Do you really need to learn how to write AI prompts?

Do we really need to learn prompt-writing? Isn’t GenerativeAI all about natural language?

These are the questions on my mind this week.

Take ChatGPT, for example. A GenerativeAI model built on a Large Language Model (LLM), it is designed to process and generate human-like text. Natural Language Processing (NLP) enables ChatGPT to interpret, analyse, and generate text in a way that mimics human communication.

The last point, mimicking human communication, is the crucial breakthrough when it comes to human technology interaction. It removes the barrier of needing to learn a separate language (computer programming language) to interface with AI.

So why is prompt-engineering still relevant as GenerativeAI matures?

There are 5 reasons:

1. Not everyone knows how to write clearly or understand how AI processes language
2. For precision and efficiency
3. When repeatability and output control is critical
4. To guide AI toward objectivity and reduce bias
5. To improve AI's handling of complex, multi-step, or technical tasks

Let’s review.

1. Not everyone knows how to write clearly or understand how AI processes language

My background in User Experience (UX) Writing has been a great advantage when it comes to using AI.

This is because prompting for AI is in many ways just like writing for humans.

Here are some parallels between UX writers (including instructional design writers) and Prompt Engineers:

• Clarity & Precision – Both require writing instructions that are easy to understand and act upon.
• Understanding the audience – Both need a deep understanding of the audience (for UX writers, it’s users; for prompt engineers, it’s the AI).
• Iterative Testing – UX writers test and refine copies for usability, prompt engineers refine prompts for better AI responses.
• Guiding Behaviour – UX writing nudges users toward the right action; prompts guide AI toward desired outputs.

To write better prompts, it is helpful to have a basic understanding of how AI processes language.

When given a prompt, AI will do the following:

1. breaks down the input into tokens
2. analyses context based on training data and prior tokens
3. predicts the next mostly likely token step by step
4. generates response based on probabilities

As much as NLP makes ChatGPT sound like a human conversationist, the reality is that AI relies on statistical patterns and probability-based predictions. No true comprehension actually took place.

Just as a UX writer starts with with a clear understanding of the target readers, keeping the above in mind helps formulating prompts that can be best processed by AI.

2. For precision and efficiency

You can learn prompting through iterations. In fact, I believe this is one of the best ways to get to know each GenerativeAI.

Prompt, adjust, repeat for optimal outcomes. In time, you will have a good collection of useful prompts.

However, iterative prompting takes time. Therefore, I understand why there are times you may want to learn from or copy others’ tried-and-tested prompts. This is particularly true when if your AI work involves specific domain knowledge.

3. When repeatability and output control is important

AI does not give fixed answers, as its outputs are probability-driven. If you desire consistent outcomes, it is important to structure your prompts to optimize repeatability.

Take these prompts for example:

“Define leadership in one sentence. Use clear and concise language." will yield many different one-sentence answers.

Define 'leadership' in exactly 15 words: 'Leadership is...' " is more limiting and forces a nearly identical output each time you use this prompt.

4. To guide AI toward objectivity and reduce bias

AI models like ChatGPT are trained on vast amounts of publicly available text hence unfortunately do inherit biases, misinformation, and overgeneralization.

Clever prompting can help mitigate this issue.

For instance, instead of asking “What is the best approach to X?”, you can ask AI to “Compare different approaches to X and discuss the pros and cons”.

Similar prompts (suggested by ChatGPT) to help avoid biases:

• “Provide research-backed information on X with sources”
• “Give an unbiased explanation of X from several viewpoints”
• “Explain this from the perspective of a scientist”

Just like formulating questions in an interview, your prompt can guide AI provide the most neutral perspective.

5. To improve AI's handling of complex, multi-step, or technical tasks.

As per point 1 above, if you don’t write clearly, AI will provide sub par outcomes. The situation gets progressively worse with complex requests.

I recently built a webapp using Cursor, an AI code editor. I could have told Cursor everything I wanted in one big request - and let it to take over and generate codes. However, that actually is a penny-wise pound-foolish approach. AI, while powerful in vast data processing, struggles to break down large, multi-step requests into logically structured segments on its own.

Like a child, AI benefits from step-by-step guidance. Humans are the experienced adults who are superior in structuring workflows, defining priorities, and ensuring logical progression.

Final Thoughts

You don’t need to formally learn how to prompt to make use of Generative AI. Use it often, use it enough, and over time you’ll naturally develop a good sense of what works.

That said, there are plenty of good reasons to read up on the topic and learn from skilled prompt engineers.

Much like you don’t need to attend culinary school to use a knife, professional chefs master their tools for a reason - refined skills leads to better results.

Happy Prompting!

For more articles, visit and subscribe my Substack.