Annotation System I've Been Using

This is the second in a three-part series. The first piece is here — it’s about why LLM research sessions (ChatGPT, Claude, Gemini, Perplexity) lose their best thinking before you even reach the end.

I’ve been annotating by hand for most of my working life. If you picked up one of my notebooks, you’d see the same marks on almost every page. A circled A for something I had to do. A circled Q for a question I needed to answer. A circled I for something I’d just learned. A star for something that mattered enough to find again.

A page from my notebook — the same system I’ve used for decades.

Nothing I invented deliberately. It developed out of necessity — out of every situation where something came up that I couldn’t act on immediately.A Q in a meeting was a question I wanted to ask the presenter but couldn’t interrupt for. Or a question that surfaced in a group discussion where it wasn’t the right moment to raise it. Later, when I was studying, it was something from a

textbook I didn’t understand and needed to go back to. The common thread: something unresolved that I couldn’t leave behind but also couldn’t chase right now.

An A was an action item — pure and simple. Something coming out of a meeting that I had to do. Something I needed to remember to handle. Pay a bill. Follow up with someone. Send a document. It didn’t matter how big or small. If I had to do it, it got an A.

A star was for moments when something really landed. In a meeting, something we weren’t doing that we should be. In a textbook, a concept that reframed everything I’d understood before it. The feeling was always the same — this is important and I cannot let it get buried.

The circled I came later, as I started taking more structured notes. New knowledge. Something that changed my understanding. Distinct from important — you can learn something without it being critical, and something can be critical without being new.

By the time I started doing deep research with LLMs, these marks were pure instinct. Decades of meetings and exams and books had made them automatic.

The Q did a lot of work on paper. A question I needed to answer, a thread I wanted to follow, something I was curious about but couldn’t chase right now — all of it went under Q. The volume was low enough that the distinction didn’t matter. I could look at my Qs at the end of a session and know intuitively which ones were urgent and which ones were just interesting.

LLMs broke that.

The number of threads that branch off a single LLM session is an order of magnitude higher than anything I encountered reading with a pencil. Suddenly my Q list was a mess — genuine questions I needed to answer sitting alongside rabbit holes I might never go down, and no way to tell them apart at a glance. The distinction that hadn’t mattered on paper became essential.

So F split off from Q. Not by design — by necessity. A Follow-up is curiosity consciously deferred. A Query is unresolved tension that needs resolution. They feel different when you mark them. They belong in different places in the digest. That’s the only symbol the system gained in moving from paper to LLMs.

Five symbols. 

Here’s what each one represents — and more importantly, what mental state you’re in when you reach for it.

'Q' — Query is unresolved tension.

Something in the conversation has opened a question you don’t have an answer to yet. Your brain is uncomfortable. It wants to chase it. The 'Q' mark is how you tell yourself: ‘I know this is open. I haven’t forgotten it. Keep moving.

'I' — Information is new ground.

Something just changed your understanding of the domain. A fact, a term, a framework you didn’t have before. Not an opinion, not a conclusion — actual new knowledge. The feeling is quiet but distinct. You know when it happens.

'★' — Important is a moment of clarity.

Something landed. An insight, a connection, a realisation. It may or may not fit neatly into a category — it doesn’t matter. The feeling is: this is the kind of thing I came here for. Use this sparingly. If you mark everything as

important, the signal disappears.

'A' — Action is commitment.

Not a thread to explore — something you have to do. The mental state is different from curiosity. It’s obligation. An 'A' either gets done or it doesn’t. And one that doesn’t get done isn’t a loose thread — it’s work that was supposed to happen and didn’t. Mark it because it has weight.

'F' — Follow-up is curiosity, consciously deferred.

A rabbit hole you want to go down — but not now. You’re aware of it, you’re choosing to leave it, and you want to be able to find it again. Unlike an 'A', an 'F' stays open by design. It may never get resolved. That’s fine. The

Follow-up is intellectual appetite, not obligation.

At the end of a session, these marks become what I call the annotated digest — everything pulled out, separated by type, independent of the conversation that produced it.

Queries become a research backlog. Information is what you learned. Starred items are the headline findings. Actions are the task list. Follow-ups may seed the next session.This is the document you actually use. Not the chat. Not the model’s summary.

Your own signal, structured and portable.

Each session feeds the next. The work compounds.

I used this system manually for months. It worked. But doing it by hand at the end of a long session required discipline I didn’t always have — and the manual step meant some sessions still lost their annotations. So I built a tool around it. A Chrome extension that lets you mark anything in an LLM session — or any webpage — with these five symbols, and generates the

digest automatically.

That’s what the third piece is about.

Bijal Sanghavi is the founder of Trika.ai, a product and AI consulting firm. 27 years building

enterprise software, most recently as CTO at Netcore Cloud, Zycus, and Saba.