DeepSeek Prompt Generator
DeepSeek’s reasoning models behave differently from the rest — they think before they answer, and prompting them like ChatGPT actively makes them worse. Describe your task below and get a prompt built for how they actually work. Free, no account.
Try:
What makes a DeepSeek prompt work
Do not tell a reasoning model to think step by step
This is the big one. R1-style models already reason internally before answering. Bolting "think step by step" on top makes them second-guess a chain they have already worked through, and the answer often gets worse. State the problem cleanly and let it run.
Skip the role-play preamble
"You are a world-class expert…" earns its keep on chat models and does very little here. Reasoning models respond to the problem, not the costume. Spend those words on the actual constraints instead.
Write in the language you want back
DeepSeek is genuinely bilingual — it is strong in Chinese, not merely translated into it. Prompt in Chinese and you will get more natural Chinese out, with better handling of idiom and context than a translated English prompt produces.
Play to its strengths: maths, logic, code
This is where DeepSeek is most competitive with far more expensive models. Give it the whole problem and the constraints in one go — it does better with a complete, well-specified task than with a conversation you steer turn by turn.
The same request, before and after
What people usually type
You are an expert developer. Think step by step and fix my code.
What actually gets a usable answer
This function should return the median of a list of numbers, but it returns the wrong value for lists of even length.
```python
def median(xs):
xs.sort()
return xs[len(xs) // 2]
```
Constraints:
- Must not mutate the caller’s list
- Must handle an empty list without raising
- Standard library only
Give me the corrected function and one sentence on what was wrong. No explanation of what a median is.No role-play, no "think step by step" — just the actual problem, the real constraints, and a clear scope for the answer. That is what a reasoning model wants.
Common questions
Is this DeepSeek prompt generator free?
Yes, with no account needed. The Chrome extension adds one-click prompting inside chat.deepseek.com.
Does this work for both DeepSeek R1 and V3?
Yes, but the advice differs. R1 is the reasoning model — do not ask it to think step by step. V3 is a conventional chat model and behaves much more like ChatGPT, so role and format instructions help there.
Should I write my prompt in Chinese or English?
Write in whichever language you want the answer in. DeepSeek is natively strong in both, so a Chinese prompt gives you better Chinese than an English prompt plus a translation step.
Prefer one-click on every model?
Install the free Chrome extension →