Two pages can hold the same information and get wildly different results in AI search. The facts are rarely what separates them. The shape is. Answer engines favour content they can pull apart and reuse, and you can learn to write it.
The short answer: to get quoted by ChatGPT, Perplexity or Google's AI Overviews, put the direct answer first, break the page into self-contained blocks under question-style headings, and publish what only you can say, like your own numbers and first-hand experience. A clear, specific sentence gets cited. A vague, meandering one gets skipped.
Plenty of businesses know their subject better than the pages that get quoted ahead of them. They just write it in a shape a machine cannot use.
Why AI quotes some pages and ignores others
An answer engine does not read your page the way a person does. It looks for passages that resolve a question cleanly, still make sense when lifted out of context, and come from a source it has reason to trust. Then it stitches those passages into an answer and, if you are lucky, names you.
That changes what "good content" means. A beautifully written essay that builds slowly to its point can be a joy to read and almost useless to a model. The point is buried, the paragraphs lean on each other, and no single sentence stands up on its own.
Meanwhile a plainer page with a crisp first sentence and one idea per section gets quoted all over the place. Same facts, different shape, different result. For how this fits alongside traditional SEO, our breakdown of AEO, GEO and SEO covers the whole alphabet.
Lead with the answer
Ditch the slow windup. State the conclusion in the first sentence or two, then support it underneath. Bury your answer three paragraphs down and you make the model dig for it, while a competitor's page offers the same thing in line one.
Newspapers call this the inverted pyramid and have used it for well over a century. Poynter traces the form to the telegraph era after the American Civil War. The most important thing goes first, the detail follows, the background comes last. It works for machines for the same reason it works for busy humans. Nobody has to hunt.
Test your own page. Read only the first two sentences under any heading. Do they answer the question the heading asks? If they are scene-setting or company history, move them down the page.
- Weak opener: "When it comes to choosing how to bill clients, there are many factors every business needs to consider, and every situation is different."
- Strong opener: "For most small service businesses, a monthly retainer beats hourly billing if you want predictable cash flow, while hourly billing suits one-off jobs where the scope is unclear."
The second version can be lifted straight into an answer. The first cannot.
One idea per chunk, headings as questions
Break content into tight, self-contained blocks under headings phrased the way people ask, like "How much does X cost?" or "X vs Y: which is better?" Each block should stand alone as a complete answer a model can lift cleanly.
"Self-contained" is the key word. If a paragraph starts with "As mentioned above" or "This is why", it depends on something the model may not grab. Name the subject again. Repeat the key noun where you would normally write "it" or "this". It will feel slightly redundant to you. A model extracting one block at a time needs it.
The questions come from your inbox and your sales calls. Whatever customers ask before they buy is what they type into an assistant. Keep a running list and turn the best ones into headings.
The formats AI loves most
- Original statistics: A specific, sourced number is about the most citable thing you can publish. Run a survey, share your own data, coin a stat.
- Comparisons and "vs" tables: Buyers ask engines to compare options constantly. Give them the clean side-by-side to quote.
- Numbered steps and lists: Easy to scan, and easy for a machine to reproduce in an answer.
- Plain-English definitions: Clear "X is…" statements get pulled straight into explainer answers.
When a team from Princeton and IIT Delhi tested ways to make pages more visible in generated answers, adding statistics, quotations and cited sources worked best, lifting visibility by 30 to 40%. Keyword stuffing did close to nothing.
This is also where people get tempted to cut corners. "Coin a stat" means publish a number from your own work. It never means make one up. Average turnaround time across your jobs this year. The most common fault you found in the last hundred inspections. What your customers spend before they come to you. Small, honest data from your own business beats a recycled industry figure, because nobody else can publish it.
Comparisons work because they match how people shop. Nobody asks an assistant "tell me about heat pumps". They ask "heat pump or gas hot water, which is cheaper to run?" If your page answers that directly, with the trade-offs laid out honestly, the model has a ready-made reply with your name on it.
Write for the human, structure for the machine. The pages that do both get read and quoted.
How to rewrite an existing page for AI citation
You do not need to start from scratch. Most businesses have pages with the right knowledge in the wrong shape. Work through the rewrite in this order.
- Pick one page that should already be winning. Usually a service page or your most useful guide.
- List the questions it should answer. Five to ten, in customer language. Check them against the enquiries in your inbox.
- Turn the best ones into headings. Phrase them as questions or as clear statements of what the section resolves.
- Rewrite the first sentence of every section as the answer. Then support it. Then add nuance.
