Optimising for ChatGPT means two separate jobs that people constantly conflate: being retrievable, so the search step can find your page, and being quotable, so the model picks your sentence over someone else’s. Most advice covers the first and skips the second, which is why sites get crawled and still never cited.
Open ChatGPT and ask the five questions your customers actually ask. Write down what it says and who it cites. That is your baseline, and it takes fifteen minutes. Everything below is guesswork until you have it.
Part one: be retrievable
Let the right crawlers in
Check your robots.txt before anything else, because this is a single line that silently undoes all the other work. The trap is that one company runs several agents for different purposes: blocking GPTBot keeps you out of training, while blocking OAI-SearchBot and ChatGPT-User keeps you out of today’s answers. People who meant to do the first often did the second by accident. The crawler article lists which is which.
Render content server-side
Retrieval fetches your HTML. It does not reliably run your JavaScript. If your article body only appears after a script executes, the fetched page may be an empty shell. Check the hard way:
curl -s https://example.com/your-post/ | grep -c "a distinctive sentence from your article"Zero means the crawler cannot see your content either. Standard WordPress themes are fine here; heavy page builders and JS-rendered blocks are where this bites.
Be fast enough not to be dropped
Retrieval fetches under a timeout. A page that takes eight seconds may simply not be there when the answer is assembled. This is ordinary performance work — caching, image sizes, fewer plugins — and it pays off in more than one place.
Part two: be quotable
This is the half that gets neglected, and it is mostly a writing problem rather than a technical one.
Answer in the first two sentences
Retrieval reads a chunk of your page, not all of it. If the answer to your own headline arrives in paragraph six after a warm-up about how the landscape is evolving, the chunk that gets read may not contain the answer at all.
Put the answer first, then explain. It is better for human readers too — nobody has ever complained that a page answered them too quickly.
Make sections self-contained
A chunk gets extracted without its neighbours. A section beginning “As mentioned above, this means…” is useless out of context. Each H2 section should make sense if it were the only thing anyone read, which sometimes means repeating a noun where you would otherwise have used “it”.
Be specific enough to be worth quoting
“Improves performance significantly” is unquotable. “Cut time-to-first-byte from 800ms to 180ms” is quotable, because it contains something a model can attribute to you rather than paraphrase from anywhere.
Numbers, versions, named constraints, dated facts. Every specific detail is a hook. Every vague claim is interchangeable with a competitor’s vague claim.
State when it was written and keep it true
For anything time-sensitive, a visible date and a dateModified in your schema help a model choose between sources. The corollary matters more: a stale page that confidently states last year’s price is worse than no page, because it will be read and quoted exactly as written.
Put a FAQ on pages that warrant one
Question-and-answer pairs match how people actually prompt, and a direct question with a direct answer under it is close to ideal retrieval shape. Two rules: every question and answer must be visible on the page, and questions must be ones people ask rather than ones you wish they asked.
What to do in WordPress specifically
- Use headings as structure, not styling. H2 and H3 are chunk boundaries now. Skipping levels for visual effect fragments your page badly.
- One canonical page per topic. Three near-duplicate posts targeting keyword variants split your evidence and leave nothing obviously canonical.
- Check your excerpts. An auto-generated excerpt that truncates mid-sentence is sometimes what gets read. Write them for the important pages.
- Audit your archive before adding to it. Fixing a stale page that contradicts your current pricing is worth more than a new article.
Measuring it
There is no Search Console for this. What you have:
- Ask, on a schedule. The same questions monthly, answers recorded. Crude, and the only direct measure that exists.
- Referral traffic. Visits with
chatgpt.comorperplexity.aias referrer. Real but a large undercount, since most people read the answer without clicking. - Crawler hits in your access log. Proves you are reachable. Proves nothing about being cited.
Anyone selling you a definitive AI visibility score is selling an estimate. Treat it as one.
Common questions
How long until changes show up in ChatGPT?
For live retrieval, as fast as your page is re-crawled — days, sometimes less. For what the model knows without looking anything up, that depends on training runs you have no visibility into and cannot schedule around. Optimise for the retrieval path, since it is the one that responds to your work.
Do I need llms.txt for this?
No. No vendor has committed to reading it and there is no evidence it is weighted today. It is cheap and harmless, so publish one if you like, but do it after the items above rather than instead of them. Details here.
Is this different from normal SEO?
It overlaps heavily. Crawlability, speed, structure and accuracy serve both. What is new is writing for extraction — self-contained sections, answers stated up front, specifics worth quoting — and the loss of ranking data to measure against.
Should I add an FAQ to every page?
No. Add one where people genuinely have recurring questions. A padded FAQ on a page nobody asks about is filler, and filler makes the rest of the page harder to extract from.
Does ChatGPT favour big brands?
There is a corroboration bias rather than a brand preference: things mentioned consistently across many sources are easier to state confidently. That is winnable on a narrow topic where you genuinely are the best source, and not winnable by writing broad articles about subjects a hundred larger sites already cover.
Where to go next
How does ChatGPT choose sources? explains the selection step. The visibility checklist is the broader version of this page.
Did this answer it? Tell us if not and we will fix the article.