Think Step by Step

If someone shook you awake and demanded instant answers to tricky questions, you would do worse than with a moment to think. Claude is the same: on hard tasks, letting it reason out loud before answering makes it noticeably more accurate.

Thinking only counts when it is written out

One rule before the examples: the thinking must appear in the answer. Asking Claude to “think silently and reply with only the final answer” does not work — reasoning that is never written out has not actually happened.

Example: the sarcastic movie review

Read this review: it opens with praise, then admits the reviewer has been living under a rock since 1900 — meaning “fresh and original” comes from someone who has seen nothing for a century. Asked directly, Claude takes the word “unrelated” literally and calls the review positive:

YOU WRITE

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 the year 1900.

CLAUDE ANSWERS

The sentiment is positive — the reviewer describes the movie as fresh, original, and mind-blowing.

Now the same question, but with a role plus an instruction to argue both sides first:

STANDING INSTRUCTION (SYSTEM PROMPT)

You are a savvy reader of movie reviews.

YOU WRITE

Is this review sentiment positive or negative? First, write the best arguments for each side in <positive-argument> and <negative-argument> 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.

CLAUDE ANSWERS

<positive-argument>The reviewer calls the film fresh, original, and mind-blowing — strong words of praise.</positive-argument> <negative-argument>The reviewer admits to living under a rock since 1900 — so anything would seem fresh and original to them. The “praise” is actually a joke at the film’s expense, implying it is badly outdated.</negative-argument> Verdict: the review is negative — it is sarcastic.

Thinking out loud let Claude catch the sarcasm it had missed.

A curious side-note: order matters

Claude is sometimes sensitive to the order of choices. In borderline cases it leans toward the second option offered — likely a habit from its training data. Asked “negative or positive?” instead of “positive or negative?”, its verdict on a borderline review can flip. Worth remembering when you design questions with options.

Example: from wrong to right with one sentence

Ask for a fact needing recall, and Claude can guess wrong:

YOU WRITE

Name a famous movie starring an actor who was born in the year 1956.

CLAUDE ANSWERS

“Top Gun”, starring Tom Cruise.

Wrong — Tom Cruise was born in 1962. Claude jumped to a famous name instead of checking the year.

Add one sentence asking it to brainstorm first, and watch the difference:

YOU WRITE

Name a famous movie starring an actor who was born in the year 1956. First brainstorm about some actors and their birth years in <brainstorm> tags, then give your answer.

CLAUDE ANSWERS

<brainstorm> Tom Hanks — born 1956. Mel Gibson — born 1956. Tom Cruise — born 1962 (does not qualify). </brainstorm> A famous movie starring an actor born in 1956 is “Forrest Gump”, starring Tom Hanks.

REMEMBER

On any hard or tricky question, add one line: “Think it through step by step before you answer, and show your reasoning.” Written-out thinking regularly turns wrong answers into right ones.