A Hacker News commenter said they use Google's NotebookLM to answer 'where does it say that?' questions about ISO technical standards, traffic codes and other documents. You upload the documents and ask questions, and the answers are tied to references in those documents.
Using only the documents in this notebook, answer this question: [YOUR QUESTION, E.G. WHAT DOES THE CODE SAY ABOUT PARKING NEAR A JUNCTION?]. Quote the exact wording that answers it, and give the section or clause number for each quote. If the documents don't answer the question, say so clearly rather than guessing, and point me to the closest related section.
Replace the [PLACEHOLDERS] with your own details before you send it.
How to try it
- Get official copies of the documents you need, such as a standard, a code of practice, a handbook or your tenancy agreement.
- Create a new notebook in NotebookLM (now also known as Gemini Notebook) and upload the documents.
- Ask a specific ‘where does it say that?’ question, using the prompt above.
- Click the references in the answer to read the original passage in context.
- Note the section numbers, so you can quote them in an email, a report or a conversation.
Why this works
Long rulebooks are hard to search: the word you’d use is rarely the word the document uses, and the answer may be split across several sections. NotebookLM answers from the documents you give it and links its answers to the passages it used, so you can check them for yourself. That makes it useful for settling a question at work or at home with the actual wording, rather than someone’s memory of it, which is how the commenter used it for technical standards and traffic codes. Our guide to studying with Gemini and notebooks covers how to set one up.
Tips to get a better result
- Ask one precise question at a time; ‘what’s the minimum distance?’ works better than ‘tell me about distances’.
- Upload the latest official version, as rules and standards are updated.
- For contracts and small print, our guide to understanding terms and policies with AI has more tips.
- Ask it to list every section that mentions your topic, so nothing is missed.
Common mistakes
- Relying on the answer without reading the quoted passage yourself.
- Uploading an out-of-date copy and quoting rules that have since changed.
- Treating its reading as legal or professional advice; for anything important, check with a qualified person.
Try a variation
Here’s a variation of our own to try next:
Using the documents in this notebook, make a one-page checklist of the rules that apply to [SITUATION, E.G. HOSTING A STREET PARTY], with the section number next to each item.
Related
- AI for work and business: our hub of ideas
- Study with Gemini and notebooks
- Understand official letters with AI
More work use cases
Catch up on work after a holiday with Copilot
A project manager working across Japan and the US used to lose a whole day catching up after holidays. Now he uses Copilot to sort email subjects by urgency, draft holding replies, summarise the decisions made while he was away and turn meeting transcripts into action lists, telling it to use only the information he provides.
I've just come back from [NUMBER] days off. Here are the subjects (or a summary) of the emails I missed: [PASTE]. Sort them into: needs my reply today, needs a decision from me, and just for information. Draft short holding replies for the urgent ones, and summarise the key decisions made while I was away in three bullet points. Use only the information I've given you and don't add anything.
Run a realistic mock job interview with Claude
Instead of her partner awkwardly playing interviewer, one jobseeker gave Claude the job description, a link to the company's website and her CV, then did a mock interview. She went in far better prepared and got the offer.
You're interviewing me for this role. Job description: [PASTE]. Company website: [LINK OR ABOUT TEXT]. My CV: [PASTE]. Ask me one question at a time, mixing competency, role-specific and tricky questions. After each answer, give brief feedback and a stronger example answer using my real experience. Finish with my three biggest areas to improve.
Rehearse asking a client for a higher rate with ChatGPT
A freelance content strategist who'd undercharged a long-term client for 14 months spent three evenings role-playing the conversation with ChatGPT, asking it to play a friendly but budget-conscious client and then point out where they caved. They say they came away with a 15% rate increase.
I want to practise asking [CLIENT OR MANAGER] for a higher rate. I currently earn [CURRENT RATE] for [WORK], the market rate is about [RANGE] and I want to ask for [TARGET]. Play someone friendly but careful with money who pushes back with lines like 'the budget is tight right now'. Stay in role until I say 'stop', then tell me where I caved, where I was strong and what to change next time.
Reference: Hacker News (kazinator)