Tool

Mastering AI Research

A general chat tool answers from memory, which is exactly when it makes things up. Research mode flips the default: AI may only answer from sources you trust, and it cites every claim. NotebookLM is built for it.

1. What this does for you

NotebookLM is a free Google tool that turns a stack of your own documents into an expert you can question. You upload your sources, agreements, reports, notes, PDFs, even links, and every answer it gives is grounded in what you loaded and footnoted back to the exact page. No invented facts, because it is not allowed to reach outside your sources. Use the grounding prompt below to keep it honest, then click any citation to read the passage it came from.

2. What to type into your AI

Copy the box below. Replace the words in [BRACKETS] with your own info. Then paste it into Claude or ChatGPT.

You are my research analyst. Use ONLY the sources I have loaded in this notebook. Do not add outside facts or general knowledge.

QUESTION:
"""
[ASK YOUR REAL QUESTION IN ONE SENTENCE]
"""

Answer in three parts:
1. THE SHORT ANSWER, in plain language.
2. THE EVIDENCE: each point with a citation to the exact source it came from.
3. THE GAPS: what these sources do NOT tell me, so I know what is still unanswered.

If the sources disagree, show me both sides instead of picking one. If you cannot answer from the sources, say so rather than guessing.

Try it now in Claude →

See a full worked example ▾

Questioning a grant agreement, template, and tracker at once

You are my research analyst. Use ONLY the sources I have loaded in this notebook. Do not add outside facts.

QUESTION:
"""
What does our funder require in the final grant report, and which of those do we not yet have data for?
"""

(Sources loaded: the signed grant agreement PDF, the funder's reporting template, and our internal Q1-Q3 tracker.)

Answer in three parts:
1. THE SHORT ANSWER, in plain language.
2. THE EVIDENCE: each point cited to the exact source.
3. THE GAPS: what these sources do NOT tell me.

If the sources disagree, show both sides. If you cannot answer from the sources, say so.

A special-ed lead checking six IEPs against the district handbook

You are my research analyst. Use ONLY the sources I have loaded in this notebook. Do not add outside facts.

(Sources loaded: 6 student IEPs, the district special-education handbook, the current 504 plan template)

QUESTION:
"""
Which of these six IEPs are missing the transition-plan section required for a student who turns 16 during this school year?
"""

Answer in three parts:
1. THE SHORT ANSWER, in plain language.
2. THE EVIDENCE: each point cited to the exact IEP and page.
3. THE GAPS: what these sources do NOT tell me.

If the sources disagree, show both sides. If you cannot answer from the sources, say so.

A city administrator reconciling public comments against a draft ordinance

You are my research analyst. Use ONLY the sources I have loaded in this notebook. Do not add outside facts.

(Sources loaded: draft residential parking ordinance, 41 public-comment submissions, staff report to council)

QUESTION:
"""
Do the public comments raise any concern that the current draft ordinance does not already address in its text?
"""

Answer in three parts:
1. THE SHORT ANSWER, in plain language.
2. THE EVIDENCE: each point cited to the exact comment or ordinance section.
3. THE GAPS: what these sources do NOT tell me.

If the sources disagree, show both sides. If you cannot answer from the sources, say so.

A GC checking a framing subcontract against the structural drawings

You are my research analyst. Use ONLY the sources I have loaded in this notebook. Do not add outside facts.

(Sources loaded: framing subcontract, structural drawings S-100 through S-140, RFI 042 response from the Architect)

QUESTION:
"""
Does the framing Sub's scope of work match what the structural drawings and the latest RFI response actually require?
"""

Answer in three parts:
1. THE SHORT ANSWER, in plain language.
2. THE EVIDENCE: each point cited to the exact drawing sheet, subcontract section, or RFI.
3. THE GAPS: what these sources do NOT tell me.

If the sources disagree, show both sides. If you cannot answer from the sources, say so.

