Getting Cited by ChatGPT, Perplexity & Claude: The AI SEO Playbook for 2026
Half of your future traffic won't come from Google's blue links — it will come from a sentence quoted by ChatGPT, Perplexity or Claude. Here is what actually makes AI engines cite you in 2026.
For twenty years "SEO" meant one thing: ranking blue links on Google. In 2026 that is only half the job. When someone asks ChatGPT for "the best prerendering service for React apps" or asks Perplexity to "compare Lovable and Bolt for SEO", the answer is a paragraph — with a handful of citations underneath. If you are in those citations, you get traffic and trust. If you are not, you are invisible.
The good news: getting cited by AI answer engines is not mysterious. It is a stricter, more literal version of good SEO. This playbook is the exact set of things that move the needle — no vibes, no "just write for humans" hand-waving.
Why AI SEO matters in 2026
The behavioural shift is already measurable. A material share of research and product-comparison queries now start in ChatGPT, Perplexity, Claude or Google's own AI Overviews instead of a search box. For those users you don't get ten blue links — you get one paragraph with three to five citations. Being one of those citations is the entire game.
- AI answer engines convert extraordinarily well. Users arrive with intent already resolved — they clicked because a machine said you were the answer.
- Citations compound. Once you are the quoted source for a topic, models keep quoting you as long as you keep the page current.
- Competition is thin. Most sites are still built for Google-only, so the barrier to being cited is remarkably low if you do the basics right.
How AI engines actually read your site
It's tempting to imagine AI models as super-Google. They are not. They are much more literal.
- GPTBot (ChatGPT), ClaudeBot / Claude-SearchBot (Anthropic), PerplexityBot and Applebot-Extended all fetch raw HTML. None of them execute JavaScript.
- They read the visible text of the page, plus the semantic HTML (headings, lists, tables), plus JSON-LD structured data. They largely ignore CSS visuals.
- They extract short, self-contained sentences and paragraphs — the kind that could stand alone as an answer.
- They prefer sources with consistent structure across pages: predictable headings, an obvious H1, clear entity naming.
This is the single biggest AI SEO mistake. If a curl request to your homepage returns an empty <div id="root"></div>, GPTBot and ClaudeBot see exactly that — nothing. No amount of great content matters until they can read it.
Step 1 — Let the AI crawlers in
Check your robots.txt. Many templates ship a default that blocks GPTBot and friends on the assumption you want to "protect" your content. If your goal is to be cited, that assumption is backwards — you are protecting yourself into invisibility.
The right robots.txt for a site that wants AI citations:
User-agent: *
Allow: /
# Explicitly allow AI answer engines
User-agent: GPTBot
Allow: /
User-agent: ClaudeBot
Allow: /
User-agent: Claude-SearchBot
Allow: /
User-agent: PerplexityBot
Allow: /
User-agent: Applebot-Extended
Allow: /
Sitemap: https://yourdomain.com/sitemap.xmlStep 2 — Be readable without JavaScript
This is where most modern SPAs fail. React, Vue, Svelte and every Lovable-style AI builder ship client-rendered by default. curl your own homepage:
curl -A "PerplexityBot/1.0" https://yourdomain.comIf the response is mostly a <script> tag and an empty root div, no AI engine can read you. The fix is prerendering — a lightweight proxy that renders your app in a real browser when a bot visits and serves the finished HTML. RenderlySEO does this at the DNS layer with no code changes, but the important thing is that you do it somehow. SSR, static export, prerendering — any of them work, as long as the HTML that reaches the crawler is complete.
Step 3 — Shape content for extraction
AI engines don't read your page the way a human reader does. They chunk it, embed the chunks, and pull out the ones most relevant to a query. Small structural choices make a huge difference in whether a chunk is quoteworthy.
Front-load the answer
The first 1–2 sentences under every H2 should directly answer the implied question. Save the context, the caveats and the "why" for the paragraphs that follow. Models overwhelmingly quote from the first sentence after a heading.
Use real headings, in order
One H1 per page. H2 for major sections. H3 for sub-topics. No skipping levels, no fake headings made with a big <div>. Semantic hierarchy is what tells a model where one topic ends and the next begins.
Prefer lists and tables for facts
If you're comparing options, use a real HTML table. If you're listing steps, use an ordered list. Extracted facts from a table or list survive re-ranking and re-summarisation much better than the same facts buried in prose.
