Putting Thoughts into Words

This post synthesizes the author's own thoughts with advice from Claude (Opus 4.8) and ChatGPT (5.6 Sol xhigh).

Recently, I had a conversation with someone whose communication skills really impressed me. I've met plenty of people who can give long, polished speeches or talk fluently about many things. Much rarer, in my experience, are people who can convey deep ideas in just a handful of precise words.

After that conversation, I spent an afternoon wondering how to develop that ability myself. So I asked Claude and ChatGPT for help, experimented with some of their suggestions, and organized what I found useful here.

My goal is simple: I want to turn fuzzy ideas into concise and precise words quickly. The good news is that this seems much more like a skill problem than a talent problem. And skills can be trained deliberately.

I decompose the problem into three dimensions: speed, concision, and precision. Speed is about forming a thought, concision is about compressing it, and precision is about sharpening it. (I separate them for convenience, although in practice they are tightly coupled.)

Speed: getting a thought into a usable form quickly

I usually have no problem refining a thought given enough time to think or write; what I struggle with is doing the same thing in real time.

One useful insight from both Claude and ChatGPT was that speaking and writing are not merely ways to transcribe finished thoughts. They are also tools for finishing thoughts. An idea can feel clear while it remains in working memory. Once externalized -- on paper or out loud -- it becomes something we can inspect, test, and rearrange. This is why trying to "think everything through first" can sometimes make real-time communication harder rather than easier.

So the first step is simply to get the thought out using whatever modality works. A few exercises:

Freewrite. Pick one question and write continuously for 10 minutes without editing. The purpose is not good prose; it is learning to think through writing rather than thinking first and writing afterward.
Question storm. When an idea feels fuzzy, write ten questions about it before trying to answer any of them. Fuzziness is often several unresolved questions collapsed into one feeling.
Voice memo. Explain an idea aloud to nobody -- or to your phone. Speaking is another form of externalization, and it transfers especially well to live conversation.

The second step is to get faster at converting that externalized mess into structure. With enough repetition, the process becomes compressed enough to run mentally during a conversation. Two exercises are particularly useful:

One sentence first. Before elaborating, complete: "The point I'm trying to make is ___." If I can't fill in the blank, I probably haven't found the point yet.
Pre-write recurring conversations. Before a standup, 1:1, or review, spend five minutes writing three bullets: the point, the reason, and the ask. Eventually the structure becomes habitual and requires less preparation.

And here is an exercise that combines both stages. I call it reasoning distillation:

Choose one topic or question.
Write freely for three minutes.
Distill what you wrote into one sentence.
Expand that sentence into three bullets: point, reason, example.
Explain it aloud in 60 seconds.
Explain it again in 30 seconds without losing the main idea.

This is a drill for repeatedly going through the cycle of fuzzy → explicit → structured → compressed -- which is exactly what I want to become faster at.

Concision: identifying what needs to be said

I've generally been able to explain things clearly (proud educator moment!). But explaining something clearly and explaining it concisely are different skills. According to Claude, verbosity often comes from delivering one's thinking process instead of the conclusion. For example:

"So I looked at the logs, and there were a bunch of timeouts, and I checked the config, and it turns out the pool size… anyway I think we should raise the connection limit."
"We should raise the connection limit. The logs show timeouts from an exhausted pool."

The second version is more concise because the speaker goes directly to what matters.

A natural solution is to start with the conclusion, then add supporting details. ChatGPT suggests a few frameworks to structure one's thoughts. (Just like our research lessons, model-based beats model-free in sample efficiency.)

For a detailed argument: Point → Reason → Example → Limit → Implication
For a normal explanation: Conclusion → Reasons → Evidence, or Point → Evidence → Explanation
For framing a problem: Situation → Complication → Question → Answer

I don't think the specific framework matters very much; they just force one to decide what role each piece of information plays before saying it.

Two core skills that matter here are prioritization and compression.

A few techniques to train prioritization:

Rule of three. If I have seven things to say, I should probably identify the three that matter most and let the rest come out if someone asks.
What / So what / Now what. State the fact, explain why it matters, then state the action. This is particularly useful for status updates.
The "so what?" chain. Keep asking "so what?" until reaching the thing the listener actually cares about. Lead with that. Much of what came before was probably throat-clearing.
Rank, then cut. List everything I want to say, rank it by importance, and say only the top few items.

A few techniques to train compression:

Express the same idea at progressively smaller scales: 200 words, 100 words, 50 words, one sentence; at every stage, ask what information would actually change the listener's understanding, and remove everything else.
Take a real email, message, or paragraph and cut it by 50% without losing meaning. When reviewing the result, look for common problems: Does the main point arrive too late? Do several sentences express essentially the same idea? Do examples appear before the claim they are supposed to support? ...

The goal is not to make everything short; it is to make every sentence earn its place.

Precision: finding the exact words

The last skill I want to improve is the ability to identify the exact idea I mean and find the right words for it. Claude pointed out two common sources of imprecision: excessive hedging (phrases such as "I think maybe") and weak information hierarchy (where important claims and incidental details are presented with equal weight).

Thankfully, precision can be trained simply by refusing to let vague language pass unnoticed. For example:

"Things improved." What improved? By how much? Compared with when?
"The process is inefficient." Which step wastes whose time?
"People disliked it." Which people, and what evidence shows dislike?

Interestingly, I've found that forcing myself to use more precise language often improves the underlying thought as well. During one meeting, I almost said "there are several issues with this approach", but then I stopped and tried to actually name them. That small amount of pressure produced a much more concrete list of problems, and made me realize that the original sentence had been hiding the fact that I had not yet organized the thought.

A few editing habits help expose this:

One idea per sentence. Sentences containing multiple "and"s, "but"s, and side clauses often hide several claims that should be separated and evaluated individually.
Replace categories with specifics. "Issues", "things", "people", "better", and "improve" are useful warning signs. Ask what they refer to concretely.
Use hedges deliberately. Uncertainty should be expressed when the underlying claim is uncertain -- not automatically added as verbal insurance.
Check the hierarchy. The main claim should sound like the main claim. Supporting details should not compete with it for attention.

This is also why I became interested in improving my wording in the first place. Forcing precision in language doesn't just improve communication; it is also a mechanism for sharpening the underlying thought.

Closing the loop: practice and feedback

Lastly, I've been trying the following routine suggested by ChatGPT, since reading communication advice is obviously not enough.

Daily, ~10 minutes: freewrite on one question or compress something I wrote. Alternate between generating and cutting.
Before important meetings, ~5 minutes: write three bullets -- my point, my reason, and my ask.
After speaking live: notice one moment where I rambled or buried the lead. Rewrite what I wish I had said.
Once a week: explain something I learned to a real audience in under 200 words.

I also keep an idea log: a running document where half-formed thoughts have to become complete sentences. Over time, I want the fuzzy-to-clear conversion to become faster simply from repetition.

And practice needs feedback. A few ways I've been trying to shorten the feedback loop:

Read writing aloud. If I stumble over a sentence, there is usually something structurally wrong with it.
Edit after distance. Coming back a day later makes unnecessary language much easier to see.
Ask someone, "What was my main point?" If their answer differs from what I intended, the problem is probably structure rather than vocabulary.
Occasionally record myself speaking. Listen for filler, repeated ideas, and conclusions that arrive too late.

In the end, much of this seems to reduce to one skill: finding the point. The faster I can find it, the less I need to say around it; the more precisely I can name it, the clearer both the communication and the thought become. I guess I'll just keep practicing.