The 11 habits making Claude dumber

I was a prompt hoarder and didn't know it

June 29, 2026

In partnership with

For a long time I collected AI tips like trading cards. Screenshots of other people's prompts. Saved threads. Bookmarked guides I opened exactly once. My folder of "good prompts" was huge, and my actual results were mediocre.

Then I read a post that flipped the whole thing. Instead of handing out 26 new Claude hacks, the writer argued you should unlearn 11 bad habits first. Every point stung a little, because I recognized myself in most of them.

What I liked is how unglamorous the advice is. No magic prompt, no secret menu. Just the small, lazy moves most of us make with the tool, and the simple fix for each one. Here's my walk through the ones that hit hardest, with why each one clicks.

LLM traffic converts 3× better than Google search

58% of buyers now start their research in ChatGPT or Gemini, not Google. Most startups aren't showing up there yet.

The ones that are get cited by the AI tools their buyers, investors, and future hires already use. And they convert at 3×.

Download the free AEO Playbook for Startups from HubSpot and get the exact steps to start showing up. Five minutes to read.

*Ad

The hoarding habits

These are the ones where I was collecting instead of doing.

You screenshot other people's prompts. The fix flips it: screenshot YOUR own broken thing instead. The error message, the ugly draft, the table that won't format. Drop the image in with "Fix this." Claude reads more from that screenshot than you'd ever type in a paragraph.

You wait until you "understand it" to start. The people who got good at this just opened it more often. Start a fresh chat, give yourself 10 minutes to fail, and you'll learn more than any course taught you.

You compare every model before you "commit." The advice here is blunt: pick Opus, set effort to high, go. The best model is the one you actually open. The comparison spreadsheet is just another way to not start.

The timid habits

This is the cluster that embarrassed me most, because I thought careful prompting was good prompting.

You re-roll the same prompt 9 times. Claude doesn't know your role, your audience, or your taste. Paste who you are and what you want ONCE at the top, and attempt #2 beats your old attempt #9.

You prompt like a coward. No more "maybe you could." Stop asking permission from a tool. Give it the outcome: "Book 3 meetings with this client by Friday." Name the finish line and it finds the path.

You spell out every step because you don't trust it. Every step you dictate is a ceiling you put on the model. I was treating Claude like a junior intern that needs a checklist. Delete the steps, hand over the outcome, and let it route there. It is smarter than your instructions.

You type everything when you can talk. You explain things out loud roughly 5x faster than you type, with more context and less self-editing. Dictate the whole brain dump and let it sort the mess.

The trusting-too-easily habits

The sneaky ones. They feel like progress while quietly lowering your quality.

You ship the first answer. It sounds sure even when it's wrong. Before you trust it, ask: "Argue against this answer. Where would a smart critic tear it apart?"

You nod along when it agrees with you. Call it what it is, a sycophant. It praises a bad idea as warmly as a good one, because agreeing is easier than pushing back. Add "Where am I wrong here? Argue the other side."

You ask it to 'improve' with no idea what's wrong. You have taste, Claude doesn't. Tell it exactly what's off in plain words: "Too formal. Cut it in half. Sound like a text to a friend."

You live in one huge chat you've had for 3 weeks. A long chat is a tired chat. Every old message it re-reads is noise. The thread you think is "full of context" is the thing making it dumber. Start fresh.

Your creative brief is due Friday. Viktor wrote it Tuesday.

Tell him the campaign. Viktor pulls last quarter's performance from Meta and TikTok, scrapes competitor ads, drafts the brief, posts it for review. You edit, he ships the creative requests to your designer. Inside Slack.

*Ad

The two that stopped me cold

Habit five was the spell-out-every-step one. I read it and realized I'd been doing exactly that, dictating steps because I didn't trust the model to route on its own. The moment I handed over a clean outcome instead of a checklist, the answers got better, not worse.

The long-chat one was the other wake-up call. I had a thread running for ages, convinced it held all my context. That bloated chat was the thing slowing everything down and muddying the answers. Starting fresh felt wrong for about ten seconds, then the responses got sharper instantly.

Four moves you can steal today

You don't need all 11 at once. Pick the ones that sting.

  1. Lead with the outcome. Swap "can you maybe help me with" for a clear finish line and a deadline.

  2. Build a skeptic into your prompts. End important requests with "Argue the other side" so weak answers die before they reach anyone else.

  3. Reset often. Open a new chat for a new task instead of dragging weeks of clutter along.

  4. Use your taste. You know when something feels off. Say it plainly and let the model adjust.

The thread running through all of it: most of us aren't held back by a lack of clever tricks. We're held back by timid prompting, by hoarding instead of doing, and by not trusting the tool to run with a goal. Fixing the habit beats learning the hack, every time.

Pick one and break it tonight

The post closes with a challenge I keep thinking about: pick one habit, break it today, not tomorrow. Open Claude tonight and catch yourself mid-habit. That's the only way any of this sticks.

So here's mine for the week. I'm killing the long-chat habit. New task, new thread, no exceptions. Pick yours, the one that stung reading this, and break it before you close the laptop.

HR and IT need to work as one. Here's how

When HR and IT don't talk, people fall through the cracks. This guide fixes the handoffs that matter most.

*Ad