Generative Engine Optimization (GEO): the 33-signal checklist for 2026

Generative Engine Optimization (GEO) is what SEO is becoming. The "blue link" layer of search is being replaced by the "answer" layer: ChatGPT browse, Claude artifacts, Perplexity citations, Google AI Overviews, MCP-aware assistants. They decide what to cite based on a different signal set than classical SEO.

This is the 33-signal checklist we use at AgentFix to scan sites. Hit 26+ to land in the agent-native top tier.

Tier 1 — discovery foundation (every site)

1. /llms.txt — curated table of contents for agents. Setup guide.
2. /llms-full.txt — long-form text corpus. ~20-80 KB sweet spot.
3. sitemap.xml at root — URL inventory.
4. robots.txt allowing AI crawlers — GPTBot, ClaudeBot, PerplexityBot, Google-Extended explicitly Allow. Full list.
5. HTTPS with valid cert — most agents skip HTTP-only sources.
6. Canonical <title> on every page — agents use this as the citation label.
7. Concise <meta name="description"> — your fastest pitch.

Tier 2 — semantic markup (where most SEO plugins help)

8. schema.org Organization JSON-LD on home + about. Templates.
9. schema.org WebSite + SearchAction — unlocks sitelinks search box.
10. schema.org FAQPage — agents lift answers verbatim into citations.
11. schema.org Article on every post — author, datePublished, headline.
12. schema.org Product on every product page (commerce).
13. Open Graph tags — og:title, og:description, og:image. Used for social previews and citation tiles.
14. Twitter Card — summary_large_image.
15. Heading hierarchy — single h1, h2/h3 in order. Agents anchor on heading text.

Tier 3 — agent-native protocols (where competitors don't ship yet)

16. /.well-known/agent-card.json — Agent-to-Agent (A2A) protocol manifest. Declares what skills your site exposes to other agents.
17. /.well-known/mcp/server-card.json — Model Context Protocol server manifest. Cursor, Continue, Claude Desktop bind to this.
18. /openapi.json (OpenAPI 3.x) — formal API description. Agents call your endpoints programmatically when present.
19. /.well-known/api-catalog (RFC 9727) — linkset+json catalogue of API descriptors. Aggregator for the above.
20. /.well-known/oauth-protected-resource (RFC 9728) — how an agent authenticates to your protected endpoints.
21. RFC 8288 Link headers — advertise service-desc, agent-card, llms.txt via HTTP headers, not just files. Lets agents discover via HEAD.
22. Content-Type: application/vnd.oai.openapi+json on the OpenAPI file — strict agents check this.

Tier 4 — content quality signals (most-cited sites do all of these)

23. Author byline with sameAs — link author profile to LinkedIn, Twitter, ORCID.
24. datePublished + dateModified on every post. Stale dates hurt citation.
25. Cite your sources — agents lean on pages that cite back; it signals research integrity.
26. Image alt text — agents that can't see images rely on alt for context.
27. Tap targets ≥ 44×44 px on mobile — Google's mobile-usability check, also used as a quality proxy by AI agents.
28. Page weight under ~1.5 MB — Perplexity has an undocumented size cutoff.

Tier 5 — frontier signals (most sites don't ship; outsized signal when you do)

29. /.well-known/acp.json — Agent Commerce Protocol. Lets agents transact directly.
30. /.well-known/ucp/payment.json — Universal Commerce Protocol payment manifest.
31. WebMCP client script — exposes runtime MCP server to agents that open your site in a browser.
32. DNS-AID records — DNS-based agent entrypoint discovery (_index._agents.yourdomain.com).
33. Stable cross-page entity graph — same JSON-LD @id across pages so agents can chain queries.

How to use this checklist

Three pragmatic paths:

1. Free — DIY across 1–2 weeks

Run npx agentfix-mini-scanner yoursite.com to see which of the first 12 signals you're missing. Use the guides linked above to fix each. Re-scan after each ship to verify.

2. Paid — ship everything in 5 minutes

The AgentFix Pack ($1 / $29 / $99 one-time) scans for all 33 signals and emails you a ZIP with the missing files generated from your real content — llms.txt, llms-full.txt, schema, A2A agent-card, MCP server-card, OpenAPI, install guides for WordPress / Webflow / Tilda / Shopify / cPanel.

3. Visualise — see your site as an AI agent does

workspace.agentfix.pro renders your site as an interactive iceberg. The tip is what Google sees; just below is what AI agents reach for; deep is the protocol layer; deepest is the frontier. Free, no signup.

Why this matters in 2026

Classical SEO was a popularity contest gated by Google's secret ranking factors. GEO is a specification — public RFCs, well-known endpoints, schema vocabularies. You either expose /.well-known/mcp/server-card.json or you don't. There's no algorithm guessing.

That's the first time in 25 years of the web that "discoverable to the answer engine" is a checkbox rather than a black box. The sites that complete the checklist get cited. The ones that don't get silence.

Related

• How to add llms.txt to WordPress
• llms.txt for Shopify
• schema.org JSON-LD for AI search
• robots.txt for AI bots