26 Criteria for AI-Readiness: How I Audited My Own Site (and Beat Three B2B SaaS Giants)

26 Criteria for AI-Readiness: How I Audited My Own Site (and Beat Three B2B SaaS Giants)

26 Criteria for AI-Readiness: How I Audited My Own Site (and Beat Three B2B SaaS Giants)

In May 2026 I ran a structured 26-criterion AI-readiness audit on my own site ivannikov.pro. Starting score: ~42/100. One day of fixes later: 80/100. Then I ran the same audit on Stripe, Twilio, and Linear. Twilio and Linear failed the E1 pre-condition gate — their public sites cannot be parsed by LLM crawlers without JavaScript. Stripe scored 55/100, missing fundamentals like llms-full.txt and explicit AI-policy in robots.txt.

If you run a B2B SaaS and rely on organic traffic, this matters. Discovery has shifted from Google to AI agents — ChatGPT, Claude, Perplexity, Google AI Overview. Sites that are not optimised for LLM consumption lose visibility every quarter. Most "AI SEO" advice is generic; this post gives you the concrete checklist.

The 26-criterion framework

Five categories, 100 points total, plus a hard pre-condition gate:

E1 gate (pre-condition)

Content accessible without JavaScript execution. LLM crawlers (GPTBot, ClaudeBot, PerplexityBot) do not run JS. If your site requires it to render content, the whole audit is UNRELIABLE — fix this first. This is where SPAs lose, hard.

A. Discovery (18 pts)

Can an LLM agent find your content from a cold start? Look for:

  • /llms.txt — site index in the llmstxt.org format.
  • /llms-full.txt — extended version with concatenated content.
  • /robots.txt with explicit Allow: for AI crawlers (not just User-agent: *).
  • /sitemap.xml with <lastmod> tags.

B. Per-page artifacts (22 pts)

On individual pages:

  • .md mirrors — /page.md returns clean Markdown. Single most effective fix for SPAs.
  • JSON-LD TechArticle / Article / Person / Organization schema in <head>.
  • <link rel="canonical"> present and accurate.
  • Last-Modified HTTP header or dateModified in JSON-LD.
  • OpenGraph + Twitter Card metadata.

C. API spec (25 pts) — skip if not an API product

  • OpenAPI on a predictable URL (/openapi.json).
  • Schema validates clean.
  • Per-endpoint examples (request + response).
  • Postman / Insomnia / Bruno collection linked.
  • SDK references near the docs.
  • Versioning in spec and URL.

D. Content (20 pts)

  • curl example per endpoint.
  • Realistic payloads (not {"foo": "bar"}).
  • Error codes documented.
  • Auth method clearly described.
  • Rate limits documented.
  • Glossary for domain-specific terms.
  • SDK code examples alongside REST.

E. Hygiene (15 pts)

  • No-JS access (the E1 gate above).
  • Stable URLs (no hash-routing).
  • Version in URL for citation stability.
  • TOS / AUP / AI policy published.

The full checklist with point allocations and verification methods is in the open-source repo.

Self-audit: ivannikov.pro

Before fixes (manual audit, 2026-05-16 morning):

| Category | Score | |---|---| | A. Discovery | 4/18 | | B. Per-page | 9/22 | | E. Hygiene | 10/15 | | Total (A+B+E, partial) | 23/55 → ~42/100 normalized |

C and D are partial / N/A for a portfolio site (no API product).

After fixes (same day, evening):

| Category | Score | |---|---| | A. Discovery | 15/18 | | B. Per-page | 12/22 | | E. Hygiene | 11/15 | | Total | 38/55 → ~69/100 normalized; 80/100 by automated script |

What changed in one day:

  1. Added /llms.txt with proper llmstxt.org structure (+6).
  2. Added /llms-full.txt with ~3000 words of canonical content (+5).
  3. Updated /robots.txt with explicit Allow: for 20+ AI crawlers including GPTBot, ClaudeBot, PerplexityBot, Google-Extended, anthropic-ai, cohere-ai, meta-externalagent (+3).
  4. Enriched JSON-LD Person schema with alternateName (handle variants), givenName/familyName, worksFor Organization schema, expanded knowsAbout to include AI/RAG/LLM/GEO (+2).
  5. Reordered services — AI-first (ai-optimisation, ai-development as top 2), Web3 demoted to adjacent capability.

