Research Note
AI Evidence Mode Decision Framework
Choose the evidence boundary before choosing a product.
AI Evidence Mode Decision Framework
Decision principle
Choose the evidence boundary before choosing a product.
Modes
| Job | Evidence need | Suitable starting mode |
|---|---|---|
| Explain assigned readings | Fixed, inspectable corpus | Source-grounded notebook |
| Find current developments | Open and time-sensitive discovery | Search with direct source review |
| Query authoritative records | Exact controlled data | Official database or system of record |
| Explore possibilities | Broad generation | General chat with explicit verification |
| Make a consequential decision | Authority, completeness, expertise, accountability | Qualified human process |
Decision questions
Is the approved evidence set already known?
Could new or missing information materially change the answer?
Does the work require creative generation, current discovery, exact records, or qualified judgment?
What data may enter the tool, and which account and terms control it?
Who will verify the result, and what source or test can reject it?
Product boundary
NotebookLM currently describes its chat as grounded in selected notebook sources. Current Notebooks in Gemini documentation says Gemini can also use web search and other tools. Those are different evidence paths even when a notebook appears in both products.
Named comparisons require same-day testing and current documentation. Product names illustrate modes; they do not define them permanently.
Failure boundary
A closed corpus can be authoritative but outdated or incomplete. Open research can be current but noisy and hard to reproduce. General chat can be useful for framing while producing unsupported detail. A database can be exact within its schema and still omit context.
Hybrid workflows are often appropriate, but every transition needs a visible source and review boundary.
Sources
Follow the evidence.
- support.google.com: 16322204support.google.com
- support.google.com: 17003757support.google.com
- support.google.com: 17004255support.google.com
- blog.google: notebooklm new features december 2024blog.google
- studentprivacy.ed.gov: privacy and education technologystudentprivacy.ed.gov
- support.google.com: 16212820support.google.com
- NIST AI Risk Management Frameworknist.gov
- www2.ed.gov: ai reportwww2.ed.gov
- daltonanderson.net: googles ai tutor the future of personalized learningdaltonanderson.net
- support.google.com: notebooklmsupport.google.com
- support.google.com: 16213268support.google.com
- journals.sagepub.com: fulljournals.sagepub.com
- youtu.be: 6BwWKkZ7aeAyoutu.be
- support.google.com: 16179559support.google.com
- NIST Generative AI Profilenvlpubs.nist.gov
- blog.google: notebooklm audio overviewsblog.google
- support.google.com: 16164461support.google.com
- open.spotify.com: 1YXy6yyC4u2mnqeARVB5qvopen.spotify.com
- edu.google.com: ai notebooklmedu.google.com
- support.google.com: 16215270support.google.com
- daltonanderson.ghost.io: googles ai tutor the future of personalized learningdaltonanderson.ghost.io