Power-userClaude

🧨 Make Claude think harder: the extended thinking cheat sheet

Claude has a second gear most people never engage: extended thinking, where it reasons privately before answering instead of writing the first plausible response. Knowing when and how to trigger it is the difference between an okay answer and the answer. Here's the cheat sheet.

How to trigger deeper thinking

In the Claude apps: toggle extended thinking on (look in the message controls). Best turned on per-task, not permanently.
In Claude Code: thinking keywords scale the effort — asking Claude to "think hard" allocates reasoning budget, and ultrathink requests the maximum. Newer versions also expose effort settings directly. One word in your prompt is all it takes.
Anywhere, by prompt: even without a toggle, process instructions help: "Before answering, work through this step by step, consider the angles I didn't ask about, and check your first instinct against a second approach."

When it's worth it (and when it's waste)

Use it for: decisions with real stakes, multi-constraint planning, debugging something weird, math and logic, analyzing data for the "so what," anything where your reaction to a fast answer would be "hmm, is that right?"

Skip it for: drafts, rewrites, brainstorms, formatting, lookups. Extended thinking on a caption is a long wait for the same caption. It's also slower and burns more usage — spend it like the budget guide says: on work that deserves it.

Prompts that aim the thinking

Think through this carefully before answering. Specifically: generate two genuinely different approaches, stress-test each against [my constraint], and tell me which you'd pick and what would change your mind. Then note what you almost got wrong on your first pass.

This decision matters: [context]. Take your time. Consider the angle I didn't ask about, the assumption most likely to be wrong, and the second-order effects. Conclusion last, reasoning first.

Read the reasoning, not just the answer

When Claude shows its thinking, skim it. You'll catch wrong assumptions early ("wait — that constraint isn't true"), and correcting the premise beats re-rolling the conclusion. The reasoning is also where the second-best option lives, which is sometimes the one you actually want.

The pairing that punches hardest: extended thinking + the SHARP honesty mode from the shortcuts guide. Deep reasoning that's also willing to tell you your plan has a hole — that's the advisor most people never configure.

Want this done for you?

We set up these exact systems for businesses like yours. One call, honest scope, no pressure.

Start your project