An owner-operator finding out who pays for the rooftop HVAC under her lease

You are my research analyst. Use ONLY the sources I have loaded in this notebook. Do not add outside facts.

(Sources loaded: signed commercial lease, first lease amendment, most recent CAM reconciliation statement)

QUESTION:
"""
Under my lease, who pays for a full HVAC replacement if the rooftop unit fails this winter, me or the landlord?
"""

Answer in three parts:
1. THE SHORT ANSWER, in plain language.
2. THE EVIDENCE: each point cited to the exact lease section or amendment.
3. THE GAPS: what these sources do NOT tell me.

If the sources disagree, show both sides. If you cannot answer from the sources, say so.

A development lead matching her tracker to a funder's reporting rules

You are my research analyst. Use ONLY the sources I have loaded in this notebook. Do not add outside facts.

(Sources loaded: signed grant agreement, funder reporting template, current program tracker spreadsheet)

QUESTION:
"""
Which outcome metrics the funder requires in the report are not yet being captured in my current program tracker?
"""

Answer in three parts:
1. THE SHORT ANSWER, in plain language.
2. THE EVIDENCE: each point cited to the exact grant section, template field, or tracker column.
3. THE GAPS: what these sources do NOT tell me.

If the sources disagree, show both sides. If you cannot answer from the sources, say so.

A bookkeeper checking a partnership's profit split against its returns

You are my research analyst. Use ONLY the sources I have loaded in this notebook. Do not add outside facts.

(Sources loaded: 2023 and 2024 Form 1065 with K-1s, the partnership agreement, IRS Publication 541)

QUESTION:
"""
Does the profit and loss split written in the partnership agreement match how the last two returns actually allocated income to each partner?
"""

Answer in three parts:
1. THE SHORT ANSWER, in plain language.
2. THE EVIDENCE: each point cited to the exact K-1, agreement section, or Pub 541 page.
3. THE GAPS: what these sources do NOT tell me.

If the sources disagree, show both sides. If you cannot answer from the sources, say so.

A working artist checking who owns reproduction rights in a consignment deal

You are my research analyst. Use ONLY the sources I have loaded in this notebook. Do not add outside facts.

(Sources loaded: gallery consignment agreement, my standard artist terms sheet, the email thread confirming the show dates)

QUESTION:
"""
Under this consignment agreement, who holds the reproduction and image rights to my pieces if the gallery closes or the show ends early?
"""

Answer in three parts:
1. THE SHORT ANSWER, in plain language.
2. THE EVIDENCE: each point cited to the exact agreement clause, terms sheet line, or email.
3. THE GAPS: what these sources do NOT tell me.

If the sources disagree, show both sides. If you cannot answer from the sources, say so.

3. What to do with the answer

Go to notebooklm.google.com and create a new notebook. It is free with a Google account.
Upload your sources: the documents that actually hold the answer. PDFs, Google Docs, pasted text, or links all work.
Paste the grounding prompt above and ask your real question in one sentence.
Read the answer, then click the citations. If a claim has no citation, treat it as unproven.
Use the built-in buttons to turn your sources into a briefing doc, an FAQ, a timeline, or an audio overview you can listen to on the go.
Reach for NotebookLM when citations matter. Reach for Claude or Gemini when you want open drafting or brainstorming instead.

Why this works

The thing that makes AI risky for research, that it will confidently fill gaps with invented facts, is switched off when you ground it in sources. NotebookLM only knows what you give it, and it always shows its work. That makes it trustworthy for the exact tasks where a made-up number would cost you: reports, compliance, anything with a citation attached.

Section Recap

Core Principle

AI invents facts when it has none. Give it trusted sources, force it to cite them, and research stops being a fact-checking chore.

Best-Fit Example

A program manager has to write a funder report and dreads cross-referencing the agreement, the template, and three quarters of tracker data. She loads all of it into a NotebookLM notebook and asks what the funder requires and where the data gaps are. In two minutes she has a cited checklist and a clear list of the two metrics she still has to gather.

