How to Write Content That LLMs Actually Cite (The GEO Playbook)

A practical playbook for writing articles, docs, and landing pages that generative engines read, trust, and cite — with real before/after examples.

How to Write Content That LLMs Actually Cite

Generative engines do not "rank" pages. They retrieve, evaluate, and cite. If your content is hard to parse, vaguely worded, or buried under fluff, the model skips it and cites a competitor.

Here is the playbook we use to make every page citation-worthy.

1. Open With the Answer, Not the Backstory

Answer engines extract the first clear statement. Bury it under a three-paragraph intro and you lose the citation.

Weak:

In today's fast-paced digital landscape, businesses of all sizes are increasingly recognizing the importance of structured data…

Strong:

Structured data (JSON-LD) tells search engines what a page is about in a machine-readable format, which can earn rich results like star ratings and FAQ dropdowns.

The second version is quotable. The first is wallpaper.

2. Write Q&A-Shaped Headings

Models map headings to questions. ## How do I add JSON-LD to Next.js? is a retrieval target. ## Implementation Details is not.

Turn every ## heading into a question a real user would type:

## Configuration
## How do I configure the sitemap?

3. Mark Up the Structure

Generative engines read structured data to disambiguate entities. The high-value schemas for a micro-SaaS:

Schema	What it declares	Where it goes
`SoftwareApplication`	Your app, pricing, category	`index.html`
`Organization`	Publisher identity	`index.html`
`BlogPosting`	Article metadata	Article pages
`FAQPage`	Question/answer pairs	Support, pricing
`BreadcrumbList`	Page hierarchy	All pages

SpaceZero's Discovery Pack generates the sitemap, llms.txt, RSS, and per-route WebPage schemas at build time from a single route config. Every blog and knowledge-base article is additionally prerendered to its own static page with BlogPosting + BreadcrumbList JSON-LD and full crawlable content.

4. Publish an llms.txt

llms.txt is the robots.txt of the AI era — a curated map of your site for generative crawlers. A minimal llms.txt:

code

# MyApp

> One-line description of what the app does.

## Core Pages

- [Home](https://myapp.space/): What the product is

- [Pricing](https://myapp.space/#pricing): Plans and prices

- [Docs](https://myapp.space/kb): How it works

## API

- POST /api/endpoint: What it does

Keep it under a few hundred lines. Link to the pages that answer questions, not every page.

5. Be Consistent About Entities

Models build an entity graph. If you call your product "MyApp" on the homepage, "My App" in the footer, and "myapp.io" in the schema, the graph fragments.

Pick one name, one category, one price, and use them everywhere — title, schema, footer, emails.

6. Publish a Changelog and an RSS Feed

Freshness is a trust signal. A dated changelog with an RSS feed tells both search engines and AI crawlers that the project is alive. SpaceZero generates rss.xml and a /changelog page from CHANGELOG_ENTRIES at build time.

The Checklist

[ ] First 40 words answer the page's core question.
[ ] Headings are phrased as questions.
[ ] JSON-LD covers app, org, and article.
[ ] llms.txt exists and is current.
[ ] Entity names are consistent everywhere.
[ ] A dated changelog with RSS is published.

Do these six things and your content stops being invisible to the models that increasingly mediate discovery.