The fastest way to know whether DeepSeek fits your work is to use it on your work. This lesson gets you from zero to a finished task in about fifteen minutes, in the free app, without touching the API. Keep a real document, spreadsheet or bug from your own week handy — a toy example will not tell you anything.
Set up and first task
- 1
Create an account
Go to chat.deepseek.com and sign up with an email address or a supported single-sign-on option. The mobile apps use the same account. There is nothing to pay and no card to add.
- 2
Tour the interface
The layout is the familiar three parts: a sidebar of past chats on the left, the conversation in the middle, and a prompt box at the bottom. Above the box you will find the two toggles that matter — DeepThink and Search — plus an attachment button for files and images.
- 3
Leave DeepThink off to start
DeepThink switches the reply to the reasoner model. It is slower and, on the API, much more expensive per answer. For drafting, rewriting, summarising and simple code it adds little. Turn it on only for maths, logic puzzles, or a coding problem the normal model got wrong.
- 4
Turn Search on only for current facts
Search adds live web results to the prompt. It helps with anything time-sensitive and hurts when you want the model to work from your pasted text alone, because it can drag in irrelevant pages. Off by default; on when the date matters.
- 5
Do one real task
Paste something from your actual work and give a specific instruction. For example: "Here is a 900-word draft. Tighten it to 600 words, keep the three examples, and do not change the numbers." Or attach a CSV and ask for the three most surprising things in it.
- 6
Judge it like a colleague's work
Read the output for accuracy, not fluency. Did it keep the numbers? Did it invent a source? Ask a follow-up that pushes back — "which of those claims are you least sure about?" — and see how it responds. That tells you more than any benchmark.
- 7
Decide about the API later
You only need an API key when you want DeepSeek to run without you — a script, an IDE plugin, an integration. If today's task was interactive, you are done and your cost is $0. Lesson 7 covers keys when you need one.
Settings that matter
The app has few settings, which is a feature. The ones worth a minute: the interface language, whether chat history is kept, and on mobile whether the app can send notifications. There is no model picker beyond the DeepThink toggle — the app always uses the current recommended chat or reasoner model.
What a good first result looks like
A good first session ends with one output you would actually use, and one clear miss you noticed. The miss is important: it shows you where you will need to check, or where DeepThink or a different tool is worth it. If the app produced nothing useful on a real task, that is also a valid result — it tells you before you have spent anything.
If you already pay for another assistant, run the same task there and compare. Our DeepSeek vs ChatGPT vs Claude page has task-by-task notes, but your own document is the benchmark that matters.
Knowledge check
When should you switch DeepThink on in the DeepSeek app?