How do I get my site cited by AI assistants?

Stage one: be readable at all

Nothing below matters if this is wrong, and it is wrong more often than people expect.

1. Serve content in HTML. Fetch your page with curl -sL. If your text is not in the response, AI crawlers do not see it. Use server-side rendering or static generation.
2. Allow the right crawlers. Search and live-fetch crawlers — OAI-SearchBot, Claude-SearchBot, Claude-User, PerplexityBot, Perplexity-User, Google-Extended — are the ones that decide whether you can be cited. Training crawlers are a separate choice.
3. Check for accidental blocks. A noindex tag, an X-Robots-Tag header, or a CDN bot-protection rule will each remove you independently of robots.txt.
4. Return real status codes. A page that serves 200 for content that does not exist confuses every consumer of your site.

Stage two: be extractable

Answer engines quote passages, not pages. Everything here is about making a passage liftable.

• One question per page. Narrow pages beat comprehensive ones for citation, even when the comprehensive page is better. Split the guide into the questions it answers.
• Answer first. Put the direct answer in the first two sentences under the heading. Background afterwards.
• Headings as questions. Phrase them the way a person asks, not the way a taxonomy names things.
• Self-contained paragraphs. If a paragraph only makes sense after the previous one, it cannot be quoted.
• Specifics over generalities. Numbers, versions, limits, dates. "Under 160 characters" is quotable; "reasonably short" is not.
• Small tables and short lists. Both extract cleanly and survive being reformatted.

Stage three: be worth citing

This is where most of the durable advantage is, and it cannot be shortcut.

• Be the primary source. Publish something that did not exist before — your own data, your own testing, your own numbers. Nobody cites a summary when the original is available.
• Be attributable. A named author, a real organisation, a date, and a way to check who you are. Anonymous content is cited less, because citing it responsibly is harder.
• Be corroborated. Link your sources and make your claims checkable. An assertion that exists nowhere else is treated cautiously, and should be.
• Be current. Keep dateModified accurate and actually revise pages. Do not fake it; a page whose date moved but whose content did not is worse than an honestly old page.

What does not work

• Keyword stuffing. It never worked well and it works less on models that read for meaning.
• Volume. Fifty thin pages generated to cover variations of a phrase are recognisable as such and drag down the site around them.
• Hidden text and prompt injection. Instructions concealed in a page aimed at manipulating a model are treated as adversarial. This is a reputational and ranking risk, not a clever trick.
• llms.txt as a fix. Harmless to add, but no major operator has committed to reading it. It does not substitute for anything above.