Pattern 01
Compress orientation time
Reduce the hidden tax of reconstructing project background before useful work can begin.
Mantel · Raymond Group
01 · Read the number correctly
The headline is real. Its limits matter.
Google's public data establishes scale, not a statistically typical enterprise workflow.
Google says Gemini Notebook, formerly NotebookLM, serves more than 30 million people across over 600,000 organizations. That makes a strong headline. It does not tell us what percentage of those organizations use the product in HR, marketing, engineering, research, or operations.
Google Cloud separately maintains a much broader catalog of enterprise generative-AI deployments. By April 22, 2026, that catalog had expanded to 1,302 use cases across Google's AI portfolio. It is not a Gemini Notebook usage survey.
The useful question is not what every one of the 600,000 organizations is doing. We do not have that data. The useful question is what patterns recur in the deployments Google has documented publicly.
Evidence boundary: customer productivity figures in this guide are vendor-reported examples, not expected results for every organization. NLM Guide distinguishes official product documentation, reported customer outcomes, and editorial interpretation.
02 · The new unit of work
Notebookization is not converting documents into notebooks.
It is converting organizational knowledge into bounded, reusable AI context for a coherent job.
A conventional chatbot begins with a prompt. A useful enterprise notebook begins with a context boundary.
Instead of repeatedly explaining the product, policy, previous quarter, research history, and approved sources to a general assistant, a team assembles that context once and reuses it. The documents are not the innovation. The reusable context layer is.
Authoritative sources+One coherent job=Reusable AI context
A competitor becomes a notebook. A policy library becomes a notebook. A research project becomes a notebook. A customer's feedback history becomes a notebook. A team's project context becomes a notebook.
That is more consequential than “chat with your documents.” It changes the basic unit of AI work from a disposable prompt to a designed information environment.
03 · Six adoption patterns
Different departments are converging on the same architecture.
The visible use cases vary. The underlying move is consistent: build a bounded context once, then reuse it for a recurring knowledge task.
Pattern 02
Reuse expert context
Let specialists curate the evidence once, then reserve their time for ambiguity and judgment.
Taboola
Pattern 03
Make procedures interactive
Turn SOPs, standards, inspection reports, and checklists into a queryable operating layer.
Forbes Marshall
Pattern 04
Query unstructured evidence
Interrogate interviews, surveys, reviews, support messages, and qualitative feedback repeatedly.
Ocado Retail
Pattern 05
Package recurring management work
Build bounded contexts around repeated preparation tasks instead of starting from zero.
Docusign
Pattern 06
Centralize engineering context
Help engineers understand the system and its history before asking a coding tool to change it.
Shopify · Airwallex
Pattern 01 · Context compression
Companies are compressing the cost of getting up to speed.
Before someone can make a good decision, they often have to reconstruct the context around it: read the papers, find the old report, ask a senior employee what happened last time, locate the right spreadsheet, and determine which version is current.
Australian consultancy Mantel uses NotebookLM as a shared knowledge hub containing research papers and internal documents. The company reports that producing a roughly 30-page thought-leadership paper once took as long as five months. With NotebookLM in the process, that timeline can fall to around two months.
Up to 5 monthsPrevious production time
→About 2 monthsReported with NotebookLM
Raymond Group applies the same pattern to financial analysis. Its teams load peer-company disclosures into NotebookLM and ask targeted questions against the documents. Google reports that extracting relevant signals now takes minutes instead of hours.
The important gain is not simply that AI writes faster. It is context compression: less human time is spent reconstructing enough background knowledge to begin useful work.
Pattern 02 · Expert-curated context
Expert knowledge is becoming reusable instead of repeatedly explained.
Taboola's analysts curate important sources and context inside shared notebooks. Colleagues can ask questions, follow up, and inspect the underlying sources without requiring the analyst to answer every repeated question personally.
This creates a new organizational object: the expert-curated notebook. An expert defines the problem, selects authoritative material, adds useful context, shares the notebook, and lets other employees explore that knowledge independently.
It is not quite a wiki. A wiki asks someone to rewrite knowledge into pages. It is not quite a shared drive. A drive gives employees documents but still expects them to locate and interpret the right material. A notebook preserves the sources while adding a source-grounded interface over them.
The expert remains important. What changes is when the expert needs to be involved. Routine context transfer can happen through the notebook. Expert attention can be reserved for ambiguity, judgment, and genuinely new problems.
Patterns 03-04 · Interactive evidence
Static documents and unstructured evidence are becoming queryable.
Industrial engineering company Forbes Marshall uses NotebookLM across sales, service, manufacturing, engineering, IT, HR, and supply chain. Its documented workflows include comparing RFQs, quotations, inspection reports, SOPs, programs, and checklists against organizational standards.
Instead of opening a 90-page procedure and hunting for one requirement, an employee can ask which section applies. Instead of manually comparing a quotation against internal standards, the employee can ask where the proposal deviates.