A resource teacher drops six real IEPs, the district handbook, and the 504 template into one notebook, then asks the compliance question she would otherwise chase across six PDFs by hand. NotebookLM answers in plain language, points to the exact IEP and page for each finding, and flags that two plans have no transition section at all. Just as useful, it lists the gap: the handbook copy loaded does not state the exact age-16 deadline, so she still confirms that against state rule. She walks into the meeting with a cited checklist instead of a hunch, and the AI never invented a fact because it could only read what she uploaded. She stays the editor.

A public-sector staffer has 41 comment submissions and one draft ordinance and needs a neutral read before the council packet goes out. She loads all three documents and asks whether residents raised anything the draft does not already cover. NotebookLM groups the concerns, cites each one to the specific comment number and the ordinance section that does or does not respond, and shows both sides where residents disagree. It also names the gap: the submissions do not include the traffic-count data that would settle the loading-zone question, so that stays a human call. The result is a defensible, non-partisan summary grounded only in the record, not in the model's opinion.

Before the Sub mobilizes, the GC loads the framing subcontract, the S-series structural sheets, and the Architect's RFI 042 response into one notebook and asks whether the scope actually matches the drawings. NotebookLM finds that the drawings call for a heavier header the subcontract never priced, and cites the exact sheet and subcontract line for each point. It shows the conflict plainly: the RFI changed the connection detail, but the subcontract still references the old one. It also flags the gap, that the geotech report is not loaded, so bearing questions stay open. The GC catches a scope gap before it becomes a change order dispute, and the AI stayed inside the contract documents.

A shop owner loads her signed lease, the amendment, and the latest CAM statement, then asks the one question that could cost her thousands if the rooftop unit dies. NotebookLM gives the short answer in plain English, cites the maintenance clause and the amendment that narrowed it, and shows where the two documents seem to conflict on capital repairs versus routine upkeep. It also names the gap: neither document defines the unit's remaining useful life, which is the exact thing that decides repair versus replacement. She now knows what to ask the landlord and her attorney, instead of guessing, and every answer is tied to a page she can point to.

Three months before the report is due, a program lead loads the grant agreement, the funder's reporting template, and her own tracker, then asks what she is not yet collecting. NotebookLM lines the required metrics up against her columns, cites each required field to the grant section and the template, and shows exactly which two outcomes have no data source at all. It flags the gap where the agreement and the template use different definitions of the same metric, so she resolves that with her program officer rather than guessing. She fixes the tracker now instead of scrambling at deadline, and the finding is grounded only in the funder's own documents.

Onboarding a new partnership client, the bookkeeper loads two years of 1065s and K-1s, the partnership agreement, and IRS Pub 541, then asks whether the agreed split matches what was actually filed. NotebookLM shows the short answer, cites each partner's percentage to the exact K-1 line and agreement section, and surfaces a year where the allocation drifted from the written terms. It names the gap plainly: the sources do not explain why the 2024 numbers changed, and no special-allocation language is in the documents loaded. This is general guidance for the review, not filed advice, but it tells her exactly what to raise with the client and the CPA before anything is signed.

Before signing, an artist loads the gallery's consignment agreement, her own standard terms sheet, and the email thread setting the dates, then asks who controls reproduction rights if the gallery folds. NotebookLM answers in plain language, cites the rights clause and the return-of-work clause, and shows where the gallery's contract and her own terms sheet flatly disagree on image use. It flags the real gap: neither document says what happens to promotional images already posted online, which is the thing that bites artists later. She negotiates that clause on purpose instead of discovering it after a closing, and every claim points back to a line she can read for herself.

Key Takeaways

NotebookLM answers only from sources you load and cites every claim to the page.
A sharp one-sentence question is half the research.
Click the citations; a claim with no source is unproven until you verify it.
Use NotebookLM when citations matter, a chat tool when you want open drafting.

← All Tools