I used to think I was being clever. Short prompts, quick messages, fewer words to save Claude tokens. Turns out I had it completely backwards, and a post from this AI professional just flipped my whole mental model.
The original poster makes a bold claim right out of the gate: stop writing short prompts to save tokens. Write the longest prompt you can. I’ll admit I raised an eyebrow at first, but once the creator explained the mechanics, it clicked hard.
Why short prompts secretly cost more
Here’s the part most people miss, and the expert lays it out beautifully. Claude re-reads your ENTIRE conversation on every single message. That’s the key fact everything else hangs on.
- Message 1 is cheap.
- Message 30 re-reads all 29 previous exchanges first.
- A short prompt is basically a wrong guess, which leads to “no, I meant…”, which forces another full reload.
- Every reload burns tokens. Tokens are money.
The vague short prompt feels efficient, but it triggers a chain of follow-ups that each reload the whole thread. That’s where your budget quietly disappears.
Approach 1: The short-prompt trap
This savvy professional paints the failure pattern in a way that felt uncomfortably familiar to me:
- You type “make it better.” Lazy and vague.
- Claude guesses wrong, because it has nothing to work with.
- You send more context, and Claude re-reads everything again.
- You’re out of tokens by 2pm.
The cost adds up not because you wrote too much, but because you wrote too little and paid for it in repeated reloads.
Approach 2: The long-prompt fix
Now flip it. The creator’s alternative is to front-load every bit of context into one message, so Claude reads once and starts working instead of guessing.
- You speak roughly 4x faster than you type, so you naturally give more context.
- One message. One read. One good answer.
- Claude has everything up front, so it stops re-reading and starts working.
The post’s author even shares a real example of what a rich, specific prompt sounds like:
“Tone’s too stiff. Make it sound like I’m texting a friend who runs a 200-person company. Keep the data. Only redo section 2.”
Notice how much is packed in there. The fix, the tone, the constraint, and the exact scope. No back-and-forth needed.
Side by side
Short prompts: feel fast to send, but trigger wrong guesses, endless follow-ups, and repeated full-context reloads that drain your tokens.
Long prompts: take a few more seconds to produce, especially by voice, but deliver one clean read and one strong answer.
The recommendation from this contributor is clear, and I’m on board: go long, and do it by talking instead of typing.
How to actually do it
The author shares a simple voice-driven workflow so the long prompt doesn’t feel like a chore. Here are the steps:
- Open Claude, but don’t type your first message yet.
- Install Wispr .ai for free.
- Set your shortcut key on the keyboard. The creator uses shift.
- Back in Claude, hold the shortcut key and start talking.
- Don’t type. Say everything in one shot: the task, the context, the format, the tone.
- Press enter to send the prompt.
- The prompt already carries the entire context.
Why it matters: the original poster makes a sharp point that stuck with me. You’re not lazy because you write short prompts. You write short prompts because typing is slow. So remove the typing, and the long prompt becomes the easy one.
Tips to get more from this
A few practical ways I’d apply what the expert shared:
- Before you talk, picture the finished output. Describe the format you want, not just the task.
- Name your audience out loud. “Explain this to a non-technical founder” beats “explain this.”
- Set the scope. If only one section needs work, say so, like the creator’s “only redo section 2” example.
- Batch your context. Task, background, tone, and constraints in a single pass.
What I love about this idea is how it connects to a bigger trend. As more of our work moves into long-running AI chats, the cost isn’t really per word, it’s per reload. The people who learn to give complete context fast will get better answers for less. The mind behind this post basically handed us a cheat code for that.
This is how you turn a wall of follow-ups into one prompt that just works. ♻️
Want the full breakdown and the exact wording the author used? Check out the original LinkedIn post for all the details.