Ocado Retail applies a related model to unstructured customer feedback. The company uses NotebookLM to summarize qualitative evidence and identify emerging trends and issues, estimating that the workflow saves analysts around two days of work per week.
Unstructured feedbackReviews, surveys, support messages
→Queryable evidenceRecurring patterns and changes
The higher-value role for AI may be upstream of content generation. Instead of asking AI mainly to produce the message, use it to understand the evidence that should determine the message.
Patterns 05-06 · Recurring work
Managers and engineers are packaging context around repeatable jobs.
A Docusign manager reported spending four to five hours per employee preparing quarterly performance reviews. With separate notebooks built around existing notes, that manager could query the record and prepare for the meeting in less than 30 minutes. Google reports this as roughly a 90% reduction in preparation time.
The pattern is portable: employee notes plus goals become a performance-review notebook; customer history becomes an account notebook; a policy set becomes an HR notebook; and launch material becomes a product notebook.
Engineering exposes the other side of the same problem. Modern teams rarely lack code. They lack unified context across repositories, architecture documents, incident reports, API documentation, design decisions, issue histories, and migration plans.
Shopify says employees use NotebookLM to collect project information in one place. Airwallex uses NotebookLM alongside Gemini on Vertex AI, although its published engineering productivity figures reflect the broader AI program rather than NotebookLM alone.
The boundary is useful: coding assistants are increasingly good at changing the system. Notebooks are especially useful for understanding the context surrounding the system.
04 · The threshold change
The notebook can now become infrastructure.
When an automation can query an existing notebook, curated context moves from a human-facing destination to a reusable service inside another workflow.
In May 2026, Google added NotebookLM integration to Google Workspace Studio. An automation can use an existing notebook as a knowledge source through an Ask NotebookLM step and retrieve a grounded response from the notebook's research and source material.
The architecture is no longer limited to documents, a notebook, and a human question. It can now extend into downstream work.
Trusted sources
Policies, reports, research, specifications, and other approved evidence.
Team notebook
A bounded context with a defined owner, scope, and refresh process.
AI workflow
A human or automation queries the notebook for grounded context.
Downstream output
A briefing, response, onboarding step, report, or other operational result.
A support workflow could query a product-knowledge notebook. An onboarding automation could query a policy notebook. A recurring briefing could query market intelligence. A sales workflow could retrieve product context before preparing a meeting brief.
Once a notebook feeds automation, source quality becomes infrastructure quality.
05 · The uncomfortable part
Grounding can turn curation bias into institutional certainty.
A cited answer can still be wrong when its source collection is incomplete, stale, contradictory, or politically selected.
Grounding feels safe. The answer cites company documents, and employees can inspect the evidence. That is clearly better than asking an unrestricted model to invent organizational knowledge from scratch.
But grounded does not mean correct. Imagine an HR notebook containing an outdated policy, an old benefits guide, three current documents, but not the latest legal clarification. The model may answer exactly as its sources imply. The citations will look reassuring. The answer can still be wrong.
This creates a subtler failure mode than hallucination: curation error.
Who decides what counts as trusted?
Who selects the documents? Who resolves conflicts? Who removes obsolete sources? Who owns the notebook when its creator leaves? Who reviews permissions? Who knows when the knowledge was last validated?
06 · Five responsibilities
Long-lived notebooks need architecture, not just better prompts.
The universal rule is one coherent job per notebook, supported by explicit ownership, source authority, maintenance, and decision capture.
Scope
Give the notebook one coherent job, not necessarily one narrow question and never “everything related to work.”
Ownership
Name a human responsible for sources, permissions, contradictions, refreshes, and retirement.
Authority
Distinguish approved policy from working context and external evidence instead of treating every source equally.
Refresh
Decide whether the corpus is frozen for reproducibility or maintained on a defined trigger and cadence.
Decision capture
Preserve important conclusions with date, owner, sources, confidence, and unresolved uncertainty.
The decision ledger
Gemini Notebook does not necessarily preserve the full path of every chat exchange. Sources can persist, and pinned or exported artifacts can persist, but important conclusions should not depend on an unsaved conversation.
For high-value notebooks, record: decision → supporting sources → date → owner → confidence → unresolved uncertainty. The notebook should not become institutional memory without an institutional record.
07 · Notebook Maturity Ladder
Not every notebook carries the same organizational risk.
Governance should increase as a notebook becomes more reusable, more shared, and more connected to downstream action.
Level 1
Disposable query
Upload material, ask a question, and leave. Useful, but little context is reused.
Level 2
Project notebook
The source set persists around one project, research problem, or decision.
Level 3
Reusable knowledge asset
The notebook supports recurring work, has a clear purpose, and preserves important outputs.
Level 4
Shared team context
Multiple employees rely on one curated notebook. Ownership, permissions, and freshness matter.
Level 5
Operational context infrastructure
The notebook feeds another AI workflow or automation. Errors can propagate without a human initiating every query.
08 · Measuring value
Count the workflow improvement, not the prompts.
