How to Win Docs and Influence Robots: Improving Robot Accuracy Through Better Docs
summary
Details: tracks the same user prompts across multiple LLMs (ChatGPT Gemini Meta etc), builds internal tooling called Project Mercury, measures results across ~8 answer engines and reports 93% accuracy, uses CDN logs to see crawlers, tests 500 then 2,000 real user questions, fixes errors by making docs explicit (add “maximum value”), adds schema.org/semantic HTML, prunes from ~50k to 25,000 pages, and observes corrections return after about 3 weeks.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 3:10Robots' information-seeking shift
- 4:14Six things robots like in documentation
- 8:07Prioritize search data
- 9:31Less content can be more
- 10:35Project Mercury testing
- 13:28Fixing LLM inaccuracies
- 15:17Doc team prioritization
- 16:42Q&A
- 27:43Balancing control and robots
related talks
39mDocs AI Tooling is Better (and Better for Us) than You Think
09mThe Robots are coming for your job, and that's okay
47mCan You Trust Your ChatBot: Techniques for Testing LLM Responses
49mNext-Gen Documentation with AI
34mProving and improving the value of your docs
25mWinning the AEO/GEO "Answer War": Why Structured Data is the Only AI Moat That Matters
26mMastering Large Language Models: Prompt Engineering for Precision & Performance
43mA journey from static documentation to generative AI