Build a Website AI Agents Love
That Scores 75+ on AX
Based on what the highest-AX sites do differently (Netlify, Stripe, Anthropic docs), here's a practical guide to making ChatGPT, Claude, and Perplexity read, understand, and recommend your site.
On this page
How the AX Score Works
Your AX (Agent Experience) score is a 0–100 measure of how well AI agents can read, understand, and act on your site. It is not a guess: AXray first probes your live site for the concrete signals agents rely on, then has a panel of frontier models score what they actually see.
We fetch llms.txt, robots.txt, sitemap.xml, JSON-LD, AGENTS.md, and test content negotiation — real signals, not opinions.
Multiple frontier models each score your site against those measured signals, from their own perspective as an agent.
The council is aggregated into one AX score plus an Agent Net Promoter Score — would an agent recommend you?
A score of 75+ means agents can find your key facts, trust them, and take the next step without guessing. Below ~40, agents skip you or hallucinate details.
AX Is the Technical Layer of AI SEO
If you're investing in AI SEO (also called GEO, Generative Engine Optimization, or AEO, Answer Engine Optimization), visibility tools tell you if ChatGPT, Perplexity, and Google AI Overviews cite you. Your AX score tells you why not — and what to fix.
Answer engines can only cite what agents can parse. Sites with a clean llms.txt, valid JSON-LD, and an open robots policy are far easier for answer engines to cite accurately — and far less likely to get their pricing or features hallucinated. Every prompt on this page is, in practice, a technical AI SEO fix: raising your AX score raises the ceiling on your AI visibility.
The difference in ambition: AI SEO stops at being cited. AX goes one step further — making your site something agents can act on (compare, sign up, book, buy) on a user's behalf.
The AX Council
One model's opinion is noisy. AXray runs an AX Council: several independent frontier models evaluate your site in parallel, then a head-of-council model reconciles them into a single, defensible score. You see each model's take and where they agree or disagree.
Copy-Paste Prompts for Your AI Coding Agent
The fastest way to raise your AX score: paste these into Claude Code, Cursor, Codex, or any AI coding agent inside your project. Each one targets a specific signal the AX Council measures. Start with the master prompt, then run the individual ones for anything still missing.
Run your site through AXray first, open the recommendations, and paste the specific gaps into the master prompt. Then re-run the evaluation to watch the score move.
Master AX Upgrade
One prompt that audits everything and fixes the whole site for agents. Start here.
You are an Agent Experience (AX) engineer. Audit this website/repository and make every change needed so AI agents (ChatGPT, Claude, Perplexity) can read, understand, and recommend it. Target an AX score of 75+.
Do all of the following, creating files where they are missing:
1. Add /llms.txt at the site root (llmstxt.org format): what the product does, who it is for, pricing, and links to the key pages and docs.
2. Add /AGENTS.md at the repo root: how an agent should navigate the product, the primary action and the exact URL/endpoint to complete it, and any API.
3. Add schema.org JSON-LD to every page: Organization, Product or SoftwareApplication, FAQPage, and BreadcrumbList where relevant.
4. Update robots.txt to allow reputable AI crawlers (GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended) and link to sitemap.xml.
5. Generate a complete sitemap.xml.
6. Use semantic HTML: exactly one <h1> stating the value proposition, logical <h2>/<h3>, <nav>, <main>, <article>, and alt text on all images.
7. Add complete meta tags: title, description, canonical, Open Graph, and Twitter Card.
For each change, show me the diff and explain in one line how it raises the AX score.Add llms.txt
The single highest-leverage file: a plain-text brief agents read to understand you.
Create an /llms.txt file at my site root following the llmstxt.org spec. Include: a top-level "# " H1 with the product name, a "> " blockquote one-line summary, a short factual description of what we do and who it is for, then "## " sections (Docs, Pricing, API, Guides) each with markdown links and a 1-line description per link. Keep it concise so an LLM can answer questions about us from this file alone.Add JSON-LD Structured Data
Machine-readable facts (schema.org) so agents quote your pricing and details correctly.
Add schema.org JSON-LD structured data to my site. On the homepage add Organization + SoftwareApplication (or Product) with name, description, url, logo, offers/pricing, and aggregateRating if we have reviews. On the FAQ add FAQPage. On docs add BreadcrumbList. Inject each as a <script type="application/ld+json"> in the page <head>, and make sure all required fields validate against schema.org.Add AGENTS.md
Tells an agent exactly how to complete the primary action on your product.
Create an AGENTS.md file describing this project for AI agents. Include: what the product does in 2 sentences, the primary user action and the exact URL/endpoint to complete it, the key pages and what each is for, any public API with an example request, and the rules an agent should follow (auth, rate limits). Write it so an agent could complete the main task using only this file.Open robots.txt to AI Crawlers
If GPTBot / ClaudeBot / PerplexityBot are blocked, agents never see you at all.
Update my robots.txt to explicitly allow the major AI crawlers so agents can discover and cite us: GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-Web, PerplexityBot, and Google-Extended. Keep sensitive paths (admin, checkout, account) disallowed, and add a "Sitemap:" line pointing to my sitemap.xml.Serve Clean Machine-Readable Content
Give agents a Markdown/plain-text version of key pages, stripped of boilerplate.
Make my key pages available in a clean, machine-readable form. Add a plain-text or Markdown version of each important page (for example respond to "Accept: text/markdown", or publish /page.md next to /page). Strip navigation, ads, and boilerplate so an agent receives just the substance of the page.Create a /for-agents Page
A dedicated, structured landing surface built for agents rather than humans.
Create a /for-agents page (and link it from the footer) that gives AI agents a structured, unambiguous summary of the product: what we do, who it is for, current pricing as a JSON block, the primary action with its URL, our public API endpoints, and links to llms.txt and AGENTS.md. Use clean semantic HTML plus JSON-LD so both crawlers and LLMs can parse it.What the Council Measures
These are the measured signals that move your AX score, roughly in order of impact:
Discoverability & Access
- • robots.txt allows AI crawlers, links sitemap
- • sitemap.xml is present and complete
- • llms.txt gives a concise product brief
- • Content negotiation / clean text versions
Structure & Meaning
- • JSON-LD (Organization, Product, FAQ)
- • Semantic HTML — one H1, clear headings
- • Meta tags — title, description, OG, canonical
- • Alt text so agents parse images
Agent Instructions
- • AGENTS.md with the primary action
- • /for-agents surface
- • Documented API / endpoints
Clarity & Trust
- • Unambiguous value proposition
- • Transparent pricing agents can quote
- • Consistent facts across page, JSON-LD, and llms.txt
AX Score Checklist (Target: 75+)
Files agents look for
- /llms.txt with a clear product brief
- /AGENTS.md with the primary action + endpoints
- robots.txt allowing GPTBot, ClaudeBot, PerplexityBot
- Complete sitemap.xml
- Optional /for-agents page
On every page
- schema.org JSON-LD (Organization, Product, FAQPage)
- One H1 stating the value prop, logical H2/H3
- Title, description, Open Graph, canonical tags
- Alt text on all images
- Pricing and key facts in machine-readable text
What Hurts Your AX Score
- Blocking AI crawlers in robots.txt — agents never see you
- Content only in JavaScript with no server-rendered text or clean fallback
- No structured data — agents guess (and get pricing/facts wrong)
- Vague value proposition — "transform your business" vs. a specific outcome
- Hidden pricing — "contact us" instead of quotable numbers
- Inconsistent facts across the page, JSON-LD, and llms.txt
See your AX score first
Run your site through the AX Council, then paste the exact gaps into the prompts above. Re-run to watch the score climb.