How I got ChatGPT and Perplexity to actually cite my site (the exact setup)

AI search is quietly eating a chunk of the clicks that used to go to Google. When someone asks ChatGPT, Perplexity, or Google's AI Overview a question, they often get an answer with sources and never click through to ten blue links. So the new game is not just to rank, it is to get cited.

I run a static content site (about 130 articles, plain HTML, no framework). Over the last few weeks I instrumented it for AI search and a few of our pages started showing up as cited sources in Perplexity and inside ChatGPT's web answers. Here is the honest, build-in-public version of what actually moved the needle, with the code, so you can copy it onto your own site today. None of this needs a CMS or a plugin.

1. Ship an llms.txt

This is the robots.txt of the AI era: a plain-markdown file at your root that hands models a curated map of your best content, so they do not have to guess from your nav. Spec is at llmstxt.org. Mine looks like this:

# AI Tools Insider

> Independent, hands-on reviews of AI tools, plus free no-signup utilities.

## Best buyer-intent guides
- [Best AI SEO Tools 2026](https://aitoolsinsiderhq.com/articles/best-ai-seo-tools.html): ranked comparison
- [ChatGPT vs Claude vs Gemini](https://aitoolsinsiderhq.com/articles/chatgpt-vs-claude-vs-gemini.html): which to pick

## Free tools
- [AI Search Visibility Checker](https://aitoolsinsiderhq.com/ai-search-visibility-checker.html): score any page for AI citation odds

Keep it short, link only your genuinely best pages, one line of context each. Then reference it from robots.txt so crawlers find it.

2. Turn your existing Q&A into FAQPage JSON-LD

This was the single highest-leverage change. LLMs love clean question/answer pairs because they map directly onto how people prompt. If your article already has a visible FAQ section, you are 90% there, you just mirror it as structured data. No new content:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [{
    "@type": "Question",
    "name": "Is Semrush worth it in 2026?",
    "acceptedAnswer": {
      "@type": "Answer",
      "text": "For teams doing serious SEO, yes. Solo creators can start with the free tier and Google Search Console."
    }
  }]
}
</script>

Rule I follow: the schema answer must match the visible answer on the page. Do not invent Q&A that is not really there, that is exactly what Google's spam systems and the LLMs' own quality filters punish.

3. Answer-first structure (write the lede like a snippet)

Models lift the first 1-3 sentences under a heading. So I rewrote the top of each page to lead with a direct, self-contained answer, then expand. A quick test: read the first sentence after each <h2> out loud. Does it answer the heading on its own? If not, rewrite it.

4. Give yourself an entity (Organization + sameAs)

LLMs reason about who is saying something. A linked Organization block with sameAs pointing at your real profiles, plus knowsAbout topics, helps the model connect your pages to one consistent author entity instead of treating each URL as an orphan.

How to check it without guessing

Validate every JSON-LD block in Google's Rich Results Test (it catches a single stray comma). To gauge citation-readiness specifically, I built a free, no-signup page scorer that grades the same signals above out of 100 with a priority fix list: https://aitoolsinsiderhq.com/ai-search-visibility-checker.html

The honest caveat

This is not a growth hack and it is not instant. AI citations compound over weeks as crawlers re-index and your entity gets reinforced. What it is: a set of one-time, zero-cost structural changes that make your existing content legible to the systems answering more and more queries.

If you want the running list of what we publish and the free tools, the map is at https://aitoolsinsiderhq.com/llms.txt, which is, fittingly, the same file the models read.