Using AI for research
Principles
- AI suggests; you verify — Treat every factual claim as untrusted until it matches a primary or high-quality secondary source.
- Ask for sources — Prefer tools that expose links (Perplexity) or your uploaded files (NotebookLM).
- Separate “ideas” from “facts” — Brainstorming and summarizing your own notes are lower risk than medical, legal, or financial claims.
- Save the trail — Bookmarks, quotes with URLs, and dates — future-you needs auditability.
Suggested workflow
| Step | What to do |
|---|---|
| 1. Frame the question | One sentence goal + audience (e.g. “Options for hosting a Next.js app for a solo dev”). |
| 2. Broad pass | Use search+citation tool or manual search; collect 5–10 candidate sources. |
| 3. Read primaries | Official docs, standards bodies, vendor status pages — not only blog summaries. |
| 4. Synthesize | Use AI to compress what you already read — paste excerpts, not blind trust. |
| 5. Check conflicts | If two sources disagree, note the disagreement; don’t average into false certainty. |
Prompt patterns that help
- “List assumptions and unknowns after you answer.”
- “Quote short excerpts I should verify; don’t invent citations.”
- “What would change your answer if it were wrong?”
Related
- What can go wrong — injection, data you should not paste
- Free chat AIs — when chat is enough vs when you need research-specific tools
Last reviewed: April 2026.
Last updated on