Your second brain is only as good as your tags

By Gerald · 6 July 2026

Wooden library card catalog drawers with small labelled index cards, representing notes filed so they can be found again

Most people building a second brain obsess over the same things. Which app. Which folder structure. Which method, PARA or Zettelkasten or whatever is trending this month. Almost nobody talks about tags, and I think tags are the most underrated feature in note-taking.

The test of a second brain is simple: three months from now, can you find the note you wrote today? Not by scrolling. By pulling it up in seconds, together with everything else related to it. Folders alone fail this test. Tags, done properly, pass it.

Folders answer one question. Tags answer the rest.

A folder gives a note a place. A tag gives it meaning.

Every note can only live in one folder, which forces a bad decision at capture time. Say you write a note about using AI to speed up SEO deliverables for a client. Does it go in the client folder? The SEO folder? The AI folder? Whichever you pick, the note disappears from the other two contexts.

Tags fix this because a note can carry several. Tag that same note seo, ai and client-work and it now shows up in all three contexts at once.

This is what people miss about tags. They add context, and context is what makes a second brain worth having. When you pull up the seo tag, you're not opening a folder. You're reconstructing everything you've ever thought about SEO: meeting notes, article ideas, research, client work, across every project and every year. Memory works by association, and tags are how you give your notes the same property. A tagged library feels alive. A foldered one feels like a filing cabinet.

Why tags fail for most people

Four plain paper tags hanging on strings against a white wall — A tag is only useful if future you can predict it. One format, no exceptions.

If tags are this good, why do most people abandon them?

Because they tag without a standard. You tag one note Marketing, the next marketing, a third mktg. You create a new tag every time you're unsure, and six months later you have 200 tags, most used exactly once. Search marketing and you get a third of your library plus none of the notes filed under the other two spellings. At that point your tags lie to you, you stop trusting them, and you tell everyone tags don't work.

Set a standard and stick to it

This is the whole game. A tag is only useful if you can predict it. Future you, typing into a search bar, should be able to guess exactly what past you named the tag. That only happens with rules. Mine:

One format, no exceptions. All lowercase, singular, hyphens for multi-word tags: client-work, not Client Work, ClientWork or client_work. Boring on purpose. Predictable beats clever.

A small, stable set. Somewhere between 15 and 30 tags covering the topics you keep coming back to. Before creating a new tag, check the existing list first. If a tag will only ever sit on one note, it doesn't deserve to exist.

One concept, one tag. Don't run ai, artificial-intelligence and llm side by side for the same idea. Pick one, merge the rest.

Tag the subject, not the contents. Ask "what is this note about" and answer in one to three tags. If you need five, the note is probably two notes.

Don't duplicate your structure. If a note already sits in your Clients notebook, tagging it clients adds nothing. Tags earn their keep by cutting across notebooks, connecting notes that live in different places.

Prune quarterly. Merge duplicates, retire dead tags. Twenty minutes, four times a year, and your tag list stays trustworthy.

Write your standard down somewhere you'll see it. The specific rules matter less than having rules at all. A consistent system beats a clever inconsistent one every time you search.

Your tags now have a second reader

There's a newer reason to care about this, and I'd argue it has become the biggest one: you're no longer the only reader of your notes.

MCP (Model Context Protocol) is the open standard that lets an AI assistant connect straight to your apps. Flow Productivity has this built in: connect your notes over MCP to ChatGPT, Claude, Grok or Perplexity, and the model you already use every day gets full context over everything you've captured. No exporting, no pasting notes into a chat window. (And anything you mark confidential stays out of the connection entirely.)

Once your notes are connected, tags stop being a personal filing habit and become machine-readable context. Ask your assistant to "pull together everything I have on GEO before my client call" and watch the difference. On an untagged library, it keyword-guesses and hopes. On a tagged one, geo returns the exact set, and the assistant builds its answer from notes you deliberately grouped together.

Tags draw the map

Tags do one more thing in Flow that most note apps never attempt: they draw a map. Flow's Mind Map builds a knowledge graph from your library, where notes, tasks and tags are the points and the relationships between them are the lines. Notes that link to each other are connected. Notes attached to tasks are connected. And every shared tag connects the notes that carry it. Tag twenty notes seo over a year and the map shows them as one cluster, even when they live in five different notebooks.

That same graph is exposed to your AI through MCP. A connected assistant can ask for the whole graph, or for everything related to a single note: the notes it links to, the tasks attached to it, and the sibling notes that share its tags. So when you ask about one client note, the assistant doesn't stop at that note. It follows the edges and pulls in the surrounding material you'd have had to remember to paste in yourself.

This is where a tag standard compounds. Every consistent tag is another edge in the graph, and every edge is context your assistant can follow instead of guess at. Sloppy tags don't just make search worse, they leave holes in the map.

A consistent tag is metadata your AI can act on. Your tag standard becomes a shared language between you, your notes and your tools. This is the point where a second brain goes from nicely organised to super useful: you stop searching your notes and start asking questions of them.

Frequently asked questions

How many tags should a second brain have? Somewhere between 15 and 30 for most people. That is enough to cover the topics you keep returning to, and small enough that you can remember the whole list. If a tag will only ever sit on one note, it does not deserve to exist.

Should I use tags or folders for my notes? Both, for different jobs. A folder or notebook gives a note one home. Tags cut across those homes and connect notes that live in different places. The failure mode is using tags to duplicate your folder structure, which adds effort and no new information.

What is a good tag naming convention? Pick one format and never deviate: all lowercase, singular, hyphens for multi-word tags, like client-work. The exact rules matter less than consistency. A tag is only useful if future you can predict what past you named it.

Can AI assistants actually use my note tags? Yes, if your notes app exposes them. Flow's MCP connector gives ChatGPT, Claude, Grok and Perplexity direct access to your notes and their tags, so a request like "everything tagged geo" returns the exact set of notes you grouped, instead of a keyword guess. Shared tags also form connections in Flow's knowledge graph, which assistants can traverse to pull in related notes you did not name.

Two seconds now, years of payoff

Tagging a note properly costs about two seconds at capture. That's the entire investment. The return is a library you can trust: notes resurface when you need them, related ideas find each other, and both you and your AI pull the right context on demand.

Folders store your notes. Tags make them think together.

I built tags into Flow with exactly this in mind. The tag layer is wired straight into Flow's MCP, so the moment you connect your AI, every tag you've maintained becomes context it can work with. If you want a second brain your assistant can actually read, give Flow a try.