How to write better ChatGPT prompts: a guide for beginners

Most disappointing answers from ChatGPT come from short prompts. "Write an email to a client about the delay" gives the model almost nothing to work with, so it fills the gaps with the most average choices available: an average tone, an average length, an average client. The result reads like a template because, in effect, it is one.

Better prompts are not clever. They are specific. Four pieces of information make the largest difference.

1. Who the output is for

Name the reader and what they already know. "A client who is annoyed, has been waiting two weeks, and does not care about our internal reasons" produces a very different email from "a client". The model adjusts vocabulary, length and tone to the reader you describe. If you describe no one, it writes for no one in particular.

2. What the output will be used for

State the purpose. An email meant to keep a client calm is not the same as one meant to document the delay for a later dispute. Say which it is.

3. The material it needs

The model knows nothing about your situation unless you tell it. Paste in the relevant facts: the original deadline, the cause of the delay, the new date, what you can offer. This is the step beginners most often skip, and it is the one that matters most. A plain prompt with good context beats an elaborate prompt with none.

One caution. Before pasting in work material, check that your organisation permits it and that the material contains nothing confidential. The guide to using ChatGPT safely at work covers what to leave out.

4. What a good result looks like

Describe the format and the limits. "Under 120 words. No apology in the first sentence. End with the new date and one concrete offer." If you have an example of the style you want, include it. Examples communicate tone far better than adjectives such as "professional" or "friendly", which the model interprets in the blandest way possible.

A before-and-after example

• Before: "Write an email to a client about the delay."
• After: "Write an email to a client who has waited two weeks for a website update and is frustrated. The delay was caused by a third-party payment provider. The new launch date is 14 October. We can offer one month of free hosting. Keep it under 120 words, do not open with an apology, do not blame the provider by name, and end with the new date and the offer."

The second prompt takes forty seconds longer to write. It usually saves several rounds of correction.

Then treat the answer as a draft

Even a good prompt rarely produces a finished result on the first attempt. Tell the model what is wrong and ask it to change that one thing: "Shorter, and less formal in the second paragraph." Two or three focused revisions are normal. When a prompt works well, save it. A tested prompt for a task you repeat every week is one of the most useful things a beginner can build.

A plain prompt with good context beats an elaborate prompt with none.

What to ignore

Lists of "magic" prompts and secret phrases promise more than they deliver. Phrases such as "act as a world-class expert" have a small effect at best. The four elements above do almost all of the work.

The free Course Parse course teaches this in modules 3, 4 and 6, and its final project has you build three tested prompts for your own weekly work. For a deeper treatment, including structured outputs, examples, evaluation and prompts that other people can reuse, see Prompt Engineering.