Google renamed NotebookLM to Gemini Notebook on July 16, 2026. It remains a standalone research product for asking questions about selected sources, reviewing inline citations, and creating research artifacts from that source set.
The practical use case is straightforward: collect eligible material in a notebook, ask a focused question, inspect the cited passages, and then decide whether the result is suitable for a brief, table, or project record. Gemini Notebook is useful when an answer should stay grounded in chosen sources, not when an AI system should conduct unattended research or publish findings without review.
This guide focuses on the operating details that matter in a real research workflow: source completeness, citation review, provenance, exports, plan limits, and a manual handoff to ClickUp. Google announced the name change in its Gemini Notebook announcement.
What NotebookLM is called now, and what it does
Use Gemini Notebook, formerly NotebookLM, when you need answers grounded in a defined collection of source material. You can create a notebook, add eligible files or links, select the sources relevant to a question, and ask for an answer with inline citations. The product can also generate reports, briefing documents, study guides, notes, mind maps, audio or video overviews, flashcards, quizzes, slide decks, infographics, and data tables. Access to particular outputs depends on the account, plan, region, language, and feature rollout.
Gemini Notebook answers are not the same as ordinary Gemini app answers. Google describes Gemini Notebook responses as grounded in the notebook sources, while Gemini app responses may use web search and other Gemini tools. Google now documents web-source discovery and Deep Research in some Gemini Notebook experiences, but availability varies. Do not describe standard notebook chat as unrestricted web search.
Inline citations help you locate the source material used for a response. They do not prove that the generated interpretation is correct, and they do not turn Gemini Notebook into a general APA, MLA, or Chicago citation manager. If a finding matters, open the cited passage and compare it with the wording of the generated claim.
Treat Gemini Notebook as a source-grounded research step, not as an approval step. The reviewer who checks the passage remains responsible for deciding whether a claim is ready for reuse.
How to use Gemini Notebook: a source-to-output workflow
Begin with a defined question and a manageable source set. For example, a team comparing two research studies could add accessible copies of both studies, select those sources, and ask: “Compare the reported methods and stated limitations in these sources. Give one row per study, distinguish the authors’ statements from your own inference, and cite each row.” The prompt identifies the task, scope, output structure, and evidence requirement.
After Gemini Notebook produces a response, inspect the citations rather than accepting a polished paragraph at face value. For a comparison, check every material row against the relevant passage. If the answer combines evidence from several documents, record which source supports which assertion. A useful output is often a reviewed data table or brief, not the original chat response.
Reports and data tables can be exported to Google Docs or Google Sheets. The export is a snapshot: edits made in the exported document or sheet do not synchronize back to the notebook. Google also documents citation information in a separate sheet tab when a data table is exported. Preserve the notebook reference, source references, and review date with the export.
Check whether a source will import completely
Gemini Notebook supports many source types, including PDFs, Google Docs, Slides and Sheets, Word and PowerPoint files, text, Markdown, CSV, images, audio, ePub, web URLs, public YouTube videos, notes, Play Books ebooks, and Gemini Chats. Actual availability can vary by account, plan, language, region, and product rollout. See Google’s source and import documentation for current conditions.
Import behavior is as important as file type. Web pages are generally imported as text, so images, embedded video, nested pages, and paywalled content may not be included. YouTube imports rely on transcripts from public videos with captions, so the complete audiovisual content is not automatically available as source evidence. Google Drive sources require access, and Google file comments and footnotes are not imported.
For local uploads, Google documents a limit of up to 500,000 words or 200 MB per source. That is a per-source limit, not the notebook-wide source allowance. Copy-protected PDFs may fail, and an image, chart, handwritten page, comment, or footnote may contain evidence that the imported representation does not preserve.
Use deterministic checks for known intake conditions: file size, supported format, accessible URL, Drive permission, public YouTube status, and caption availability. Do not ask an AI model to guess whether an inaccessible document imported completely. If the evidence depends on omitted content, obtain an accessible version or route the source for manual capture and review.
Turn generated answers into traceable research records
A team brief should not store a generated answer as one undifferentiated research record. Separate the record grains:
- A source record represents one imported or synchronized source.
- A claim record represents one editorial assertion.
- A citation record represents one source passage supporting one claim or response.
- A response or run record represents one prompt execution or generated artifact.
- An aggregate record represents a rollup such as claims reviewed per project or citations per day, not an individual citation.
The following is a hypothetical editorial schema, not a Google data model. It shows one claim record. Supporting passages should be stored as separate citation records, each with its own citation identifier and relationship to the claim.
{
"claim_id": "claim-project-01-0007",
"project_id": "project-01",
"claim_text": "Illustrative finding about a reported method",
"claim_type": "reported_evidence",
"source_ids": [
"source-study-a-version-2025-04-12"
],
"citation_ids": [
"citation-run-0042-claim-0007-01"
],
"source_published_date": "2025-04-12",
"source_retrieved_at": "2026-10-26T10:30:00Z",
"source_version_or_hash": "version-recorded-by-research-team",
"claim_status": "needs_review",
"reviewer": "Research editor",
"review_timestamp": null,
"notes_on_inference": "Confirm whether the wording reports the authors' statement or an editorial inference."
