Make Claude Think Step by Step
Make Claude Think Step by Step
Imagine being woken up by a friend who immediately hits you with complex math problems, demanding instant answers. You’d probably be dazed, right?
But if you had time to wake up properly and work through it step by step with pen and paper, things would be different.
Claude is the same.
Sometimes Claude gives irrelevant answers or simply gets it wrong. Not because it’s not smart enough, but because you made it “react too quickly.”
Why Slow Down AI?
When facing complex problems, if you demand an immediate answer, Claude often “cuts corners” – relying on quick intuition instead of careful analysis.
It’s like a teacher demanding only the final answer without showing work. You might guess right or wrong, but definitely not as reliable as working through it properly.
The key: thinking must be “out loud” to count.
You can’t tell Claude “think about it first, then tell me the answer.” If it only thinks in its “head,” that’s the same as not thinking at all. You must make it write out the thinking process – only then is it really thinking.
This technique in AI prompting is called “Chain of Thought” – basically making the AI show its reasoning process.
A Real Example
Take this movie review:
This movie blew my mind with its freshness and originality. In totally unrelated news, I have been living under a rock since 1900.
You can tell this is sarcasm. The first part praises the movie, the second part is joking that the reviewer has been “living under a rock.”
But if you directly ask Claude:
Is this movie review sentiment positive or negative?
This movie blew my mind with its freshness and originality. In totally unrelated news, I have been living under a rock since 1900.
Claude might get stuck because it takes the word “unrelated” literally, thinking the two parts really aren’t connected, and doesn’t know how to judge.
Make Claude “Think Out Loud”
Change the approach – make Claude write out its thinking:
Is this review sentiment positive or negative? First, write the best arguments for each side in <positive-argument> and <negative-argument> XML tags, then answer.
This movie blew my mind with its freshness and originality. In totally unrelated news, I have been living under a rock since 1900.
Now Claude will analyze carefully:
<positive-argument>
- Uses positive words like "blew my mind," "freshness," "originality"
- Expresses appreciation for the movie's creativity
</positive-argument>
<negative-argument>
- The second part about "living under a rock" seems unrelated
</negative-argument>
Answer: Positive
Making Claude write out the reasoning process helps it correctly identify this as a positive review.
An Interesting Quirk
Claude has a “quirk”: when things are ambiguous, it tends to prefer the second option.
Change the prompt to:
Is this review sentiment negative or positive? First write the best arguments for each side in <negative-argument> and <positive-argument> XML tags, then answer.
“Negative” is first now. In this case, Claude might actually judge it as negative because it “favors” the second option.
You can use this knowledge to nudge Claude toward a particular answer. But mainly, you should let it truly understand the problem, not play word games.
Practical Example: Actor Birth Year
Ask Claude:
Name a famous movie starring an actor who was born in 1956.
Ask directly, and Claude might guess randomly – possibly wrong.
But make it “brainstorm” first:
Name a famous movie starring an actor born in 1956. First brainstorm some actors and their birth years in <brainstorm> tags, then give your answer.
Claude will:
<brainstorm>
- Tom Hanks: born 1956
- Mel Gibson: born 1956
- Sean Penn: born 1960 (nope)
</brainstorm>
Answer: Cast Away (starring Tom Hanks)
Let it list candidates first, then choose – accuracy immediately improves.
How to Do It
Key points:
Explicitly require thinking process
Don’t just ask for the answer. Say:
Please first...then...
First list in...tags...then give answer
First...Second...Finally...
Use XML tags to organize thinking
<brainstorm>...</brainstorm>
<positive-argument>...</positive-argument>
<negative-argument>...</negative-argument>
<reasoning>...</reasoning>
Give enough thinking space
Don’t let Claude be lazy. Explicitly require it to think through each step:
Please analyze pros and cons of each option in detail
Please list at least 3 supporting reasons
Please derive step by step, don't jump
Use role prompts when necessary
Give Claude an identity to help it think deeper:
You are an experienced film critic...
You are a careful data analyst...
You are a detail-oriented engineer...
Pitfalls to Avoid
Don’t “think silently”
Think about it first, then tell me the answer – no
Please write your thinking process in
Don’t skip steps
Give me the conclusion directly – no
Please analyze in three steps: first…second…finally… – yes
Don’t only ask one direction
Analyze why this is positive – no
List both positive and negative arguments, then give conclusion – yes
Remember This
Make AI think out loud for reliable answers.
For complex problems, make Claude “brainstorm” or “list arguments” first. Use XML tags to organize the thinking process. Require step-by-step analysis without jumping. When ambiguous, try adjusting option order. Give clear role identities for deeper thinking.
Next Steps
Making Claude think step by step is just one part of prompt engineering. Coming up: how to control output format more precisely, how to make Claude “fill in the blanks,” how to handle complex multi-step reasoning tasks.
Combine these AI prompt techniques and you’ll find Claude’s capabilities completely different.
Ever had AI give you an obviously wrong answer? What was the problem? How did you solve it? Share your experience in the comments.
Found this article helpful?
- Like: If it helped, give it a like to help others find it
- Share: Forward to friends or colleagues who might need it
- Follow: Follow Dream Beasts Programming for more practical tech articles
- Comment: Have questions or thoughts? Join the discussion in the comments
Your support is my biggest motivation to keep creating!