How to write high-quality AI prompt words

More and more people are starting to use AI. A common problem is that they know they need "good cue words" but don't know what counts as "good". Communicating with AI can sometimes be like taking a stroke of luck: typing a piece of text casually may yield good results; Repeatedly modifying for a long time, the output is still incorrect.

This article does not discuss grand concepts such as "prompt word engineering", but rather breaks down my habit of writing prompt words. They can be summarized into four key points.

Let's be clear: AI cannot guess everything. Many people think that AI is smart enough, so there is no need to go into detail about the requirements. The actual situation is that what you didn't say, AI will complete it on its own; If you express yourself vaguely, AI will guess. If you guess wrong, you will feel that AI is not easy to use, and AI doesn't know why you are not satisfied.

The essence of writing prompt words is to translate your expectations into instructions that AI can execute. The more you want to save time and let AI "watch and act", the more random the results will be. The following four habits come from repeated attempts. They may not be advanced, but they can increase the probability of each output falling within the expected range.

An effective approach is to construct a structure for the prompt words. Most people may only write one sentence when writing prompts, 'Help me write a weekly report'. AI can also write such prompt words, but how far they differ from the results you want depends entirely on luck.

I usually break down the prompt words into four modules. Role setting, used to tell AI who it is, for example, 'You are the operations manager of a toB product, reporting to the product director'. Task description, used to tell AI what to do, such as "Write a weekly report of about 400 words based on the 5 work records of this week". Constraints are used to tell AI what not to do, such as "don't write clich é s, don't use nonsense like 'successfully completed this week', data should be retained, but don't list it in a separate column". Output format, used to tell AI what the final content should look like, for example, "divided into three subheadings, this week's focus, next week's plan, and risks that need attention, with a maximum of 3 items per section".

Before writing prompt words, think carefully about these four modules. Even if it only takes 30 seconds longer than before, the result may be different. At first, it may seem troublesome, but as you write more, you will realize that this is actually organizing your own thoughts.

Another easily overlooked aspect is providing AI with sufficient context to prevent it from guessing. This is the place where I have stepped on the most pitfalls. Before, when I wrote prompt words, I always thought that clear task writing was enough. Later, it was discovered that task description only accounted for half of the effect, with the other half being context.

For example, if you ask AI to write an article on official account, the topic is how to write prompt words. Of course it can write. But it is highly likely to be written as a general science popularization article, starting with the phrase 'In the era of AI, prompt words are becoming increasingly important'. I don't want this outcome.

So I will add who the target readers are, not everyone, but professionals who have already used AI but find it difficult to use. I will explain the tone, like a colleague teaching, not like a teacher training, using short sentences and colloquial language. I will tell AI what has been written before, what the first few articles in this series have talked about, and which article this one is. I will also clarify what cannot be written, not reviews, not hot topics, and not selling anxiety.

The more context you provide to AI, the fewer parts it can guess. If you guess less, the deviation will be small. A simple criterion for judgment is that if the output of AI makes you feel "okay, but not what I want", it is likely that the context is not given enough.

Constraints are often more important than task descriptions. When I first started using AI, all my attention was focused on what it was supposed to do. Later on, it was discovered that many times, what not to let it do is more important than what to let it do.

For example, I asked AI to help me write a manuscript. When not mentioning constraints, it will add a bunch of "must say", "can be seen", and "in the context of today's era", which are pronounced incorrectly.

After adding constraints, I will make it clear not to use "must say", "can see", "in today's era", "worth noting", and not to write a summary paragraph, with each paragraph not exceeding four lines and the entire text not exceeding two emoticons. The essence of these constraints is to include the phrase 'no, this is not right' in the prompt when you revise the manuscript.

If you don't say 'no', AI won't know it's not possible. It will only follow its default habits. My habit is to place the constraints after the task description, starting with "do not", one line at a time. Don't write 'please keep it concise', write 'no more than four lines per paragraph, no more than two emoticons'. The more specific, the better. 'Please be concise' is nonsense, AI doesn't know what simplicity is.

The output format is also worth explaining separately, and the more specific the better. This is a point that many people overlook. If you ask AI to write an article, it will default to outputting paragraphs of pure text. You want subheadings, you want lists, you want code blocks, AI doesn't know, you have to say it yourself.

I usually write the output format as follows: the first paragraph should not exceed three lines; The main text is divided into three to four sections, with each section using secondary subheadings; Each section should not exceed five paragraphs; Bold key sentences; Wrap up with a short sentence at the end, no more than two lines.

Why is it so specific? Because you don't explain the format clearly, the default format of AI is likely not what you want. Its default is to write an essay, and what you want is a formal article. The more specific the formatting instructions, the closer you get to a state where you can use it directly, without having to spend time adjusting formats, changing segments, or adding titles.

Ultimately, these four habits may seem simple, but not many people can persist in using them. Most people still slack off when using AI, typing 'help me write a weekly report' and sending it out, then spending fifteen minutes revising it back and forth. Actually, if you spend thirty seconds organizing the prompt words, you don't have to waste fifteen minutes.

Ultimately, it's about speaking clearly. We can be vague when talking to people, we can add a sentence, but not with AI.