The mental model: you are writing for a passage lifter
A search engine indexes your page and ranks it. A model retrieves a passage from your page and decides whether to lift it into an answer it is already composing. Every technique below follows from that one difference.
Useful test before publishing: copy any single paragraph out of your page, paste it somewhere with no context, and read it. If it no longer makes sense or no longer says who it is about, a model cannot use it either.
What reliably increases citation rate
- Answer first, then expand. Open each section with one or two sentences that fully answer its heading, before any context or caveats. This is the single highest-leverage change and it costs nothing.
- Write self-contained passages. Repeat the subject instead of using "it" or "this", because the paragraph will be read without the ones above it.
- Be specific and quantified. "Forty to sixty prompts" gets lifted; "a reasonable number of prompts" does not. Numbers, ranges, dates and named conditions are what models quote.
- Use headings that are the actual question. A heading reading "How often should I sample" retrieves better than "Cadence", because it matches the shape of what buyers ask.
- State facts models otherwise guess. Pricing, limits, supported integrations, coverage, availability. When your site is silent, the model fills the gap from stale third-party pages.
- Add structured data that matches the visible text. FAQPage, Product with offers, DefinedTerm and Organization all give unambiguous facts, but only if the markup and the page agree.
- Show freshness honestly. A visible published and updated date helps retrieval on pricing, version and comparison questions. Changing the date without changing the content does not.
- Build corroboration off-site. Models cross-check across sources, so a claim that appears only on your own domain carries less weight than one echoed in independent coverage, directories and documentation.
The page types that get cited most
Citation is not evenly distributed across a site. Four formats do most of the work, and most B2B sites are under-invested in all four.
- Definitional pages. A glossary entry that defines a term cleanly in one sentence is among the most liftable content that exists.
- How-to guides with numbered, concrete steps and stated constraints.
- Comparison pages, provided they are honest. Models penalise pages that only flatter the publisher by preferring third-party comparisons instead.
- Documentation and pricing pages, because they are the canonical source for facts nobody else can state authoritatively.
What to stop doing
- Do not gate the answer. Content behind a form or a login cannot be retrieved, so a gated report earns no citations at all.
- Do not render the substance in JavaScript only. Several AI crawlers do not execute it, and the page arrives functionally empty.
- Do not write prompt-injection style instructions into pages. Attempts to instruct the model in hidden text are detected, ignored, and put your domain at risk.
- Do not publish thin pages at volume. Retrieval rewards the one page that answers a question completely, not thirty that circle it.
How to know whether it worked
Citation rate moves slowly and unevenly, so judge it on the right horizon. Re-crawling and index refresh alone typically take one to three weeks, and the first thing to change is usually which of your pages gets cited, not how often you are mentioned.
Track cited URLs per prompt, not just a site-wide count. A rewrite that moves citations from a blog post to your pricing page has changed something real about how the model describes you, even if the total stays flat.
Common questions
Make sure the page is crawlable and server-rendered, then structure it so each section opens with a self-contained answer containing specific facts, numbers and the subject's name rather than pronouns. Pages that state facts nobody else can state authoritatively, such as pricing and documentation, are cited most often.
Re-crawling and index refresh alone typically take one to three weeks after a page changes. The first visible effect is usually a shift in which of your pages gets cited, rather than an increase in how often your brand is mentioned.
It helps by removing ambiguity about facts, but only when the markup matches the visible text on the page. Structured data that contradicts the page, or describes content that is not there, is discounted and can damage trust in the domain.
Yes, if they are honest. Comparison questions are among the most common buying-stage prompts, but models tend to prefer third-party comparisons over pages that only flatter their publisher, so a page that states genuine trade-offs is more likely to be cited than one that does not.