What is chain-of-thought prompting?
Chain of thought (CoT) is a prompting technique that asks the model to show its reasoning step by step before answering. Phrases such as “think step by step” greatly improve quality on logic, maths and complex trade-offs.
DEFINITION
Chain of thought means the language model does not jump straight to an answer but states its reasoning explicitly, step by step—it “thinks out loud,” which usually produces stronger results.
The simplest pattern: add “Think step by step” or “Explain your reasoning first.” The model then surfaces its working and can self-check along the way.
CoT helps especially with:
- Math problems where intermediate work must be visible.
- Logical inference such as if–then chains and case splits.
- Decision trade-offs including pros/cons and risk views.
- Planning where ordering and dependencies matter.
Why it works: language models generate token by token. Writing intermediate steps makes those tokens available for the next ones, so later reasoning can build on earlier steps. Without CoT, many intermediate steps stay implicit and errors slip in.
Variants
- Zero-shot CoT: “Think step by step.”
- Few-shot CoT: you include examples with full solution traces.
- Auto-CoT: the model decides when to externalise its reasoning.
CONNECTIONS
Leadership
Situational leadership needs context and judgment. CoT prompts help leaders use AI for decision support by walking through situational variables instead of receiving a generic verdict.
Agility
Retrospectives benefit from structured analysis paths. CoT keeps teams from leaping to fixes before causes are thought through.
Project management
Critical path and dependency reviews are natural CoT tasks: the model sequences work and spots bottlenecks more reliably than with a one-shot answer.
KEY POINTS
- “Think step by step” is the simplest CoT trigger.
- CoT strongly improves logic, maths and nuanced judgment tasks.
- Models can build on their own written intermediate states—that is the mechanism.
- Few-shot CoT with worked examples usually beats zero-shot CoT.
- CoT makes AI answers easier for humans to audit.
EXAMPLE
Without CoT: “Should we fund project A or B?” → a snap recommendation, often thin.
With CoT: “Analyse step by step which project we should prioritise. Consider strategic fit, resource use, risk profile and expected ROI.” → the model works through each lens, weighs trade-offs and justifies the call so others can review it.
MISCONCEPTIONS
Doesn’t CoT make answers too long?
Sometimes. On hard tasks the extra length buys quality. If you only need a crisp conclusion, append “Then summarise your recommendation in two sentences” to keep both trace and brevity.
Should I use CoT for every question?
No. Trivial prompts gain little and run slower. Use CoT when analysis depth and the reasoning path matter.