You've probably had this experience: you know you took notes on something - you remember the conversation, the key insight, and some details, but you can't find it. This happened to me recently, and I was annoyed that I wasn't quickly able to use my firm-approved AI to search OneNote.

If you're using OneNote, you probably know by know that the search function is notoriously unreliable; it's tedious at best. Yet I still use OneNote because it is the easiest and quickest note-taking platform accessible on the firm computer. So I thought, there must be an AI solution to remediate the search issue. The solution isn't great; you have to export all your OneNote files to PDF and then upload them to your firm's AI platform.

But here's what I've realized: even if you move your notes into an AI platform like ChatGPT Enterprise, Legora, or Harvey, the quality of your search results depends almost entirely on how you took the notes in the first place.

AI search is powerful because it understands meaning, not just keywords. You can ask "what did the witness say about the timeline?" and it will find relevant content even if you never wrote the word "timeline." But AI search still needs context to work with. If your notes say "she denied involvement," the AI has no idea who "she" is, what she denied involvement in, or why it matters, unless you included clues.

The good news: you don't need to become a meticulous note-formatter. You just need to adopt a few habits during note-taking, then let AI do the heavy lifting on structure and summaries afterward.

Here's the workflow I've been testing.

The Storage Problem (And Why It Matters)

Before getting into note-taking habits, we need to address where your notes live. This might be the hardest part to get right.

The goal is a single storage location that serves all your AI tools without constant exporting and converting. If your firm uses iManage, I think that's the way to go. Both Harvey and Legora integrate directly with iManage, which means notes saved there become immediately searchable in those platforms without any extra steps.

My recommendation: take notes directly in Word, saved to iManage.

I know, I know. Word feels clunkier than OneNote for quick capture, and that is precisely why I switched to OneNote midway through my first year at the firm. But the trade-off is worth it:

  • Single storage location serves all three AI platforms
  • No export or conversion steps
  • Notes remain editable for ongoing additions
  • Firm compliance requirements automatically satisfied

If you really prefer OneNote for capture, you can make it work. The hybrid approach: take notes in OneNote during the session, then copy/paste into your AI tool for processing, and save the processed output as a Word doc to iManage. Your raw notes live in both places (OneNote original, plus preserved at the bottom of the iManage doc). It's more friction, but it keeps OneNote in the capture phase if that's important to your workflow.

Folder Structure

Keep it simple and consistent:

[Matter Name]/
├── Working Notes/
│   ├── Team Meetings/
│   ├── Client Calls/
│   ├── Meet and Confers/
│   ├── Hearings/
│   ├── Research/
│   └── Strategy/

File naming convention: YYYY-MM-DD_[Type]_[Brief Description].docx

Examples:

  • 2025-01-03_Depo_Jane-Smith-VP-Engineering.docx
  • 2025-01-05_Call_Expert-Witness-Prep.docx
  • 2025-01-06_DocReview_March-2023-Emails.docx

Five Note-Taking Habits That Make AI Search Work

These are the only things you need to do differently while taking notes. Everything else gets handled by AI afterward.

1. Start Every Session With a Context Block

Before taking any notes, spend 15 seconds writing:

Matter: [Full matter name and client/matter number]
Date: [Today's date in YYYY-MM-DD format]
Type: [Depo/Hearing/Call/Doc Review/Research/Strategy]
Re: [What this is about in plain English]

This gives AI the scaffolding it needs to categorize and connect your notes.

2. Use Full Names on First Reference

Instead of "Smith" or "the witness," write "Jane Smith (TechStart VP Engineering)."

Instead of "the email," write "March 15, 2023 email from Smith to Williams."

Instead of "the motion," write "Defendant's Motion to Compel (Dkt. 45)."

3. No Orphan Pronouns

This is important. "She said she wasn't involved" is not very helpful in isolation. "Smith said she wasn't involved in the migration decision" is more useful and searchable.

Instead of "He objected," write "Defense counsel objected."

Instead of "They produced it late," write "TechStart produced the Slack export late."

4. Capture Document References With Specificity

When referencing a document, include enough detail to find it later:

  • Bates number if available
  • Exhibit number if marked
  • Date + author + recipient if an email
  • Page or paragraph if citing specific content

5. Flag Your Own Uncertainty

If you're unsure about something, mark it explicitly:

  • [CHECK: Was this March or April?]
  • [CONFIRM: Smith's exact title]
  • [NEED: Find the email she referenced]

This prevents AI from treating your uncertainty as established fact when it summarizes.

What You Don't Need to Do

  • Write summaries (AI handles this)
  • Format with headers (AI handles this)
  • Organize by topic (AI handles this)
  • Write in complete sentences without typos (fragments are fine)
  • Avoid abbreviations (just be consistent within each note)

Post-Session Processing: Two Options

After you finish taking notes, you need to run them through AI to add structure and summaries. This is where the magic happens, but it's also where hallucination risk lives. The prompts below are designed to be conservative: AI is explicitly told to only use what's in your notes and to flag anything unclear rather than guess.

💡
If you take decent enough notes, you might be able to skip the AI optimization portion below and still have relatively searchable notes.

