♟️ Better Prompting Techniques: Reasoning
Simple yet very powerful way to improve the quality of AI answers
The quality (especially accuracy) of AI responses can be vastly different depending on the prompt you provide. Just asking AI a question in the first way that comes to mind will often work just fine, but at times will lead to completely wrong answers. Applying several techniques in such cases can help extract much more value from AI answers and turn wrong answers into correct ones. Over the next few posts, I’ll share my favorite techniques for improving prompts. Today, we’ll focus on one of the most powerful: asking AI to reason before answering.
The Power of Reasoning
Among all the methods for enhancing AI responses, prompting it to provide reasoning before answering is absolutely transformative. By encouraging AI to articulate its thought process, you’re more likely to receive accurate, nuanced answers.
You can approach this technique in two ways:
Ask the AI to reason freely.
Help guide its reasoning by outlining specific steps.
Either way, the goal is to engage the model in problem-solving rather than rushing to a conclusion, just with more or less help.
Why Does This Work?
AI doesn’t just respond to the text you provide in one shot — instead it generates word after word and every time it considers what it already said so far. This process is known as autoregression. Asking AI to provide reasoning plays into this trait and ensures that there is more detailed information available for AI to consider before giving the final answer — some of it from original question and documents and some from what AI already reasoned about.
And you have control over this process by providing more or less detailed instructions — from just “reason about” to allow AI to figure out reasoning steps by itself or providing specific algorithm/information to look for/questions to answer to define the reasoning steps on your side.
Example: Identifying Mutual vs. Unilateral NDAs
In one of my earlier posts, I explained how AI can struggle with distinguishing whether an NDA is mutual or unilateral. Let’s revisit that example with a reasoning-focused approach.
If you ask:
“Is this NDA mutual or unilateral?”
AI might give a quick answer but miss critical nuances. Instead, prompt it with reasoning:
“Is this NDA mutual or unilateral? Before answering, provide your reasoning. Focus on determining whether each party can act as both a Discloser and a Receiver (mutual) or whether the roles are predefined (unilateral).”
This adjusted prompt encourages the AI to analyze the contract in a specific way and pay attention to the actual question that needs to be answered (whether each party can play both disclosing and receiving roles) which subsequently allows it to accurately answer the original question.
Conclusion
Prompting AI to provide reasoning — and sometimes guiding it by listing steps — has a dramatic impact on answer quality. It’s a simple yet effective way to unlock better insights.
While new advanced models, like OpenAI’s o1 model already employ reasoning under the hood, this technique empowers you to elevate any model’s performance. With practice, you can easily steer the AI in the right direction and achieve the answers you need in just a few iterations.
Start experimenting with reasoning prompts today — you’ll be amazed at the difference it makes!
Stay tuned for more prompting techniques in future posts. Whether you're crafting Markdown prompts or exploring reasoning strategies, there’s always more to learn about getting the best from AI.