Skip to content

How to write content for ChatGPT and Perplexity

A language model does not read your page end to end; it takes fragments. Writing so each fragment stands alone is its own skill.

An answer engine does not read your page the way a person does. It splits the content into chunks, selects the ones most relevant to the query, and assembles an answer. That changes how you should write.

Chunkability

The core rule: every paragraph must make sense without the one before it. Three habits break this:

  • Back-references. "As described above", "this method" — the chunk becomes meaningless in isolation.
  • Pronoun chains. "It also affects that" says nothing without context.
  • Long build-ups. Prose that delivers the answer three paragraphs in gets filtered out before selection.

A structure that works

  1. Make the heading a question — phrased the way a user would ask it.
  2. Make the first sentence the answer — 20–40 words, complete and standalone.
  3. Then expand. Reasons, exceptions, examples.
  4. Close the section. Do not bridge to the next one; each section is its own door.

What gets cited

Answer engines carry these kinds of sentences noticeably more often:

  • Sentences with numbers. "Average load time was 2.4 seconds" travels far better than "pages should be fast".
  • Definition sentences. The "X is a Z that does Y" pattern.
  • Comparisons. "The core difference between A and B is…"
  • Step lists. Numbered, one action per step.

Why original data decides it

When a model sees the same fact on fifty sites, it must choose which to cite and usually picks the most authoritative. When a fact exists in only one place, there is no choice to make.

That is why the most valuable content is data nobody else has: ratios from your own customer base, results from your own tests, measurements specific to your field.

Putting your brand in the sentence

If the quoted sentence does not contain your brand, you gain little recognition even with a link. When describing your own method, make the brand the subject: "Across 40 projects, Loopcode measured an average of…". That way the brand travels with the quote.

What not to do

  • Stuff keywords — models detect this better than humans do
  • Pad length by restating the same point in different words
  • Write introductions that delay the answer
  • Put critical information only inside an image

A test you can run

Copy a section you have written and read it with no context. Is it accurate, complete and understandable on its own? If not, rewrite it. That simple test covers most of answer engine optimisation.