Your page ranks on the first page of Google. Then a prospect asks ChatGPT the same question and it recommends three competitors instead. Nothing is broken. The page simply is not built the way a generative engine reads.
Knowing how to write content AI will cite is now a formatting discipline as much as a writing one. The engines are not judging your prose. They are looking for passages they can lift, quote and attribute without risk.
Below are five content structures that make that possible, with an example of each. None of them require new research or a rewrite of your site.
Key Takeaways
- AI engines quote passages, not pages, so each section has to stand on its own.
- Question-led H2s and answer-first paragraphs are the two highest-return changes you can make.
- Structured data helps machines confirm what your page is about, and who wrote it.
- Inconsistent business details across the web weaken your chances of being recommended.
- Citing real, linked sources makes your claims safer for an engine to repeat.
- Most sites need 10 to 20 pages restructured, not the whole library.
Why does structure decide whether AI engines cite you?
Generative engines do not read a page from top to bottom the way a person does. They retrieve and quote passages. A passage only gets used if it makes sense in isolation, so your structure decides which parts of the page can be lifted at all. Badly chunked content gets skipped, whatever its quality.
That is a commercial problem, not a technical one. A citation in an AI answer puts your name in front of a buyer at the exact moment they are shortlisting.
The mechanics of that shortlist are worth understanding before you start editing. We covered them in how ChatGPT chooses which businesses to recommend.
How to write content AI will cite: the five structures
Write content AI will cite by using question-led subheadings, answering each question in the opening sentence, adding structured data, keeping your entity details consistent across the web, and citing real sources with links. Those five structures make every section quotable on its own. Apply them in that order.
1. Question-led subheadings
Turn your H2s into the questions your buyers actually type. People do not ask AI engines about “our approach to service delivery”. They ask “how much does commercial cleaning cost per square foot in London”.
A question-led heading gives the engine a clean match between the query and your content block. It also tells the model exactly where the answer begins and ends.
Example. Replace “Our Pricing Model” with “How much does commercial cleaning cost in London?” The content underneath barely changes. The retrievability does.
Commercially, this is the cheapest fix on the list. It is a find-and-replace exercise across your existing pages, and it usually takes an afternoon.
2. Direct-answer-first paragraphs
Answer the heading’s question in the first 40 to 60 words underneath it. One idea, one paragraph, no run-up.
Most marketing copy warms up first: context, a bit of positioning, then the answer in paragraph four. An engine reading that block has two bad options. Quote the whole thing, which is too long, or pull a fragment from the middle and risk misrepresenting you.
Example. Under “How long does an office fit-out take?”, open with “A standard 5,000 square foot office fit-out takes 8 to 12 weeks from design sign-off to handover.” Then expand on the variables.
You can still write with personality. You just spend it after the answer, not before it.
3. Structured data
Add schema markup so machines can confirm what your page is, who wrote it and what it claims. Article, Organization, LocalBusiness, Service and FAQPage are the ones most SMEs need.
Google’s own structured data documentation is the reference to work from. Note that FAQ rich results in Google are now restricted to a narrow set of sites, but the markup still gives any parser a clean, unambiguous read of your questions and answers.
Example. An Organization block that states your legal name, registered address, founding date and social profiles gives an engine a verified anchor for your brand rather than a guess.
Think of schema as removing doubt. Engines avoid citing sources they cannot verify.
4. Entity consistency
Make sure your business name, address, service descriptions and founder details match everywhere they appear: your site, Companies House, your Google Business Profile, LinkedIn, directories and press mentions.
Generative engines build a picture of your business by cross-referencing sources. Contradictions lower confidence, and low confidence means no recommendation.
Example. Imagine a 12-person cleaning firm in Croydon trading as “Croydon Clean Ltd” on Companies House, “Croydon Cleaning Services” on Google and “CCS Facilities” on LinkedIn. That is three entities as far as a model is concerned, each too thin to trust.
This is the least glamorous item here and often the most valuable. Our GEO and AEO services start with exactly this audit.