Enterprise AI usage is easy to count. The better unit of value is the job that changed and what happened downstream.
| Workflow | Metric that matters | Evidence status |
|---|---|---|
| Performance reviews | Manager preparation time | Documented One Docusign manager reports 4–5 hours falling to under 30 minutes; Google’s case summary cites a 90% reduction. |
| Customer feedback | Analyst synthesis time | Documented Ocado reports roughly two days per week saved. |
| Long-form research | Time from research start to completed paper | Documented Mantel reports up to five months falling to about two. |
| Financial analysis | Time to extract signals from disclosures | Documented Raymond reports hours falling to minutes. |
| Shared knowledge | Repeated questions escalated to experts | Recommended metric. Taboola documents the workflow but not a public benchmark. |
| SOP support | Time to find and compare the right procedure | Recommended metric. Forbes Marshall documents the workflow but not a public benchmark. |
| Automated notebook workflows | Error rate, review rate, and downstream consequence | Recommended metric. Critical when notebook answers feed action. |
The last metric matters most at Level 5. Once a notebook feeds another workflow, asking whether the answer sounded accurate is not enough. Teams need to measure what happened because of that answer.
09 · A practical pilot
Start with one recurring knowledge bottleneck.
Do not begin by telling every employee to “use AI more.” Choose a bounded job with stable evidence and a measurable before state.
Choose the job
Select one recurring task, such as preparing sales meetings, reviewing an SOP, or synthesizing customer feedback.
Design the context
Define authoritative sources, scope, owner, permissions, refresh triggers, and the human review point.
Measure the change
Compare time, quality, escalation, error, and downstream outcomes before and after the notebook.
A good pilot has a bounded task, a stable source set, a clear owner, a measurable baseline, a human review point, and a defined refresh cycle. If the notebook does not measurably improve the job, collecting more prompts is unlikely to fix the underlying problem.
10 · Enterprise NLM
This article is the thesis page, not the entire category.
The enterprise cluster should grow around notebook design, governance, economics, departmental practice, and the notebook lifecycle.
Live · Start herePlanned guide Planned guide Planned guide Planned guide Planned guide
Enterprise Adoption Patterns
Why companies are turning organizational knowledge into bounded, reusable AI context.
Design a Team Gemini Notebook
Purpose, ownership, source policy, naming, permissions, maintenance, and output records.
Gemini Notebook Governance
Authority, freshness, versioning, access reviews, departing owners, and audit trails.
Notebook ROI
Published case metrics, measurement design, and scenario-based value estimation.
Department Playbooks
Marketing, HR, sales, IT, engineering, research, and strategy workflows.
Notebook Lifecycle
Create, validate, deploy, monitor, refresh, archive, and retire shared notebooks.
11 · Frequently Asked Questions
Enterprise Gemini Notebook, clarified.
Is NotebookLM now called Gemini Notebook?
Yes. Google renamed NotebookLM to Gemini Notebook on July 16, 2026. It remains a standalone product while becoming more integrated with the Gemini ecosystem.
Does Google show how all 600,000 organizations use it?
No public representative survey breaks the adoption number down by department or workflow. The use cases here come from customer stories and official product documentation.
What is notebookization?
NLM Guide uses notebookization to describe turning a bounded collection of organizational knowledge into reusable AI context for a coherent job.
Should every department build its own notebook?
No. A notebook makes sense when there is a recurring knowledge problem with a coherent source set and a responsible owner. Uncontrolled creation produces duplication, stale information, and conflicting answers.
Can Gemini Notebook be used in automated workflows?
Yes. Google Workspace Studio can use an existing notebook as a knowledge source through an Ask NotebookLM step.
Does Gemini Notebook preserve every chat exchange?
Do not assume that it does. Important conclusions should be pinned, exported, or recorded in a decision ledger with their supporting sources.
What is different about Gemini Notebook Enterprise?
The enterprise product adds Google Cloud-oriented administrative, sharing, security, and compliance controls. The exact controls available depend on configuration, region, and product documentation.
12 · Primary sources
Where the claims come from.
Product facts use official documentation. Customer outcomes retain their vendor-reported status. Strategic claims are NLM Guide's interpretation.
Sources re-checked October 11, 2026. Customer outcomes are vendor-reported and labeled that way; press and community commentary are not used as evidence for any figure on this page.
- 01Google: NotebookLM is now Gemini Notebook, July 16, 2026
- 02Google Cloud: 1,302 real-world generative AI use cases
- 03Google Workspace: Gemini Notebook for Business
- 04Google Workspace: Docusign customer story
- 05Google Workspace: Ocado Retail customer story
- 06Google Cloud: Mantel customer story
- 07Google Workspace: Raymond Group customer story
- 08Google Cloud: Taboola customer story
- 09Google Workspace: Forbes Marshall customer story
- 10Google Cloud: Shopify customer story
- 11Google Cloud: Airwallex customer story
- 12Google Workspace Updates: Use NotebookLM in Workspace Studio flows
- 13Google Cloud: Gemini Notebook Enterprise overview