TL;DR
- Create one NotebookLM notebook per client and load every source you have on them, from their website to your last call notes.
- NotebookLM answers only from your sources, so every claim comes with a citation you can click back to.
- Brief it with constraints like a researcher, not a vague question, and it returns something you can actually use.
- Use the Studio to turn the chat into a one page brief, a slide, or an audio overview before the call.
I had a prospect call last Tuesday and twelve browser tabs open. Their homepage, three LinkedIn posts, an old proposal I’d sent a competitor of theirs, and a 40 page industry report I downloaded and never read.
Reading all of it properly would have taken the day I didn’t have. So I dumped every file into a NotebookLM notebook and let it read while I made coffee.
Build one notebook from their sources
A NotebookLM notebook is only as good as what you put in it. Start a new notebook and add everything you have on the client as a source. Paste their URL, upload the PDF of their last RFP, drop in the notes from your last call. NotebookLM now accepts docs, sheets, slides, and audio, so the pile of mismatched formats you’ve collected is fine as is.
The reason this beats a generic chatbot is grounding. NotebookLM answers only from the sources in that notebook. When it makes a claim, you can click through to the exact sentence it pulled from. (The old “read it later” pile, by the way, was really the “read it never” pile. This fixes that without admitting it.)
NotebookLM also ships featured notebooks from partners like The Economist that are useful for market context. I keep one open as background while I build the client-specific one, so I’m not walking in blind on their industry either.
Keep your sources clean. If you upload a competitor’s deck by mistake, the brief will quietly blend two companies. I learned that one the hard way, and had to walk it back on the call.
Ask it like you’re briefing a researcher
Once the notebook is built, the chat is where the work happens. The trick is to give it constraints the way you’d brief a junior analyst, not to ask “what should I know about this client.”
A flat instruction gets a flat answer. A constrained one gets something you can use. Tell it your role, the call’s goal, and what format you want back. For a fractional CFO client, I fed in their last two quarterly board decks and a CSV export from their CRM, then asked what story the numbers actually told versus the one the deck told.
It came back with three gaps their own leadership had buried in appendix slides. Those went straight into the proposal.
The citation trail is the part that makes this safe to use with real clients. I’m not presenting NotebookLM’s guesses as my own analysis. I’m presenting its read of their own documents back to them, which is a different and much stronger position to be in.
Turn the answers into deliverables
The chat is the thinking. The Studio is the output. NotebookLM 2.0 can spin the same sources into a doc, a slide deck, an infographic, or a short audio overview. For client work I usually generate a one page brief for myself and a single slide to open the call with.
The audio overview is the piece I underestimated. Hearing NotebookLM argue both sides of the client’s problem on while walking my dogs is better prep than skimming my notes a third time. (Yes, I listen to a podcast about my clients now. We all have our things.)
The Prompt
You are preparing me for a [type of call: discovery / proposal / QBR] with [client name].
Sources: the client's website, their last RFP, and notes from our previous call are loaded in this notebook.
Give me:
1. Three risks or gaps in their current setup, cited to specific sources
2. Two questions I should ask that they have not answered yet
3. A one paragraph positioning statement I can open the call with
Stay strictly within the sources. If something is not in them, say so rather than guessing.Take It Further
The packaged version of this is in the Client Research kit, with the notebook template and the full prompt set. Every workflow I document lands in The Dispatch Newsletter first, so that’s the place to catch the next one.