Test if AI can navigate
and use your docs
80% of developers use ChatGPT/Claude to learn APIs. DocWorks validates that AI agents can actually find answers in your documentation.
# Testing: "How do I authenticate?"
⚠️ Found auth docs with 60% confidence
→ Searched: /intro → /api-reference → /auth
→ Found: OAuth flow documented
→ Missing: API key generation, token refresh
→ Issue: Information scattered across 3 pages
→ Found: OAuth flow documented
→ Missing: API key generation, token refresh
→ Issue: Information scattered across 3 pages
npx docworks check
No signup. Just API keys.How it actually works
1.
AI gets your llms.txt
Lists all your documentation URLs in a standard format
docs/intro.md docs/auth.md docs/api/*.md
2.
AI navigates with web_search
Uses OpenAI's web tools to explore docs like a developer would
tools: [{ type: "web_search" }]
3.
Structured output with confidence
Returns JSON with path taken, confidence score, and gaps
{ confidence: 0.6, path: [...], missing: [...] }
Real validation example
Your config (docworks.yml)
source: https://docs.api.com journeys: authentication: - How do I get API keys? - How do I authenticate? - How do I refresh tokens? error_handling: - What are the error codes? - How do I retry requests? provider: openai model: gpt-4o-mini
What DocWorks finds
Journey: authentication
✅ How do I get API keys? (95%, 1 page)
⚠️ How do I authenticate? (60%, 3 pages)
❌ How do I refresh tokens? (not found)
Missing: Token refresh docs
Journey: error_handling
✅ What are the error codes? (100%, immediate)
✅ How do I retry requests? (90%, 1 page)
Overall: 70% complete, auth needs work
Test against multiple AI models
Different models navigate docs differently. Test them all in CI/CD to ensure broad compatibility.
Multi-model test results
GPT-4o
95% confidence- found everything
Claude 3 Opus
75% confidence- struggled with navigation
Gemini Pro
60% confidence- couldn't find auth
Insight: Your docs may be optimized for OpenAI. Consider restructuring for broader AI compatibility.