Option A: Near-Zero Friction (Batch Weekly)

Time per session: ~30 seconds

After each session, just save the Word doc to iManage in the correct folder. Done.

Once per week (or before you need to search), batch-process all your unprocessed notes: upload them to ChatGPT Enterprise, run the batch prompt below, review outputs, and paste the processed versions back into each document above the raw notes.

Pros: Minimal daily friction, batching is efficient

Cons: Notes aren't AI-optimized until you process them

Option B: Process Each Session (2 Minutes)

Time per session: ~2 minutes

Immediately after each session:

  1. Copy your entire note
  2. Paste into ChatGPT Enterprise with the single-note prompt below
  3. Review AI output (30 seconds)
  4. Paste AI output at the TOP of your Word doc
  5. Add a divider line and "RAW NOTES" header
  6. Your original notes remain below, untouched
  7. Save to iManage

Pros: Every note is immediately AI-optimized and searchable

Cons: Requires discipline after every session

The Prompts

Single-Note Processing Prompt

I'm going to give you raw notes from a legal matter. Your job is to structure and summarize these notes WITHOUT adding any information that isn't explicitly stated.

CRITICAL RULES:
1. ONLY include facts, statements, and observations that appear in my notes
2. If something is unclear or ambiguous, flag it with [UNCLEAR] rather than guessing
3. Preserve all specific names, dates, document references, and quotes exactly as written
4. If my notes contain uncertainty markers like [CHECK] or [CONFIRM], keep them
5. Do not infer what documents say beyond what my notes state
6. Do not add legal analysis unless my notes contain my own analysis

OUTPUT FORMAT:
---
**[Auto-generate a descriptive title based on content]**
Matter: [Extract from context block]
Date: [Extract from context block]
Type: [Extract from context block]

**Summary**
[2-4 sentences capturing the key substance and significance of this session]

**Key Points**
[Bullet points of the most important facts, testimony, or observations from the notes]

**Action Items / Follow-Up**
[Any tasks, questions to pursue, or next steps mentioned or implied by the notes]

**Document/Exhibit References**
[List any documents, exhibits, or evidence referenced, with identifying details]

**People Referenced**
[List of individuals mentioned with their roles/affiliations as stated in notes]
---

Here are my raw notes:

[PASTE NOTES HERE]

Batch Processing Prompt

I'm going to give you multiple sets of raw notes from a legal matter. Process each one separately using the format below.

CRITICAL RULES:
1. ONLY include facts, statements, and observations that appear in the notes
2. If something is unclear or ambiguous, flag it with [UNCLEAR] rather than guessing
3. Preserve all specific names, dates, document references, and quotes exactly as written
4. Keep any uncertainty markers like [CHECK] or [CONFIRM]
5. Do not infer content beyond what the notes state
6. Do not add legal analysis unless the notes contain the attorney's own analysis
7. Process each note completely before moving to the next
8. Clearly separate each processed note

OUTPUT FORMAT FOR EACH NOTE:
---
**[Auto-generate descriptive title]**
Matter: [Extract from context block]
Date: [Extract from context block]
Type: [Extract from context block]

**Summary**
[2-4 sentences on key substance and significance]

**Key Points**
[Bullet points of important facts/testimony/observations]

**Action Items / Follow-Up**
[Tasks or next steps from the notes]

**Document/Exhibit References**
[Documents/exhibits with identifying details]

**People Referenced**
[Individuals and their roles as stated]

---
[END OF NOTE]
===

Process the following notes:

[PASTE ALL NOTES, SEPARATED BY CLEAR MARKERS]

Why Raw Notes Stay at the Bottom

Every processed document should preserve your original notes at the bottom, clearly marked:

═══════════════════════════════════════════════════════════════
RAW NOTES (UNEDITED)
═══════════════════════════════════════════════════════════════

[Your original notes exactly as written]

This serves two purposes. First, it's your reference when the AI summary feels off and you want to verify what you actually wrote. Second, it's your protection against hallucination. If AI-generated content ever becomes an issue, you have the ground truth right there.

Accessing Your Notes Across Platforms

Harvey: Also integrates with iManage (and has Vault for document storage). Connect your iManage workspace if your firm has enabled the integration, or upload notes directly to a Harvey Vault project.

Legora: Integrates directly with iManage. Once your processed notes are saved, use Legora's Assistant to query across all notes in the matter, or use Tabular Review to extract patterns across multiple notes.

ChatGPT Enterprise: No direct iManage integration. Create a Project for each active matter and periodically upload your processed notes. Or upload relevant docs directly into a conversation when you need to query them.

Practical approach: Use Legora or Harvey for comprehensive matter-wide searches (since they connect to iManage). Use ChatGPT Projects for matters where you want conversational back-and-forth analysis.

Try It and Share What You Learn

I think this is significantly better that letting months worth of notes sit in OneNote without being able to quickly and easily search them.

But I'm sure there are improvements to make. If you try this (or a variation), I'd genuinely like to hear how it goes. What worked? What was too much friction? What would you change?

Whatever approach you land on, the core principle holds: if you want AI to find your notes, you need some system that makes them findable. The five habits above are a starting point. The prompts are freely available to copy and modify. Take a shot, adjust as needed, and let me know what you learn.