What AI actually cites: the first 30%, and a table
Two findings shape whether an engine lifts your answer or someone else's, where the answer sits on the page, and whether it's a table. Here's what the research shows, and what it means for how you write.
Steen Stones · Reviewed
An AI engine doesn't recommend a page. It lifts a self-contained answer it can quote and cite. So the question that decides whether you're the source isn't “is my content good?” It's “can an engine find a clean answer, quickly, and is it mine?” Two findings from the public research change how you should write for that.
1. The first 30% of the page does the work
Front-loading isn't a style preference; it's where the citations are. If your answer is buried after four paragraphs of preamble, the way most website content is written for a human skim, an engine is more likely to quote a competitor who put the answer up top. Lead with the plain answer, in a paragraph that stands on its own, then put the evidence beneath it.
2. Tables get lifted more than prose
Where information is genuinely comparable, fees, timelines, options, criteria, a clean table is far more liftable than the same facts in a paragraph. Nectiv's analysis found 33% of ChatGPT citations included a table, against 13% of Google results. The lesson isn't “add tables everywhere”; it's “when the content is comparable, structure it so a machine can lift it whole.”
What it means for you
- Lead every page with the answer, in the first paragraph, not after a warm-up.
- Use a real table wherever the information is comparable, never as decoration.
- Write each answer to stand alone: an engine should be able to quote it without the surrounding page.
- Keep it compliant answer by answer, the sentence that gets lifted has to clear the rules on its own.
The unit of work isn't the article. It's the extractable answer.