Your hub for trends, best practices and resources
542,000+ Visitors Annually!
Tools and Trends
Scrabble pieces spelling out AI DeepSeek

Using DeepSeek:What Happens to the Information You Enter?

What DeepSeek collects, where your information goes, what you can control and what it costs


The chat at chat.deepseek.com and the phone app are free. DeepSeek’s published pricing covers the developer interface only and lists no consumer subscription. There is no equivalent of ChatGPT Plus or Claude Pro to buy. DeepSeek does not publish a message limit for the free chat, and we did not reach one in a day of heavy use. That doesn’t necessarily mean no limit exists.

Costs begin when you use DeepSeek’s API, a separate paid service that developers use to connect DeepSeek to other software — for example, a marking tool. API use is charged according to the amount of text sent to and generated by the model, measured in tokens. As a rough guide, one token is about three-quarters of an English word.

The cost can be very low. At current off-peak rates, sending a 1,500-word essay to V4 Flash and asking for 600 words of feedback would cost about one-tenth of a cent. Doing the same for a class of 40 would cost about four cents. Using the more capable V4 Pro model would cost about 12 cents for the same class set.

During off-peak periods, DeepSeek currently charges US$0.22 per million uncached input tokens and US$0.66 per million output tokens for V4 Flash. The corresponding V4 Pro rates are US$0.66 and US$1.98. During peak periods, these prices are twice as high.

DeepSeek publishes its peak periods in Coordinated Universal Time (UTC). During our tests in late August 2026, those periods translated to 9:00 p.m. to midnight and 2:00 a.m. to 6:00 a.m. Eastern Time on weekdays. Normal daytime use in Ontario therefore fell within the less expensive off-peak period.

These rates took effect on August 16, 2026, when DeepSeek introduced peak and off-peak API pricing. Anyone who priced the API before then would have seen different figures.

Yes. Under DeepSeek’s terms of use, before using its services you should register with an email address or a supported third-party account, and the official app signs you in by email, Google or Apple.

You will also find websites carrying the DeepSeek name that let you use its models without signing up. These are third-party services that provide access to DeepSeek models but are not operated by DeepSeek itself. Their privacy policies, data collection and retention practices may be different from DeepSeek’s.

Nothing in this post about what DeepSeek collects, where data goes, how long it is kept or what settings you can change necessarily applies to those services. If deepseek.com is not in the address bar, you are not using DeepSeek’s own service, regardless of what the page is called.

Everything you type, and more. The privacy policy for all DeepSeek’s tools names your text input, voice input, prompts, uploaded files, photos, feedback and chat history. Automatically, it collects your device model, operating system, IP address, device identifiers, system language, crash reports and performance logs, as well as which features you use. It assigns you a device ID and a user ID and uses them to recognize you across devices and work out your approximate location from your IP address.

It also says the service is not designed to handle sensitive personal data — health, religious belief, racial or ethnic origin and immigration status are among the categories — and asks you not to provide it about yourself or anyone else.

To the People’s Republic of China. DeepSeek collects, processes and stores your personal data on its cloud servers, and the policy that says so covers its apps and websites, not only the developer interface. The company is Hangzhou DeepSeek Artificial Intelligence Co., Ltd. The use of the data is governed by mainland Chinese law, and disputes go to the court where it is registered.

DeepSeek does not publish a single retention period. It says how long it keeps things depends on the type of information, why it processes it and what the law requires. It gives one worked example: your account information, inputs and payment information are kept for as long as you have an account, and longer if you breach the terms.

DeepSeek’s model mechanism disclosure says the pairs used to finetune its models are mostly written by its research team, with a small portion potentially drawn from user input, de-identified first. For users of the developer interface, DeepSeek may use inputs and outputs to develop or improve the service. If you refuse, you can opt out by turning off a setting called “Improve the model for everyone”, a toggle in settings. The privacy policy separately gives you a right to opt out of your personal data being used for model training or optimization.

The limit matters. Turning it off stops your material being used to improve the models. It does not stop collection, storage in China, the retention above or sharing with DeepSeek’s service providers, its corporate group or law enforcement. Those are separate arrangements that the toggle does not touch.

You can delete individual conversations through the settings and copy your history out. You can delete the account yourself, after which DeepSeek says you cannot reactivate it or retrieve any of its content, although it may still retain data required by law. How long a deleted conversation persists in its systems is not known, and the policy does not promise that deleting a conversation from the interface removes every underlying copy.

DeepSeek’s privacy policy states that personal data submitted to its hosted service is collected, processed and stored in the People’s Republic of China.

Two clauses in the terms of use bear on this, and neither is obvious. The first is that by submitting anything, you warrant that you hold all the rights, licences and permissions necessary for DeepSeek to process it. Pasting a student’s essay is a representation on your part that you are entitled to do so.

Whether you actually are is a question about your institution’s rules, not DeepSeek’s, and those rules are what will govern you.

Several universities and colleges have teaching guidance that treats generative AI tools they haven’t approved as raising concerns around privacy, security and student intellectual property. These guidelines say you can’t require a student to accept a third party’s terms or hand over their personal data or intellectual property. The general position on ownership is the unsurprising one: The instructor owns the assignment question and the course materials while the student owns the answer.

The second clause is closer to home. Where output informs a decision with legal or material consequences for a person and the DeepSeek terms name educational decisions explicitly, alongside credit, employment and medical ones it must undergo human review, with the intention that any liability rests with you alone. DeepSeek is advisory, not a decision tool.

An unreleased examination, an unpublished paper or non-public course materials are inputs like any other: collected, stored in China, retained as above, available for training unless you have turned that off. DeepSeek also notes that conversations you share by link, once on the open web, can be picked up by net-crawlers despite its countermeasures.

No. If you use DeepSeek’s website, app or developer interface, the information you enter is sent to DeepSeek and handled according to its privacy and data practices.

There is another option, although it is mainly for institutions rather than individual instructors. DeepSeek makes some of its models available for organizations to run on their own computer systems. In that case, prompts and other material can stay within the organization rather than being sent to DeepSeek.

For example, Germany’s academic cloud provider GWDG hosts a DeepSeek model for its member universities and states that prompts and message contents are not stored on its systems.

There are limitations. The models that organizations can host themselves may not be the same or most current versions available through DeepSeek’s website, and running a large AI model requires significant computing resources. For most individual instructors, using DeepSeek will therefore mean using its online service and sending information to DeepSeek.

Three things change what happens to what you enter: which DeepSeek service you are using, whether model improvement is turned on, and whether the material is yours or someone else’s.

This table summarizes what we found.

Sources checked August 30, 2026: DeepSeek Terms of Use, DeepSeek Privacy Policy, DeepSeek API Pricing and DeepSeek model documentation. All descriptions of DeepSeek’s current arrangements in this post are based on those sources unless otherwise stated.

Tweets from @ContactNorth

Trending Articles

new-pedagogy-1140x400
A New Pedagogy Is Emerging... ...
Changes in society, student expectations, and technology are motivating university and college faculty and instructors...
1140x400-10guidingprinciples2
Ten Guiding Principles for the...
The following ten principles are intended to provide a (far from definitive!) guide for reflecting...
1140x400-microcredential
Ten Facts You Need to Know Abo...
There is no absolute definition for micro-credentials in Canada or anywhere else, in terms of...