We scored our own website 29/100 on AI-agent readiness. Here's how we fixed it in one afternoon.

Aireadify scored its own homepage 29 out of 100 for AI-agent readiness, an F grade, and then raised it to 83 out of 100 (an A) in a single afternoon. The team implemented three high-ROI fixes: serving clean markdown via content negotiation, adding an `llms.txt` index, and including JSON-LD structured data. These changes reduced the token cost for AI agents from up to 90,000 tokens to as few as 1,000 tokens per page.

In my last post https://dev.to/aireadify/i-scanned-106-chip-company-websites-to-see-if-ai-agents-can-read-them-average-grade-f-4b76 I scanned 106 semiconductor sites and the average score was 42/100 — an F. Then I ran the same scanner on our own homepage . 29/100. Also an F. So I spent one afternoon fixing it. We got to 83/100 A . Here are the three highest-ROI changes, ranked by effort. All of them are free and most take under an hour. The single biggest waste: when an AI agent requests Accept: text/markdown , most servers still dump a wall of HTML. A typical chip-company homepage costs an agent 40,000–90,000 tokens to parse. The same content as clean markdown is 1,000–2,500 tokens — a 95% reduction. If you use a modern framework Next.js, Astro, SvelteKit, etc. you can do this in one middleware block: if req.headers.get 'accept' || '' .includes 'text/markdown' { return new Response markdown, { headers: { 'Content-Type': 'text/markdown; charset=utf-8' } } ; } Or at the edge Cloudflare Workers / Vercel Edge : js export default { async fetch request { const url = new URL request.url ; if request.headers.get 'accept' ?.includes 'text/markdown' { const md = await getMarkdownForPath url.pathname ; return new Response md, { headers: { 'Content-Type': 'text/markdown' } } ; } return fetch request ; } }; Impact: ~25 point jump on our score. llms.txt 20 min llms.txt is a simple markdown index at /.well-known/llms.txt that tells an AI agent which pages matter and what they contain. Think of it as a robots.txt for language models. Aireadify AI-agent readiness scanner and scoring for B2B websites. Products - Scanner https://aireadify.ai/scan : Free 0–100 score for any URL, ~2s, no signup - Leaderboard https://aireadify.ai/leaderboard : 106 semiconductor sites ranked Content - Why AI-agent readiness matters for B2B https://aireadify.ai/blog/why-ai-readiness - Methodology: 20 signals we check https://aireadify.ai/blog/methodology That's it. No schema, no JSON, no XML. Just markdown links with descriptions. Impact: ~20 point jump. Most B2B sites have product catalogs that are invisible to AI agents because they're rendered client-side or buried in unstructured HTML. Add JSON-LD Product schema to each product page: { "@context": "https://schema.org", "@type": "Product", "name": "CT8000 3D Hall Sensor", "description": "±40 mT, I²C / SPI, AEC-Q100 Grade 1", "sku": "CT8000-WL-TR", "brand": { "@type": "Brand", "name": "YourCompany" } } And expose a simple MCP endpoint so agents can query it directly instead of scraping: // POST /mcp/search parts { "query": "3D hall sensor I2C automotive" } Impact: ~10 point jump. | Fix | Time | Point gain | |---|---|---| | Markdown content negotiation | 15 min | ~25 | | llms.txt | 20 min | ~20 | | JSON-LD structured data | 30 min | ~10 | | robots.txt + sitemap.xml | 10 min | ~5 | | Open Graph + Twitter Cards | 15 min | ~5 | Total | ~90 min | ~65 | We went from 29 → 83 in roughly that order. The last 17 points are diminishing returns — semantic HTML, dark-mode meta tags, RSS feeds — nice-to-haves. Buyers are researching inside ChatGPT, Claude, and Perplexity now. If your site costs an agent 90k tokens to parse, it gets deprioritized or hallucinated. If it serves clean markdown with an index, it gets cited. The score is a proxy for "how likely is an AI to recommend you?" Free, no signup, ~2 seconds: https://aireadify.ai/scan https://aireadify.ai/scan It checks the same 20 signals and gives you a per-category breakdown with exact fixes. If you want the full 106-company ranking, it's here: https://aireadify.ai/leaderboard https://aireadify.ai/leaderboard Disclosure: we built the scanner and do agent-readiness work. Happy to share methodology or argue about weights in the comments.