Benchmark: B2B SaaS giants

I ran the automated subset of the audit against three popular B2B SaaS products:

| Site | E1 gate | Score | Notes | |---|---|---|---| | ivannikov.pro | ✅ PASS | 80/100 | After Week 1 fixes — SSG + llms.txt + robots AI policy + sitemap | | Stripe | ✅ PASS | 55/100 | Has llms.txt (no H1 though), no llms-full.txt, generic robots.txt, no sitemap.xml at root, no OpenGraph | | Twilio | ❌ FAIL | 32/100 UNRELIABLE | SPA — body empty without JS. LLM crawlers cannot read content. | | Linear | ❌ FAIL | 40/100 UNRELIABLE | Has llms.txt (218 lines!) and llms-full.txt, but main site is JS-rendered. |

The takeaway: two of three B2B SaaS giants are invisible to LLM crawlers from their public marketing sites. They likely get cited via secondary signals (docs subdomains, third-party mentions, training data lag), but they are leaving organic AI visibility on the table.

If your competitors are in similar shape, the first-mover window in your niche is wide open for 12–18 months.

Where most teams lose

Three patterns I see consistently:

1. SPA-by-default (the E1 trap)

Single-page apps render content client-side. To you it looks fine; to a crawler it's an empty <body>. Stripe got this right by serving HTML directly. Twilio and Linear are paying the SPA tax. Fixes range from "switch a few key pages to SSR" to "add per-page .md mirrors via a route handler". The Markdown mirror is the highest ROI fix.

2. No explicit AI policy in robots.txt

A bare User-agent: * is technically permissive, but explicit Allow: directives for GPTBot, ClaudeBot, etc. are a positive signal — and a place where you can also disallow if you want to. Most teams just have neither.

3. JSON-LD missing or stale

If you have a <head> populated with Person / Organization / Article JSON-LD, agents can resolve entity references. If you don't, you are betting on the LLM's own retrieval. With explicit sameAs (linking GitHub, LinkedIn, Telegram, X, Upwork), you become one entity in the agent's mental model — instead of three handles with a similar name.

Why this matters more for founder-led brands

For studios, agencies, and freelancer-led SaaS, there is a second dimension I call entity authority — does the LLM recognise who you are as one person, or three different ones?

If you have:

  • GitHub your-name
  • LinkedIn your.name
  • Twitter your_name
  • Telegram yourname

…that's potentially four separate entities. Schema.org alternateName is a starting point, but the long-term fix is bio consistency (the same one-liner everywhere) + linkback discipline (every profile links back to your domain) + a steady drip of technical writing under your name.

This is a separate companion checklist — 13 criteria, 50 points, covering platform footprint, schema signals, consistency, and external mentions. Same open-source repo.

Try it on your site

The methodology, prompt, bash script, and example reports are MIT-licensed:

github.com/ivannikov-pro/ai-readiness-audit

Three ways to use it:

  1. Bash script for a 30-second automated subset:

    ./scripts/check-ai-readiness.sh https://yourcompany.com

  2. Claude / ChatGPT prompt — paste prompts/audit-prompt.md into your LLM with your URL.

  3. Manual checklist — work through audit-checklist.md for the full 26 criteria with point allocations.

If you want someone to actually do the audit, prioritise fixes, and track your AI Visibility Score over time — that's the paid service.

TL;DR

  • AI agents are increasingly how customers discover B2B SaaS.
  • Most popular B2B SaaS sites fail basic AI-readiness — SPAs that need JS, no llms.txt, no explicit AI policy.
  • The fix is a known, scoreable list of 26 criteria.
  • I shipped my own site from 42 to 80 in one day. The tools are open source.
  • First-mover window in most niches is open for the next 12–18 months.

If you found this useful: follow me on X, Telegram, or the channel. The full strategy and methodology will be documented as the methodology evolves.

— Aleksandr Ivannikov / IVANNIKOV.PRO

Is your Web3 project ready for launch?

Download my comprehensive 50-point checklist covering smart contract security, tokenomics, legal, and pre-launch marketing.