Name entities the same way every time
Pick one canonical name for your product, competitors and concepts, and use it consistently across every page. "RenderlySEO" everywhere, not "Renderly SEO" on one page and "the Renderly platform" on another. Consistency is what lets a model recognise you as the same entity across sources.
Step 4 — Ship the right structured data
JSON-LD is the single highest-leverage AI SEO tactic per hour of effort. It gives models a machine-readable version of what your page is about, and they use it heavily when composing answers.
The four schemas that matter most:
- Organization / SoftwareApplication on your homepage — tells engines what you are and what category you compete in.
- FAQPage on any page with Q&A content — models extract entire question/answer pairs verbatim.
- HowTo on step-by-step guides — great for tutorials and setup docs.
- BreadcrumbList on every internal page — helps engines understand your site's hierarchy.
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [{
"@type": "Question",
"name": "What is prerendering?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Prerendering is a technique where a server renders a JavaScript app into complete HTML for crawlers, while real users still get the interactive single-page app."
}
}]
}
</script>Step 5 — Publish an llms.txt
llms.txt is the emerging convention for telling AI crawlers what your site is about and which pages matter. Think of it as a short, prose-friendly sitemap written for language models. It lives at /llms.txt at your project root.
# RenderlySEO
> Prerendering and JavaScript SEO for modern apps. Get indexed by Google and cited by AI answer engines in 5 minutes with a single DNS record.
## Core pages
- [How it works](/how-it-works): DNS, rendering and cache flow.
- [Features](/features): AI search, crawler proof and fixes.
- [Pricing](/pricing): Plans from $14.99 with a 7-day trial.
## Guides
- [Get your Lovable app indexed](/blog/get-lovable-app-indexed-by-google)
- [AI SEO playbook](/blog/ai-seo-playbook-chatgpt-perplexity-claude)Step 6 — Build citation-worthy authority
Technical hygiene gets you into the candidate pool. Authority is what gets you picked out of it. AI engines lean heavily on signals of trust — mostly the same signals classical SEO cares about, but weighted differently.
- Original data and benchmarks beat opinion. If you have real numbers, publish them.
- First-hand experience beats aggregation. Models can tell (and increasingly filter for) primary sources.
- Consistent author bylines with real names and roles are a positive signal.
- Inbound links from category-relevant sites still matter — they are how models discover you exist in the first place.
How to measure AI visibility
You can't optimise what you don't measure. Two lightweight practices that actually work:
- Pick 10–20 queries your customers would ask an AI engine. Run them weekly in ChatGPT, Perplexity and Claude. Log whether you're cited, which page is cited, and which competitors show up.
- Watch your server logs (or Cloudflare / RenderlySEO analytics) for GPTBot, ClaudeBot, PerplexityBot and Applebot traffic. Rising crawl volume on a page usually precedes rising citations.
Mistakes that get you filtered out
- Client-rendered pages with no prerender — AI engines see nothing.
- Blocking AI bots in robots.txt "just in case" — you are opting out of the citation lottery.
- Thin, keyword-stuffed pages — models score coherence, not density.
- AI-generated content with no original data or point of view — increasingly discounted.
- Cloaking or serving different HTML to bots and users beyond format — treat prerendered HTML as a faithful copy of what users see.
FAQ
Is AI SEO different from regular SEO?
It's the same foundation with stricter requirements. AI engines don't execute JavaScript, extract from clear structural signals, and reward self-contained answers. Everything that helps AI SEO also helps classical SEO.
Do I need to block AI crawlers to protect my content?
Only if you specifically don't want to appear in AI answers. For most sites the traffic and trust from being cited far outweigh the marginal risk of training-data inclusion.
How long until I start getting cited?
Once crawlers can read you and your structured data is in place, expect first citations within 2–6 weeks. Compounding is the interesting part — once you're the source for one query, related queries follow.
Does llms.txt actually do anything today?
It is not yet universally consumed, but publishing it is cheap and forward-compatible. Early signals suggest multiple major engines read it, and it also serves as a useful sitemap for your own reference.
Where does RenderlySEO fit in?
RenderlySEO handles the readability half — the DNS-level prerendering that makes sure GPTBot, ClaudeBot and PerplexityBot see complete HTML instead of an empty React shell. Everything else in this playbook — headings, JSON-LD, llms.txt, content shape — you own.
AI answer engines are still an underweighted opportunity for most sites. Get the basics right — crawler access, prerendered HTML, clean structure, honest structured data, a real llms.txt — and you'll be pulling citations from ChatGPT and Perplexity while your competitors are still arguing about whether AI search matters.