How to Set Up an AI Second Brain You Can Keep Using

To set up an AI second brain, choose one project, collect its reliable sources, and separate reference notes from decisions and next steps. Then test whether an assistant can answer real questions using that record, with evidence you can open.

You can keep documents in a notes app and use fenbs for the project context different assistants need to share. The useful starting point is a small, accurate record you can maintain. A large archive is optional; a clear answer about what happens next is essential.

What should you set up first?

1. Pick one active outcome, such as preparing a class presentation or revising a client proposal. Write what a finished result must include.
2. Gather the few sources you already use to make decisions. Save their titles, locations, and the dates or versions that matter.
3. Create separate places for incoming material, checked notes, human decisions, and current work. Give every saved item a clear home.
4. Ask your assistant three questions you can answer yourself. Compare each response with the original evidence before expanding the collection.

For example, a student preparing a presentation might begin with the assignment instructions, two readings, and feedback on an outline. Adding an entire semester of downloads would make the first test harder to understand. Start with enough evidence to answer the next question, then add material when a real question requires it.

If you want the underlying concept before choosing a structure, read what an AI second brain is. This setup is about making that concept useful in an ordinary week.

How should the information be organized?

Use a structure based on how information changes. Incoming material needs review. A checked reference may remain useful for months. A decision needs an owner and a reason, while a next step can become outdated tomorrow. Putting these together in one long note hides those differences.

Connect a note to its project through a link or a clear project name. Avoid building a complicated tag vocabulary before you know what you will search for. A title such as “Presentation: evidence for the opening example” is easier to recognize than “Interesting ideas 4.”

An AI brain becomes useful when this information is available at the moment of work. Organization alone does not make an assistant read it; your session instructions must say where to begin.

How do you keep notes connected to their evidence?

Preserve the original separately from an AI summary. Give the summary a source title and a precise location, such as a section name or page number. Label your own interpretation so a future reader does not mistake it for the author’s claim.

For research collections, Zotero documents notes attached to individual sources↗ (opens in a new tab) as well as standalone notes. That distinction is useful in any system: evidence belongs with its source; a synthesis can connect several sources while pointing back to each.

• Source: the original document, its author or owner, and a stable location.
• Checked takeaway: the claim you verified and where the source supports it.
• Uncertainty: missing evidence, conflicting information, or a detail that may change.
• Use: the question, project, or task this information helps you address.

Do not save information simply because an assistant generated it. If you cannot explain its future use or check its basis, leave it in the working conversation. The second brain should reduce the amount you must inspect next time.

How do you know your assistant can retrieve the right context?

Test a fact, a comparison, and a missing answer. Ask which assignment requirement governs the presentation length, how two readings disagree, and something the collection does not establish. A useful system identifies the relevant evidence and admits the gap.

Ask the assistant to identify supporting passages before explaining its conclusion. Anthropic’s prompting guidance↗ (opens in a new tab) recommends grounding work on long documents in relevant quotations. You can use the same review habit with your chosen assistant, then check the quotations yourself.

When an answer fails, diagnose the cause. Was the source missing, was the correct version hard to identify, or did the assistant misread a passage it found? Each problem needs a different repair. Adding more files is useful only for the first one.

What should become lasting project context?

Promote only reviewed information that changes future work. A useful project summary says what the goal is, which constraints apply, what a person has approved, and what still needs checking. Leave the full reading notes available through source links.

Keep proposals separate from approvals. If an assistant suggests dropping an example, record it as an option until a person chooses it. Once chosen, record the reason so the next session can continue without reopening the same discussion.

In fenbs, Decisions and rules and Where we left off give that reviewed context a home; portable AI memory explains why keeping it outside one conversation matters.

How to do this with fenbs

1. Create a free account, then add the project you used for the initial test.
2. Add current tasks and short context notes. Include source locations and state what remains uncertain.
3. Record what people have decided in Decisions and rules. Keep questions that still need a person clearly open.
4. Connect your assistant through Integrations. Ask it to read the AI context before starting and save progress before stopping.

fenbs is free for one person, with no credit card. It is not the right fit when your only need is storing and annotating long research documents; keep a document or reference tool for that job. It also does not import your chat history.

How do you keep the system useful after setup?

At the end of a work session, update the next action and mark which checks actually passed. During a regular review, empty the inbox, flag obsolete notes, and check the sources behind active decisions. Archive completed work without removing the reasoning you may need later.

For an app-specific version, use the guide to setting up a second brain in Claude. For more approaches, browse the second-brain guides. Keep using the same small retrieval test when you change your setup so you can tell whether it improved.