use. Each sub-question goes looking for a real answer, and pages that contain real answers get drawn on.

The shift, in one sentence: you are no longer trying to rank for the question — you are trying to be useful to its fragments.

What this means for your content, practically

1. Cover the follow-up questions, not just the headline one. If you sell a service, the fragments are predictable: what it costs, how long it takes, what is included, how it compares to alternatives, who it suits, where you operate. Each deserves a genuine answer on your site.
2. One clear topic per page or section, with a heading that states it. A heading phrased as the question, followed by a direct answer in the first sentence or two, is the easiest format for these systems to lift and cite.
3. Answer first, elaborate second. Burying the answer at the end of eight paragraphs of preamble worked when humans skimmed; machines pull the direct statement.
4. Be specific where competitors are vague. "Setup takes about two weeks and costs from NZD 1,450" can be used in an answer. "Fast and affordable" cannot.

The part nobody can sell you

Because the fan-out is generated dynamically and shaped by the individual user, no tool can hand you the exact list of sub-questions to target. Anyone claiming otherwise is guessing at machine behaviour that changes weekly. The durable strategy is less glamorous: know the questions your customers genuinely ask — you hear them on the phone every week — and make sure each has a clear, honest, findable answer on your website.

That is also why this shift mildly favours real businesses over content farms. The fragments of a genuine buying question are things only someone who does the work can answer well. If you write those answers down, plainly, you are doing the majority of what "AI optimisation" actually consists of.