Chain of Thought prompting: definition, examples and difference with prompt chaining

Article published in September 2024, updated in October 2026: examples of prompts, place of Chain of Thought in relation to reasoning models, prompt chaining in agents.

Chain of Thought prompting refers to a technique that structures a reasoning task into intermediate steps. It is used for problems requiring logic, calculation, or decomposition. It should not be confused with the prompt chaining, which consists of chaining several distinct prompts in order to advance a task step by step.

What is Chain of Thought prompting?

Chain of Thought prompting (CoT) is a prompt engineering technique aimed at improving the performance of language models on tasks requiring logic, computation, and decision-making, by breaking the problem down into steps.

How does the Chain of Thought work?

When faced with a complex problem, one strategy is to break it down into simpler sub-problems. Prompting can replicate this structure by instructing the model to process the necessary elements sequentially before producing the expected result.

The work published in 2022 under the title "Chain-of-Thought Prompting Elicits Reasoning in Large Language Models" has helped to popularize this approach for various arithmetic, symbolic and common-sense reasoning problems.

Examples of Chain of Thought prompts

There are generally two ways to trigger a chain of thought.

1. The "zero-shot" CoT: The model is simply asked to detail their reasoning before concluding.

Our team of 12 is switching to working remotely 4 days out of 5. We currently rent 14 workstations at €650 per month each. How many workstations do we need to keep to cover 90% of our on-site presence, and what annual savings will this represent? Work through the results step by step, then present them in a table.

2. The "few-shot" CoT: The model is shown one or two solved examples, along with their reasoning, before the real question is posed. The model reproduces the solution structure shown.

Example: an invoice of €1,200 excluding VAT with 20% VAT and a 5% discount on the pre-tax amount. Reasoning: discount = €60, net pre-tax amount = €1,140, ​​VAT = €228, total including VAT = €1,368. Your turn: an invoice of €3,450 excluding VAT, 20% VAT, 8% discount on the pre-tax amount.

What is the purpose of Chain of Thought prompting?

• Break down a complex problem into sub-tasks.
• To structure a logical solution or a calculation.
• Organize the necessary information before the final answer.
• To make the reasoning readable, and therefore verifiable by a human.
• To reduce certain errors related to a solution that is too direct.

The Chain of Thought in the Age of Reasoning Models (2026)

Since 2024, major software publishers have offered so-called "reasoning models" or extended thinking modes: these generate an internal thought process before responding. In consumer and professional assistants, this mode often appears as a "thinking" option or a dedicated model in the selector.

A practical consequence of this is that, with these models, adding "reasons step by step" is much less useful than it was in 2023. Developers recommend instead providing high-level instructions and letting the model organize its reasoning. The prompt's role shifts to three key elements:

• The context : the data, constraints and assumptions that the model cannot guess.
• Success criteria: What constitutes a good answer, and what should lead to a solution being rejected.
• The verification: ask the model to check its result (recalculation, consistency of figures, sources) before delivering it.

The explicit Chain of Thought remains useful with fast models, which do not reason internally, and whenever we want to impose a precise method: an analysis grid, a control procedure, a regulatory calculation order.

To choose between a rapid model and a reasoning model depending on the task, see our article. The right AI model for each task.

Limitations of the Chain of Thought

A structured response does not mean that a model reasons like a human. A model can produce a plausible explanation while still arriving at an incorrect conclusion, and the displayed reasoning does not always accurately reflect the calculation actually performed by the model. Therefore, the breakdown of a task must remain linked to the verification of facts, calculations, and the final result.

Chain of Thought and Prompt Chaining: what's the difference?

Prompt chaining is particularly useful when a task involves several separable operations, such as searching for information, synthesizing it, and then transforming the synthesis into a deliverable. Chain of Thought is more concerned with structuring a problem that requires sequential reasoning.

Example of prompt chaining

1. Prompt 1: "Here are 40 customer testimonials. Classify them by theme and count the occurrences."
2. Prompt 2: "Based on this ranking, identify the three main irritants and cite two representative verbatim quotes for each."
3. Prompt 3: "Write a one-page memo for the management committee outlining these three irritants and providing one recommendation for each irritant."

Each step can be reviewed and corrected before moving on to the next, which limits the propagation of errors. By 2026, agents and delegation tools of this type Cowork They apply this logic automatically: they break down a request into steps, chain the processes together, and only return to the user with the deliverable. Understanding prompt chaining helps to better define the tasks assigned to them. See our Cowork.

To delve deeper: this topic echoes our reflection on the future of the prompt engineer profession and to our article on the job of Prompt Designer.