}Keep publication date separate from retrieval date. A source may be old but newly reviewed, or it may change after it was first imported. For a citation record, use a unique identifier tied to the claim, source, response or run, and citation location where available. Do not use a notebook title, date, or daily citation count as the unique key for individual observations.
If multiple workers can process the same source, a search-then-create check is not race-safe. Use a database-enforced unique constraint and a transactional upsert. A proposed external key for one claim-citation row could combine project_id, normalized source identifier, source version, claim identifier, and citation locator. These are implementation recommendations, not native Gemini Notebook fields.
Implementation patterns for a research workflow
The table below describes proposed operating patterns around documented product actions. The validation and record fields are editorial design choices, not built-in Google or ClickUp schemas.
| Trigger | AI job | Validation | Destination or fallback |
|---|---|---|---|
| File, URL, Drive item, or public video selected | Summarize or answer from eligible notebook sources | Check access, format, size, captions, and likely omissions | Notebook for accepted material; intake queue for partial or inaccessible sources |
| Question comparing selected sources | Draft cited claims or a structured data table | Check every material row against its cited passage | Reviewed brief or Google Sheet; return unsupported rows to the editor |
| Approved finding becomes project work | ClickUp Brain may summarize, rewrite, translate, or create action-oriented text from content already in ClickUp | Confirm reviewer, excerpt, source date, status, and permissions | User-mediated ClickUp Doc or task; hold the item if provenance is incomplete |
For the first pattern, the source owner handles missing permissions, paywalls, copy protection, or absent captions. For the second, the research editor handles ambiguous wording and distinguishes “the source reports” from “the evidence suggests.” For the third, the project owner handles destination permissions while the research reviewer resolves disputed provenance.
Limits, sharing, plans, and currentness
Each notebook is independent, so Gemini Notebook cannot retrieve across several notebooks at once. Google documents private sharing with Viewer or Editor access and public sharing. A personal Gmail account can share a notebook with up to 50 users according to the current documentation, while source and artifact access, user limits, and availability depend on account and plan. Public access requires the recipient to be signed in to a Google account, and shared access may expose underlying sources and artifacts. Check permissions before sharing sensitive or licensed material.
Google’s current plan documentation lists 50 sources per notebook for Standard, 100 for Plus, 300 for Pro, and up to 600 for Ultra. Usage limits are compute-based and may vary with prompt complexity, model, conversation length, and feature. Plan limits, feature access, and usage conditions can change, so confirm the current plan and source limits before procurement or rollout.
Google Drive sources can update when the original changes, while exported Docs and Sheets are snapshots. Retain the retrieval date and source version, and recheck a finding when the underlying Drive file changes. Do not assume that a later edit to an export updates the notebook or its citations.
A citation establishes a path back to source material. Before a finding enters a brief or project system, compare the claim with the cited passage, record whether it is direct evidence or inference, and confirm that the source is still current enough for the decision.
When to use another tool, and how to hand off to ClickUp
Choose the tool by the work to be done. Gemini Notebook fits questions and artifacts grounded in selected materials. A reference manager is better when the core job is maintaining a citation library, formal bibliography, or PDF annotation workflow. A conventional notes tool suits ongoing personal capture. ClickUp Brain fits AI assistance with content already in ClickUp Docs, tasks, comments, and Chat, including summarization, rewriting, translation, and action-item generation. ClickUp AI Fields can produce structured text or dropdown outputs from task context, subject to plan, role, and AI-credit conditions.
ClickUp’s Chrome Extension can capture webpages or text into ClickUp tasks or Docs. ClickUp Brain can work with content already placed in ClickUp, and AI Fields can create structured outputs from ClickUp context. These capabilities do not establish factual accuracy or an automatic research approval process.
The safe handoff is user-mediated: a researcher reviews the Gemini Notebook output, exports or copies the approved material, and then adds it to a ClickUp Doc or task through a supported action. A direct consumer Gemini Notebook-to-ClickUp connector and automatic citation transfer were not verified. Treat the handoff as an export and intake process, not a live integration.
Keep the minimum provenance with the ClickUp item: claim text, source URL or identifier, supporting excerpt, source retrieval date, review status, reviewer, and any note that the wording is an inference. If a required field is missing, leave the item in the research review queue instead of assigning downstream work as though the finding were approved.
A short launch check for a research workflow
- Confirm that each source is accessible and contains the evidence needed for the question, including relevant text, captions, images, comments, or footnotes where applicable.
- Name the person who opens cited passages and approves material claims.
- Store source identifiers, supporting excerpts, retrieval dates, and export dates with reviewed findings.
- Check current plan limits, sharing permissions, and feature access for everyone who needs the notebook or artifacts.
- Measure observable process outcomes such as time from source intake to reviewed brief, the share of material claims with checked citations, and the number of items returned for missing provenance.
Automate routing only after the team has defined acceptable sources, required provenance fields, and the person who approves a claim. Until then, keep partial imports, unsupported findings, and currentness exceptions with a named human reviewer.