Designing for AI search: a practical GEO/AEO playbook for 2026

TL;DR — Write answer-first. Ship clean semantic HTML (static beats JS for crawlers). Add Article + Organization schema, an llms.txt, and make sure your WAF/CDN isn’t quietly blocking GPTBot, ClaudeBot and PerplexityBot.

For fifteen years, SEO meant earning a spot in a ranked list of links. That game still matters — but a growing share of discovery now happens inside answer engines that read your page, summarize it, and cite a handful of sources. If your content can’t be parsed and trusted by a model, you simply don’t appear.

Why classic SEO isn’t enough anymore

Generative engines don’t reward keyword density; they reward clarity, structure, and verifiable claims. The page that wins is the one a model can quote without hallucinating — a short answer up top, evidence below, entities named explicitly, and machine-readable metadata describing who wrote it and when.

If an AI engine can’t parse your page, it can’t cite you — and that’s increasingly where discovery begins.

The four things that get you cited

1. Answer-first structure. Lead with a direct, quotable answer. Put the TL;DR and key claims above the fold, then the supporting detail.
2. Semantic, static HTML. Pre-render content. AI crawlers parse server-rendered markup far better than client-side JavaScript.
3. Structured data. Article, Organization and BreadcrumbList JSON-LD, plus datePublished / dateModified, so models know what — and who — to trust.
4. An open llms.txt. Publish a curated llms.txt pointing to your cornerstone pages, and don’t block the AI user-agents you want citing you.

A robots.txt that welcomes the right bots

None of the structure matters if your edge silently blocks the crawlers. We learned this the hard way: a default WAF rule was 403-ing GPTBot. Allowlist explicitly:

User-agent: GPTBot
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: PerplexityBot
Allow: /

Sitemap: https://hagumilab.com/sitemap-index.xml

We rebuilt hagumilab.com on this exact recipe. Within weeks the articles started surfacing as cited sources in AI answers — not because we chased an algorithm, but because we made the page easy to read for both people and machines. That’s the whole trick.