10 commonly used DeepSeek universal prompt word templates | Tips

When DeepSeek was asked to write a short video script, it received an essay written by an elementary school student. When DeepSeek was asked to write a product details page copy, AI wrote a product manual. When DeepSeek was asked to analyze sales data, AI only produced a bunch of useless tables.

It’s not that DeepSeek is not powerful enough, it’s that your prompt words don’t hit the “Ren and Du Meridians” of AI! The most commonly used technique for writing prompt words is to include role + task + constraint + output specification. 

The wrong way to write prompt words, such as "Help me write a product copy." Although one of the most powerful abilities of AI is that it can answer almost any question , at this point even the most powerful AI will basically start making up product copy, and may even default to the product type you need.

We use AI not to get answers to questions, but to get answers that are useful to us.

The correct questioning technique (writing prompts) for the above case should be like this: As a senior e-commerce copywriting expert, write a Xiaohongshu tweet for the new hyaluronic acid mask you sell, highlighting the "first aid for staying up late + primer before makeup" function. The content is divided into three modules: pain point scenarios, product selling points and user reviews.

Another secret to improving the quality of AI output is scenario-based questioning, error correction instructions, and role-playing.

For example, a wrong way to ask a question is "Compare gasoline cars and electric cars" , while a correct way to ask a question is "Use a comparison table to show the starting principles, applicable scenarios, and advantages and disadvantages of gasoline cars vs. electric cars . "

If you want the AI ​​to answer questions that are easy to understand, don’t include too many professional terms. The wrong way to ask questions is “Simply explain XX without any professional terms.”

Instead, you should ask questions like "Explain XX in plain language" or "Explain XX in language that elementary school students can understand". The answers given in this way are easier to understand.

Another common technique is to give the AI ​​a role. For example, if you want to learn Newton's first law, you can ask the wrong question " Explain Newton's first law . "

Asking this question might get a better answer: "If you were a high school physics teacher, explain Newton's first law to me in layman's terms."

Finally, there is a tip that is particularly easy to overlook. The prompt words you write are vague or mix up multiple tasks. It is like asking an intern to do a task without clear assignments or tasks. The result is likely to be not very good. The same is true for AI.

Try to describe the problem clearly, avoid using vague prompts, and try not to have multiple unrelated tasks in one question.

An example of an error in which multiple unrelated tasks appear in one question is "What are the advantages of new energy vehicles over traditional fuel vehicles? And write an operational plan for selling new energy vehicles." This question can be broken down and asked to the AI ​​twice to gradually generate the questions, which is likely to produce better results.