Google technical writer explains how to chunk content for AI retrieval, and it's mostly old SEO basics
gaganghotra_ · x · 2026-11-05
A Google technical writer detailed how to structure content for AI/agent retrieval pipelines: Agent Search caps chunks at 500 tokens (300-400 English words), and a 1,000-word unheaded article gets split into 3-4 context-less chunks.
Key guidelines:
- Use specific, descriptive headings ("Known limitations of offline caching" beats vague "Known limitations")
- Use real H2/H3 semantic tags — CSS-styled pseudo headings become plain paragraphs after Markdown conversion
- Keep sections self-contained and under the chunk limit, leaving headroom for page titles
Poster harpreetchatha notes this is mostly grade-school writing advice, and the new GEO/AI-optimization framing largely overlaps with what good SEO practitioners have done for years.
More from Apps
- Dev open-sources free AI skill that helps cancer patients and caregivers navigate medical decisions — Scobleizer · 2026-11-05
- Most new users can't figure out general-purpose chat agents — use cases are too broad — signulll · 2026-11-05
- DuckDuckGo courts schools with AI-free search amid backlash over Google's AI policy — TechCrunch AI · 2026-11-05
- OpenAI ad stars Messi using ChatGPT to reimagine his room — OpenAI · 2026-11-05
- Community fork ports QuixiDesk AI desktop app to Linux with 100+ patches, Arch and signed Flatpak builds — QuixiAI · 2026-11-05
- Lenny's Newsletter goes print: 8 AI-era career posts as a $29 magazine — lennysan · 2026-11-05