Ask ChatGPT or Perplexity a buying question and watch what happens. It doesn't hand back ten links. It writes one answer and names a small set of sources — usually 2 to 7. If you want to know how to get cited by ChatGPT, start with a simple truth: the engine is looking for pages it can quote without work. Most pages fail that test. They open with a story, bury the answer in paragraph six, and format everything as one long wall of prose.

This post covers the on-page half of the job — the concrete rewrites that make a page liftable. It assumes the AI crawlers can already read your site. If you're not sure they can, fix that first: why your brand isn't showing up in AI answers walks through the crawlability blockers. A perfectly structured page that no bot can read cites nobody.

One more framing note. Industry research puts the overlap between top Google links and AI-cited sources below 20%. That's the opening. Your competitor's rankings don't automatically carry over into AI answers — which means well-structured pages from smaller sites genuinely compete here.

Why do engines lift some pages and skip others?

An answer engine works in two steps. First it retrieves candidate pages. Then it reads them and writes one answer, crediting the sources it actually used. Retrieval gets you into the room. What gets you quoted is whether the engine can pull a clean, complete, credible answer from your page without stitching it together itself.

Think of it like a busy editor on deadline. Hand the editor a finished quote and they use it. Hand them a rambling interview transcript and they move on to the next source. Every lever below is a way of handing the engine a finished quote.

What is a direct-answer block?

A direct-answer block is a short, complete answer — 40 to 60 words — placed immediately under a question-style heading, before any context, story, or caveats. It answers the question in full, in plain language, in a shape an engine can lift verbatim. It's the single highest-leverage change on this list.

Why 40 to 60 words? That's a writing guideline, not a magic number. Shorter tends to be incomplete. Longer starts to ramble. A block that size forces you to state the answer, the key qualifier, and nothing else — which is exactly what an engine wants to quote and exactly what a skimming human wants to read.

Before and after: the same intro, rewritten answer-first

Here's a realistic example. A local bookkeeping firm has a page targeting "how much does a bookkeeper cost for a small business?"

The before version says nothing quotable for four sentences. The after version is a complete, specific answer an engine can lift — and a human gets value in ten seconds. Notice it commits to a structure (flat fee, priced by volume and complexity) instead of hiding behind "it depends."

Write headings the way people ask questions

Engines match questions to answers. A heading that mirrors the question — "How much does a bookkeeper cost?" instead of "Our Pricing Philosophy" — makes that match easy. Clever headings are a tax on your citation odds.

• Weak: "The Pricing Puzzle" → Strong: "How much does a bookkeeper cost for a small business?"
• Weak: "Making the Right Choice" → Strong: "Should you hire a bookkeeper or use software?"
• Weak: "Getting Started" → Strong: "What documents does a bookkeeper need from you?"

Use your buyers' actual words. The questions they type into ChatGPT are the questions they'd ask you on a call — pull your headings from there, not from a brainstorm.

Add a TL;DR and keep density high

A TL;DR at the top does two jobs. For humans, it earns trust in the first five seconds. For engines, it's a pre-summarized version of your whole page — a gift. Three to five bullets, each a complete claim, no teasers.

Then hold the rest of the page to the same standard: high information density. Every paragraph should carry a fact, a definition, a number you can stand behind, or a specific recommendation. Filler sentences — "in today's fast-paced world," "as we all know" — give an engine nothing to quote and give a reader a reason to leave. If a sentence could appear on any competitor's page, cut it or sharpen it.

Make the structure extractable

Engines parse structure more reliably than prose. The same information, reformatted, becomes dramatically easier to lift:

Then add the machine-readable layer: FAQ schema and HowTo schema in JSON-LD. Schema doesn't replace visible content — it labels it. It tells the engine "this is a question, this is its answer, these are ordered steps," so it can pull your answer cleanly instead of guessing at your page's shape. Keep the schema and the visible text identical; schema that says things the page doesn't is a trust problem, not a shortcut.

Make every section self-contained

Engines quote sections, not whole pages. If your answer in section five only makes sense after reading sections one through four — "as we saw above," "using the method from earlier" — the lifted quote is broken, so the engine won't lift it.

The fix is simple: write each section so it stands alone. Restate the subject by name instead of leaning on "it" or "this approach." Repeat a key number if the section needs it. That feels slightly redundant to you as the writer. It reads as clarity to a skimmer, and it reads as liftable to a machine.

Rewrite one page answer-first this week

Don't overhaul the whole site. Pick one page that targets a real buyer question and run this sequence:

1. Pick the page and the question. Choose a page tied to a question buyers actually ask an AI — a pricing, comparison, or "how do I" page. Write the question down in their words.
2. Write the 40–60 word direct answer. Answer the question completely, in plain language, with your key specifics. No warm-up, no "it depends" without the follow-through.
3. Restructure the headings. Turn each H2 and H3 into the question that section answers. Put a direct answer under each one before the elaboration.
4. Add a TL;DR and convert prose to structure. Three to five complete-claim bullets up top. Turn comparisons into tables, steps into numbered lists, and buried definitions into one-liners.
5. Add a visible FAQ with matching JSON-LD. Five to seven real buyer questions, each with a direct answer first. Mirror them exactly in FAQ schema.
6. Cut every sentence that carries no information. Read the page once, deleting anything that could sit on a competitor's site unchanged. What's left is denser — and more quotable.

Then wait, and check. Ask ChatGPT and Perplexity the target question over the following weeks and note whether you appear. One check proves nothing — answers are non-deterministic and vary by user, region, and day — so track the trend, not a single result. How to measure AI visibility turns that into a simple routine.

Read next

• Generative engine optimization: the full guide — the complete CLEAR framework this post's two levers come from.
• Why your brand isn't showing up in AI answers — crawlability, the step before this one.
• How to measure AI visibility and share of voice — proving whether the rewrite worked.