Answers only from your documents, with a source beside each one: a research assistant set up with NotebookLM


About this service
I set up a research assistant in NotebookLM with your PDFs, reports, policies and proposals. It answers only from those documents, with a source beside every claim. Ask ChatGPT about your office's leave policy or a donor's guidelines and it gives a confident answer. There is no way to tell whether that came from your document or was made up, and in a report or a client's paper, one made-up fact is danger enough. Yet the real answer is already with you, in a mountain of files: twenty PDFs, three versions of an SOP, last year's proposals. When someone new joins, they ask a senior, and the senior's day goes on giving the same answers again and again. ◆ What I do Google's NotebookLM is an AI that answers only from the sources you give it, and says so when the answer is not in the sources. We select your files, separate duplicates and old versions, arrange them into notebooks by topic, write the instructions, then test with your real questions, opening each answer's citation to check it. Each notebook also gets a source-bound briefing: 7 key facts with citations, where the sources agree and where they do not, which questions the sources do not answer, and a 10-line summary. The method comes from chapter 10 of Sabbir Ahmed's book "How I Use AI". ◆ Three packages One: one notebook on one topic, up to 20 sources. Three: three notebooks on three topics, a source map of which file is where, and sharing for up to 5 people. Five: five notebooks for a team, a list of where two documents say different things, a question bank of 50 questions for new staff, and an hour of training. ◆ Who it is for NGOs and project teams who have to keep going through donor guidelines and old proposals; companies that have SOPs, HR policies and product documents but nobody can find them when needed; and researchers, lawyers or consultants who have to read ten or fifteen reports against each other. It does not write anything new. It finds what is in the documents. If you want writing in your own style, that is a separate gig: "Custom GPT / digital me". ◆ What needs to be said honestly Not going beyond the sources also means it does not catch the sources' mistakes. If an old policy is in the notebook, the old answer will come, with a citation. So deciding which files stay is your decision. We flag suspicious files but do not remove them ourselves. Text cannot be read properly from old PDFs in Bijoy/SutonnyMJ fonts and from scanned paper. I tell you on the first day which ones are like that. NotebookLM runs in your own Google account, and the free plan has limits. If you need more, the paid plan is in your name, outside our fee. This is deliberate, so that the notebook stays in your hands even if you stop working with us tomorrow. ◆ How it goes After you order, you put the files in a Drive folder and write what your team most wants to know: 10 real questions. You open a notebook in your own account and add us as editors. We never ask for passwords. Then the notebook, tests, briefing and handover. At the end you remove our access.





