Build a Personal Knowledge Base That Answers Questions

You'll end up with: A searchable knowledge base of your notes, docs, and expertise that you can query with AI

Overview
30-45 min
Intermediate
Free to start
2 tools
Cost breakdown
Claude.ai (planning and cleanup)Free tier
NotebookLM (source-backed answers)Free
TotalFree to start
Common mistake

Dumping every file into one workspace before deciding what questions it should answer. Start with one domain, define 5-8 real questions first, then add only sources that help answer those questions.

Before you start
  • Pick one knowledge domain or project
  • Gather 10-25 useful notes, docs, PDFs, transcripts, or links
  • Decide where the knowledge base will live
  • Open Claude and NotebookLM or Notion
1

Choose the knowledge base job

Narrow the knowledge base to one domain and define the questions it must answer.

Claude.aiFreeOpen Claude.ai
Exact action

1. Go to claude.ai and start a new chat 2. Tell Claude what scattered knowledge you have: notes, docs, PDFs, transcripts, SOPs, research, or client material 3. Ask Claude to define one focused knowledge base with: audience, domain, source types, excluded topics, and 5-8 high-value questions it should answer 4. Rewrite the output until the job fits in one sentence 5. Save the final job statement and questions in your source inventory

You have a one-sentence job statement like “This knowledge base helps me answer client onboarding and delivery questions from past proposals, SOPs, calls, and notes.”
If the scope sounds like “everything I know,” split it by workstream, client type, product, recurring decision, or project. One useful narrow base beats one giant messy base.
2

Build a source inventory

List the source documents before uploading anything so the base has intentional inputs.

NotionFreeOpen Notion
Exact action

1. Create a Notion table called “Knowledge base source inventory” 2. Add columns for Source name, Type, Location/link, Why it matters, Date/freshness, Question it helps answer, and Status 3. Add 10-25 starter sources from your notes, PDFs, docs, call transcripts, SOPs, research, templates, or client materials 4. Mark each source as “use now,” “reference only,” or “skip” 5. Keep only sources that help answer at least one question from Step 1

Each source has a clear reason to exist and maps to at least one question from Step 1. You can see the whole starter set on one screen or in one filtered view.
If most sources are old, vague, or duplicated, mark them “reference only” and prioritize recent material that supports decisions you actually make.
3

Clean and label the starter set

Prepare files so the AI can cite and distinguish them later.

Claude.aiFreeOpen Claude.ai
Exact action

1. Paste your source inventory into Claude 2. Ask: “Review this source inventory for a personal knowledge base. Suggest clean file names, 3-6 source groups, duplicate removals, missing context, and a small tag set that will help retrieval.” 3. Rename files or notes with descriptive titles before uploading them anywhere 4. Group sources into categories such as onboarding, pricing, delivery, research, templates, or decisions 5. Remove obvious duplicates and sources that do not support the questions from Step 1

Your sources are grouped into 3-6 categories and named clearly enough that a future answer can point back to the right document.
If Claude suggests a giant tagging system, cut it down. Keep only tags that help retrieval, such as client-intake, pricing, delivery, research, templates, or decisions.
4

Create the searchable knowledge base

Add the curated starter set to a tool that can answer questions from sources.

NotebookLMFreeOpen NotebookLM
Exact action

1. Open NotebookLM and create a new notebook named after your knowledge base job statement 2. Upload or paste your “use now” sources from the inventory, starting with the highest-value 10 if you have a lot 3. Keep source titles readable and grouped by the categories from Step 3 4. Add a short notebook description that says what this base should answer and what it should ignore 5. Compare the source list in NotebookLM against your inventory to make sure nothing important is missing

NotebookLM shows your source list, and each important source is present once with a title you recognize.
If uploads fail or limits get in the way, start with the highest-value 10 sources and keep the rest in the inventory for a second pass.
5

Test with real questions and improve gaps

Ask practical questions and use failures to improve the source set.

NotebookLMFreeOpen NotebookLM
Exact action

1. Ask the 5-8 questions you wrote in Step 1 2. For each answer, check whether it cites or references the right source 3. Mark each question as “trusted,” “missing context,” or “too generic” in your inventory 4. Add missing docs or notes where the answer is weak 5. Re-ask weak questions after adding or cleaning sources

At least 4-5 questions produce answers you would trust enough to use as a first draft, decision aid, or starting point for client work.
If answers are generic, ask “Which source supports that?” Remove or replace sources that are too broad, duplicated, stale, or unrelated.
6

Write the maintenance ritual

Make the knowledge base sustainable with a light update process.

Claude.aiFreeOpen Claude.ai
Exact action

1. Paste your source inventory and test-question notes into Claude 2. Ask: “Create a 10-15 minute monthly maintenance checklist for this knowledge base. Include what to add, what to archive, how to tag new sources, and which test questions to rerun.” 3. Save the checklist inside your source inventory or the knowledge base itself 4. Set a recurring calendar reminder to run it monthly 5. After each maintenance pass, rerun two test questions and note whether answer quality improved

You have a repeatable 10-15 minute ritual for adding new knowledge, archiving stale sources, and checking answer quality.
If the ritual feels heavy, reduce it to three actions: add useful new sources, archive stale ones, and rerun two test questions.

All done!

You now have: A searchable knowledge base of your notes, docs, and expertise that you can query with AI

Explore more guides

Want this workflow built for your business?

Book a free audit