Chain-of-Thought Reasoning

Reviewed & published by Brayan K

Chain-of-thought prompting means asking the AI to show its work — to reason through a problem step by step instead of jumping straight to an answer. For anything that takes more than one mental hop (math, logic, planning), this simple habit dramatically improves accuracy.

Part of the free Prompt Engineering course at LearnCodingFast — hands-on lessons with worked examples and the output they print, plus practice exercises and a quick quiz.

Tools like ChatGPT, Claude, and Gemini are far more reliable when you let them think out loud. This lesson teaches you the magic phrases — and when to use them.

What You'll Learn in This Lesson

• What 'chain-of-thought' really means in plain English
• The phrase 'let's think step by step' and why it works
• Breaking big problems into smaller, ordered steps
• Getting reasoning AND a short final answer
• When step-by-step helps — and when to skip it
• Spotting where the logic went wrong and fixing it

1 Ask the AI to show its work

When a problem has several steps, an instant answer is a guess. Asking the model to reason step by step lets it build on each intermediate result. Same question, two prompts:

If a shirt costs $40 with a 25% discount
and 8% tax, what's the final price?
Just give me the number.

Demanding only the number invites a careless slip on a multi-step calculation.

A shirt costs $40 with a 25% discount,
then 8% tax. Work through it step by step,
showing each calculation, then give the
final price on the last line.

Reasoning first, answer last — easy to check, far less likely to be wrong.

2 Break big problems into ordered steps

You don’t have to leave the decomposition to the AI. Listing the sub-tasks yourself produces more complete, better-organised answers:

Naming the steps gives the model a checklist — and gives you a structure you can correct one piece at a time.

3 Reasoning AND a tidy final answer

Worried that step-by-step answers are too long? Ask for both. You get the accuracy of reasoning plus a clean conclusion you can copy:

Reason through this step by step.
Then, on the final line, write:
FINAL ANSWER: <your concise answer>

This pattern is perfect when you want to trust the answer but also paste it somewhere clean.

📋 Copy-paste chain-of-thought template

Here's the problem: <describe it>.
Work through it step by step, showing each
step of your reasoning. Check your work as
you go. Then give the final answer clearly
on the last line.

Try it for real

The reason to ask for steps is not that the model thinks harder — it is that each step becomes a claim you can test. Run this: four labelled steps, four chances to catch an error. A single number at the end is unfalsifiable, which is the whole problem with asking for one.

# "Let's think step by step" helps because it makes the steps INSPECTABLE.
# A hidden answer you cannot check is worth less than a shown one you can.

# A model's shown working on: "A shop sells 3 packs of 12 for £4.50 each.
# There is 20% off the total. What is the final price?"
steps = [
    ("packs x per pack", 3 * 12,            36),
    ("total before discount", 3 * 4.50,     13.50),
    ("discount amount", 13.50 * 0.20,       2.70),
    ("final price", 13.50 - 2.70,           10.80),
]

print("Checking each step against arithmetic:\n")
all_ok = True
for label, computed, claimed in steps:
    ok = abs(computed - claimed) < 0.005
    all_ok &= ok
    print(f"  {'OK ' if ok else 'BAD'}  {label:<22} model said {claimed:>6}   actually {round(computed, 2):>6}")

print()
print("all steps check out:", all_ok)
print()
print("This is the real value of chain of thought: not that the model 'thinks",
      "harder',\nbut that each step becomes a claim you can test. A single number",
      "at the end\nis unfalsifiable; four labelled steps are four chances to catch",
      "an error.")

# 🎯 YOUR TURN — check the working, find the bad step.

# The model's shown steps for: "A train travels 240 km in 3 hours,
# then 150 km in 2 hours. What is the average speed for the whole trip?"
steps = [
    ("total distance", 240 + 150,   390),
    ("total time",     3 + 2,       5),
    ("average speed",  390 / 5,     80),   # the model wrote 80
]

bad = []
for label, computed, claimed in steps:
    # 1) A step is wrong when the arithmetic disagrees with what was claimed.
    if abs(computed - claimed) ___ 0.005:
        # 👉 replace ___ with  >
        bad.append(label)

    print(f"{label:<16} claimed {claimed:<6} actual {round(computed, 2)}")

# 2) Report how many steps failed.
print()
print("bad steps:", ___)
# 👉 replace ___ with  bad

# ✅ Expected output:
# total distance   claimed 390    actual 390
# total time       claimed 5      actual 5
# average speed    claimed 80     actual 78.0
#
# bad steps: ['average speed']

Common Mistakes (and the fix)

• Demanding only the answer on a multi-step problem. — Let it reason first; ask for the number last.
• Using step-by-step on trivial lookups. — For a single fact, a direct answer is better.
• Leaving all the structure to the AI. — Name the sub-steps yourself for big tasks.
• Not reading the reasoning. — The visible steps are where you catch and fix mistakes.

⏱ Test Yourself — Timed Quiz

10 quick questions, 12 seconds each. Instant feedback — beat the clock!

🎉 Lesson Complete

• ✅ Ask the AI to reason step by step on multi-step problems
• ✅ “Let’s think step by step” switches the model from guessing to reasoning
• ✅ Break big tasks into ordered sub-steps yourself
• ✅ Ask for reasoning AND a tidy final answer on the last line
• ✅ Skip step-by-step for simple, single-fact questions

Practice quiz

Continue this course

• Previous: Few-Shot Prompting with Examples
• Next: Controlling the Output Format — Get JSON, tables, bullet lists or a set length — exactly as you need it

Frequently asked questions

What is chain-of-thought prompting in plain English?

It's simply asking the AI to show its work. Instead of demanding an instant answer, you tell it to reason through the problem step by step. Like a student showing each line of a math problem, this often leads to more accurate results on anything that takes more than one mental hop.

Does telling the AI to 'think step by step' really help?

Yes, especially for multi-step problems — math, logic puzzles, planning, or tangled reasoning. Walking through intermediate steps reduces the chance the model jumps to a wrong conclusion. For a simple fact lookup it's unnecessary and just makes the answer longer.

Can I get the reasoning but still a short final answer?

Absolutely. Ask for both: 'Reason through it step by step, then give just the final answer on the last line.' You get the benefit of careful reasoning plus a tidy, easy-to-copy conclusion.

How do I break a big task into steps for the AI?

List the sub-tasks yourself and ask the AI to do them in order. For example: 'Plan this in steps: 1) outline 2) draft 3) edit. Do each before moving on.' Decomposing the work makes long answers more complete and easier to follow.