NotebookLM (Gemini Notebook): how to build an AI-supported company knowledge base
NotebookLM, operating under the name Gemini Notebook since July 2026, helps you work with a selected set of sources. It only delivers value with well-organised documentation and with answers checked against the source.
A new employee asks about the complaints procedure, and the answer is in one of three documents on the shared drive: two are out of date, and the third has “final version 2” in its name. An AI tool will quickly find an answer in them, but it will not decide which document is the one that applies.
A company knowledge base makes sense when it contains up-to-date material, has designated owners and lets people quickly reach a reliable answer. NotebookLM can make working with such a set of documents easier, but it will not put the company's procedures in order for you.
What NotebookLM and Gemini Notebook are
NotebookLM is a Google service for working with materials selected by the user. In July 2026 Google renamed it Gemini Notebook. According to the vendor's announcement, existing notebooks remain unchanged, and the service is available at the existing address and in the Gemini app.
After adding sources, you can ask questions, create summaries and use citations that lead to specific passages of the material. Basing answers on selected sources makes them easier to check, although the risk of missing context or misinterpretation remains.
The range of supported sources, features and limits depends on the plan and changes over time. Before deployment, check the current service documentation, rather than basing the project on figures from older guides.
What a business can use NotebookLM for
NotebookLM supports tasks in which an employee has to go through many related materials, for example:
- finding entries in instructions and procedures;
- preparing preliminary report summaries;
- comparing versions of project documentation;
- creating onboarding materials for new employees;
- organising questions and answers about a product or process.
Treat the answer as help in finding information. A person makes the decision, especially in legal, HR, financial, medical and safety matters.
How to organise documentation before deploying AI
An AI tool will not fix contradictory procedures and will not recognise by itself which of two documents is current. Before the pilot:
- Define the scope of the knowledge base and its audience.
- Remove duplicates and mark outdated versions.
- Appoint an owner for each area of documentation.
- Introduce a review date and rules for approving changes.
- Specify what data may be processed in the chosen service.
Only on such a foundation are answers generated from the sources consistently useful.
Data privacy in NotebookLM: personal account vs Google Workspace
Do not add invoices, contracts, candidates' CVs or other documents containing personal data or trade secrets without careful thought. First establish the legal basis and purpose of processing, limit the scope of data, check the terms of the chosen version of the service and configure access according to employees' roles.
The difference between accounts matters. According to Google's information on data in the service for Google Workspace accounts, uploaded materials, questions and answers are not reviewed by humans or used to train AI models. For personal accounts, data is not used for training unless the user submits feedback, in which case Google may review the entire conversation. A company should therefore use work accounts and supplement the vendor's statement with its own risk assessment, retention rules and GDPR obligations.
How to run a safe pilot of an AI knowledge base
Good piloting is small, measurable and reversible:
- Choose one process. Start, for example, with public product instructions or internal procedures without sensitive data.
- Prepare a controlled set. Give documents unambiguous names, versions and update dates.
- Define test questions. Include simple and ambiguous questions as well as ones that the sources cannot answer.
- Check the citations. The user should open the indicated fragment and confirm that the answer makes sense.
- Grant minimum permissions. Only people who need the data for work should have access.
- Measure the effect. Compare the time it takes to reach the correct answer, the number of errors and the effort needed to maintain the materials.
- Agree on a fallback procedure. Important information must also be available when the service is down.
Limitations of NotebookLM in business
When assessing the tool, take a few limitations into account:
- the answer may be incomplete or incorrect despite the references included;
- the quality of the result depends on the quality, timeliness and readability of the sources;
- tables, scans, recordings and unusual document layout may be read inaccurately;
- features, limits and licensing rules differ between plans;
- Sharing your notebook with the wrong person may reveal the information it contains.
Therefore, NotebookLM should not be the sole documentation repository or approval mechanism for decisions with significant impact.
NotebookLM or another knowledge base solution
The choice of tool depends on where the data is stored, the integrations required, the permissions model, audit capabilities and the cost of maintenance. Sometimes a well-organised company drive with search is enough. At other times, a document management system, a company knowledge portal or your own document-based assistant will be better, which we describe in our post on an AI assistant for business.
If the company already uses Google Workspace, NotebookLM is a sensible candidate for a pilot. A security and cost-effectiveness assessment still has to be carried out.
When an AI knowledge base works well
NotebookLM makes it easier to ask questions about a selected set of materials and points to the passages its answers are based on. It gives the most to a company that has well-organised documentation, controls permissions and requires employees to check the result against the source. We run AI tool pilots, data flow assessments and integrate these tools into business processes as part of our service AI and automation.
Sources
- Google, NotebookLM is now Gemini Notebook, 16.07.2026, https://blog.google/innovation-and-ai/products/gemini-notebook/notebooklm-gemini-notebook/ (accessed 02/10/2026)
- Google, Help: data and privacy in NotebookLM (Gemini Notebook), https://support.google.com/notebooklm/answer/16164461 (accessed 02/10/2026)
As of 2 October 2026.