Writing to Be Quoted
Whether your text lands in an AI answer is decided by its construction, not its topic. Five properties that make a paragraph quotable.
Boaz Lichtenstein
The Search Console report shows it unintentionally: what gets quoted is not the best topic but the fitting construction. A model does not read an article, it lifts pieces out of it – and a piece that no longer holds on its own is worthless. Five properties make the difference, and none of them involves keywords.
1. The claim sits in the first sentence
Journalism teaches the punchline at the end. For quotability that is exactly wrong: start the paragraph with the result and justify it afterwards, and you deliver a piece that holds on its own. Build up and resolve, and you deliver two halves.
2. The paragraph stands without back-references
“As described above”, “this effect”, an “it” whose antecedent sits three paragraphs earlier: every back-reference makes a piece useless the moment it is lifted out. The test is simple – read one paragraph alone. If you have to guess what it is about, it is not quotable.
3. Numbers carry unit and date inside the sentence
“Around 40 percent” is worth nothing without a point in time and a reference base, and is rightly not picked up. “40 percent of orders in July 2026” is. The caption does not count: what belongs to the sentence belongs in the sentence.
4. Definitions stated, not paraphrased
Models grab definition sentences whole. A sentence of the form “X is …” gets picked up far more often than the same explanation spread across three sentences that elegantly circle the term. Elegant circling is a disadvantage here.
5. Form follows content, not habit
Enumerable things as a list, values as a table, relationships as prose. A table without values is a layout trick and gets ignored; a list whose items are sentences should have been prose. Headlines say what follows – no puns, because a model cannot tell from them what the section is about.
What does not help
More text does not help, more subheadings do not help, and keyword density stopped working in the results list long ago. What helps is the uncomfortable work: one claim per paragraph, evidenced, without back-references. The side effect is that such texts read better for humans too – which is no coincidence but the same reason.