Writing for answer engines without writing for robots

What actually changed

For twenty years the game was ranking a page so a person would click it. Increasingly the page is read by a model that will answer the question itself and cite whoever stated it most clearly. The click may never happen. The citation is the prize.

That sounds like a catastrophe for publishers and it partly is. But the mechanics of being quotable are not mysterious, and they overlap almost entirely with the mechanics of being useful.

How to be quotable

State the claim in one sentence. A model extracts sentences, not vibes. If your answer takes three paragraphs to assemble, nobody will lift it.

Put the answer near the question. Section headed with the question, answer in the first sentence under it, elaboration after. This is also how impatient humans read.

Attach the evidence. Figures, sources and dates make a passage safer to quote, because they can be checked. Unsupported assertions are exactly what a cautious model skips.

Say when it does not apply. Caveats make a claim more citable, not less, because they narrow it into something defensible.

What not to do about it

Do not write FAQ blocks that repeat the article in worse prose. Do not stuff the ranked keywords into sentences that a person would not say out loud. Do not add structured data describing content that is not on the page, that is the fastest route to being distrusted by both kinds of reader.

The keyword set still matters, but as vocabulary rather than as a quota. Ours come out of the language people actually used in threads, which means using them naturally is the default rather than an effort.

The overlap with just writing well

Plain claims, answers where the questions are, evidence attached, honest limits. Every one of those is what a decent editor asked for before answer engines existed. The difference now is that the cost of ignoring them is visible: you can watch a competitor get cited for saying the thing you buried in paragraph nine.