- Break up any paragraph that carries two ideas. One idea, one block.
- Add one thing only you can say. A number from your own work, or a mistake you see constantly on the job.
- Add a short FAQ. Direct questions, answers of a few sentences, each one standing alone.
- Put a name on it. Author, role, experience.
For the three edits that pay off fastest, see three edits that get your business quoted by ChatGPT.
A worked example: the plumber's hot water page
Say you run a plumbing business in Perth with a "Hot Water Systems" page. Right now it opens with two paragraphs about your family history, then a list of brands you install, then "call us for a quote". The knowledge a model wants is all in your head and none of it is on the page in usable form.
The rewrite opens with a direct answer: which hot water system suits most households in your area, and why, in two sentences. Then come headings built from the questions your office fields every week: "Heat pump vs gas vs electric: which is cheapest to run?", "How long does a hot water system last?", "What does hot water replacement cost?", "Signs your hot water system is about to fail". Under the cost heading you give honest ranges from your own jobs and explain what pushes a job to the top of the range.
Then you add the thing only you can say: the most common reason you see systems fail early in local conditions, from years of call-outs. Finally, the page is signed by the licensed plumber who wrote it, with their licence type and years in the trade.
The business is the same. The page now contains a dozen passages a model can lift cleanly, several of which exist nowhere else online. That is a hypothetical, but it is exactly the shape of the change.
Sign your work
A named author with credentials and first-hand experience signals a trust that thin, AI-spun content can't fake. Google spells it out. Its guidance for site owners strongly encourages accurate authorship information, such as bylines, and asks whether your content clearly demonstrates first-hand expertise.
In practice that means an author name on every article, a short bio stating what qualifies them, and an author page that links to their other work and their profiles elsewhere. It means writing "we tested", "on our last job", "the mistake we see most" whenever that is true, because generated filler has no last job to talk about. And it means your business details are consistent everywhere, so the model can connect the page to a real organisation. We go deeper on that in how to make AI trust your brand.
The mistakes that stop you getting quoted
- Writing for word count. Padding dilutes the good passages. The answer gets lost in the noise.
- Hiding answers behind a form or a call. "Contact us for pricing" gives the model nothing, so it quotes the competitor who published ranges.
- Clever headings. "The heat is on" tells a machine nothing. "How long does a hot water system last?" tells it what the section answers.
- Mass-producing AI copy. If your page says what a model would say anyway, it has no reason to cite you.
- Burying key content in images or PDFs. If the answer lives in a graphic, many systems cannot read it. Put the text on the page.
How to tell if it is working
Measurement is still rough, so combine a few signals. Each month, ask the main assistants the questions your pages answer and log every time you are named or linked. Watch referral traffic from chatgpt.com and perplexity.ai in your analytics, and read it as a minimum, since some AI visits get counted as Direct. Watch branded search in Search Console, because people who see you recommended often search your name next. And ask new enquiries how they found you, adding "an AI assistant" as an option.
No single signal is reliable. Four of them moving the same way over a few months is.
Content that earns the citation
The businesses that win AI search will not be the ones publishing the most. They will be the ones whose pages are easiest to quote and hardest to replace, with direct answers and their own numbers under a named author.
That is the work we do in our digital marketing and AI search service at LIQID NOISE. We restructure the pages you already have, build the ones you are missing, and make sure the expertise inside your business actually shows up where your customers are asking. You already know the answers. We make sure the machines can find them.
Frequently asked questions
How do I get my website quoted by ChatGPT?
Make your pages easy to extract and worth citing. Put a direct answer in the first sentence under each heading, keep one idea per section, and phrase headings like the questions customers ask. Add something original, such as data from your own work. Sign content with a named, qualified author.
What content format do AI answer engines prefer?
Formats that can be lifted cleanly into an answer. Plain-English definitions, numbered steps, honest comparisons between options, short FAQs and original statistics all work well. What they share is that each block makes sense on its own, without the surrounding page. Long, winding paragraphs that build slowly to a point are the hardest format for a model to quote.
Does writing for AI search hurt readability for humans?
No, it usually improves it. Leading with the answer and keeping one idea per section is good writing for busy readers. People skim before they read, and a page structured for extraction is also easy to skim. All you lose is the slow windup, and very few readers will miss it.
Should every article have an FAQ section for AI search?
Most informational pages benefit from one. A short FAQ gives answer engines ready-made question and answer pairs. Keep answers direct and self-contained, use the wording customers search with, and avoid repeating the body word for word. Three strong questions beat ten thin ones.