5. Real sources and citations
Link out to primary sources for every statistic, standard or regulation you mention. Government pages, official documentation and named published research.
Engines weigh whether a claim is safe to repeat. A number with a linked, credible source behind it is safe. An unattributed number is a liability, and the surrounding passage often gets dropped with it.
Example. If you cite a change to Building Safety Regulator requirements, link to the gov.uk page. If you cite Google’s stance on content quality, link to Google’s creating helpful content guidance.
The same principle applies to your own claims. Named authors, dates and job titles give an engine something to attribute.
What does a citable page look like next to a standard one?
A citable page differs from a standard one in five visible ways: its headings are questions, its answers come first, its markup is machine-readable, its entity details are consistent, and its claims are sourced. The underlying information is often identical. Only the packaging changes.
| Element | Typical blog page | Page built to be cited |
|---|---|---|
| H2 headings | Topic labels (“Our Process”) | Buyer questions (“How long does it take?”) |
| Opening paragraph | Scene-setting intro | 40 to 60 word direct answer |
| Paragraph length | 5 to 8 sentences, several ideas | 1 to 3 sentences, one idea |
| Schema | Basic or none | Article, Organization, FAQPage |
| Statistics | Unlinked or vague | Named source, linked |
| Author | “Admin” or unattributed | Named, with credentials |
Where should you start if you already have 80 pages?
Start with the 10 to 20 pages that already earn traffic or leads, and restructure those first. Restructuring works on existing content, so you are not commissioning new research. Prioritise commercial pages and any post that answers a genuine buying question, then move down the list.
A practical order of work:
- Export your top 20 pages by conversions and impressions.
- Rewrite every H2 as a question a buyer would ask.
- Add a 40 to 60 word direct answer under each one.
- Add or fix schema on those pages.
- Fix your entity details across every external profile.
- Test the questions in ChatGPT, Perplexity and Gemini, and record who gets cited.
That last step matters, because the three engines do not behave identically. We break down the differences in Perplexity vs ChatGPT vs Gemini: how to get cited across all three.
Sector context changes the priorities too. For a worked example in one industry, see GEO for facilities management companies.
Frequently asked questions
Do I need to rewrite my whole site?
No. Restructuring is not rewriting. In most cases you keep the information and change the packaging: question-led headings, an answer-first opening paragraph under each, shorter paragraphs and schema. Start with the 10 to 20 pages that already drive traffic or enquiries. Sites with hundreds of thin pages are a different problem, and that is a pruning decision rather than a formatting one.
How often should I update content for AI engines?
Review commercial and high-traffic pages every quarter, and update immediately when a fact, price, regulation or product detail changes. Freshness signals matter most where the answer genuinely dates. Evergreen explainers can go six to twelve months between reviews. Constant cosmetic edits achieve nothing, so change something substantive or leave the page alone.
Does schema markup guarantee an AI citation?
No. Schema removes ambiguity about what your page says and who published it, which improves your odds. It does not manufacture authority. A well-marked-up page with weak, unsourced content still loses to a plainer page with better answers. Treat structured data as a supporting signal, not the strategy.
Will writing this way make content worse for human readers?
Generally the opposite. Question-led headings and answer-first paragraphs match how people skim, so time on page and scroll depth usually improve. The risk is tonal, not structural: applied mechanically, this style reads like a manual. Keep your voice in the sentences that follow the answer, and the page works for both audiences.
How do I know whether AI engines are citing my content?
Ask the engines directly. Run your target questions through ChatGPT, Perplexity and Gemini monthly and log which sources get named. Then check your analytics for referral traffic from AI domains. Many citations carry no clickable link, so mentions and branded search volume are better indicators of progress than sessions alone.
Getting this done properly
Structure is a discipline, not a one-off project. The five techniques above will move most sites, but the gains compound when they are applied consistently across every page you publish, backed by clean entity data and monitored across all three major engines.
That is the work we do for UK SMEs and purpose-driven organisations every week.
