Every AI answer is an editorial decision. When ChatGPT, Perplexity or Google's AI Overviews respond to a question, they choose a handful of sources to quote and pass over everyone else. Watch enough of those decisions and a pattern emerges: the content AI search cites shares a recognisable set of traits, and the content it ignores shares another. This guide lays out both, so you can write for AI search deliberately rather than hopefully.
How AI search picks its sources
Generative engines are cautious quoters. Composing an answer means putting their credibility behind a claim, so they favour sources that make trusting them easy: pages that answer the question directly, sites with a consistent identity and reputation, authors who verifiably exist, and facts that can be checked against other sources. Nothing in that list is a trick, and that is the point. AI search behaves like a strict editor, and the way past a strict editor has always been better material, not louder pitching.
You can see the caution in how the engines behave. They rarely cite a page for everything; they cite it for the specific passage that answered the specific question. That has a useful implication: a modest site with one genuinely excellent, well-evidenced page can be quoted ahead of a household name whose page waffles. The bar is editorial, not architectural, and small businesses clear it more often than they expect.
The content AI search cites
Across the engines, the cited pages tend to share five traits.
- Direct answers under real questions. A heading that matches how people actually ask, and a first sentence that resolves it before the detail begins. Engines lift passages, and a passage that stands on its own is liftable.
- Original content. First-hand data, real examples, documented experience, a genuine point of view. A page that adds something new to the web gives an engine a reason to cite you instead of the hundred pages saying the same thing.
- Specifics with sources. Named figures, dates, steps and trade-offs, attributed to where they came from. Verifiable beats vague every time.
- Clear structure. Proper heading hierarchy, short paragraphs, lists and tables where they belong. This is classic on-page SEO doing double duty as machine readability.
- A visible author. A named person with demonstrable expertise, consistent across the site. Anonymous pages ask an engine to trust nobody in particular.
The content AI search ignores
The ignored pile is just as consistent, and most of it fails for one underlying reason: it adds nothing. Rewritten summaries of other people's articles give an engine nothing it could not get from the original. Padded introductions that circle a question for four paragraphs bury the quotable sentence, if one exists at all. Keyword-stuffed pages read as exactly what they are. Anonymous content with no author, no evidence and no accountability asks to be trusted on nothing. And hedged, non-committal writing, the kind that lists "several factors to consider" without ever committing to an answer, gives a quoting engine no answer to quote. None of this is about tricking a model. It is about giving a cautious system fewer reasons to say no.
The uncomfortable test: if your page vanished tomorrow, would the web lose any information? If the honest answer is no, AI search has no reason to cite it, and the fix is substance, not formatting.
Writing content for AI search without ruining it for humans
The good news is that there is no conflict. GEO content, meaning content shaped to earn AI citations, is simply disciplined writing: answer first, evidence next, structure throughout, and a real person standing behind it. Humans reward exactly the same things with their attention and their enquiries. The web is full of advice about writing "for the algorithm" as though it were a separate audience. It is not. The engine is trying to predict what a careful human would trust.
A practical drill we use: take your most important page and read only the first sentence beneath each heading. If those sentences answer the questions the headings raise, the page is quotable. If they wind up to the answer, rewrite them to lead with it. Then add whatever original content you can genuinely claim, your own numbers, your own photographs, your own documented results, because that is the material engines cannot find anywhere else.
One more habit worth building: date your important pages and keep them current. Engines answering in the present tense prefer sources that look maintained, and a page refreshed with this year's details quietly outcompetes an identical one that has clearly been abandoned.
Where this fits in a wider strategy
Content is the biggest lever in earning AI citations, but it works alongside the rest: technical health, schema markup, entity consistency and authority built beyond your own site. Our guide on how to rank in AI Overviews covers that full playbook, our GEO explainer covers the strategy behind it, and add-ons like an llms.txt file belong at the very end of the queue. Strong GEO content comes first, because it is the one ingredient nothing else can substitute for. It also compounds: every page that earns a citation makes the next one easier to earn, because the engines keep finding your site a reliable source to quote.
Wondering if AI cites you yet? Run the free AI visibility check to see which engines quote your content today, and where your competitors are being chosen instead.
