· ai-assistants

OpenAI vs DeepSeek

On what the policies say, there's nothing between them. Both are graded D.

Grades read the plan most people are on. Each company's full entry has every plan, every clause and the archived copy it was read from.

OpenAI

ChatGPT trains on your content by default on Free, Plus and Pro, with an opt-out that binds new conversations only. Business and API do not train by default.

Verdict

By default: Trains on you by default
The way out: A self-serve opt-out exists
What D means: On by default. An opt-out exists, but you have to go looking for it.
Last checked: 5 Oct 2026

Read the full verdict →

DeepSeek

DeepSeek trains its models on what you type. An opt-out is listed, but only among rights that "may be available" depending on where you live, and you have to email for it.

Verdict

By default: Trains on you by default
The way out: Opt out by contacting support
What D means: On by default. An opt-out exists, but you have to go looking for it.
Last checked: 13 Aug 2026

Read the full verdict →

OpenAI: every plan

OpenAI · what each plan can refuse · checked 5 Oct 2026

2 PLANS

Plan

Exposure

Trains

Opt-out path

Retention

De-identified

Human review

Free, Plus & Pro (ChatGPT consumer)

51 / 100

YES

Self-serve toggle

Not stated

Unknown

How the score is made51 / 100 · the rubric

Default32 / 50

On by default, with a way out

Way out15 / 33

One switch, but the policy never says which way it ships

Paywall0 / 4

No paywall on the control

Scope0 / 4

One kind of data is in scope

Human review2 / 4

The policy does not say whether people read it

De-identified2 / 2

Nothing is said about de-identification

Business, Enterprise & API

18 / 100

Nothing to switch off

Not stated

Unknown

How the score is made18 / 100 · the rubric

Default13 / 50

Training happens only if you opt in

Way out0 / 33

Nothing is on, so nothing needs refusing

Paywall0 / 4

No paywall on the control

Scope1 / 4

2 kinds of data are in scope

Human review2 / 4

The policy does not say whether people read it

De-identified2 / 2

Nothing is said about de-identification

EVERY CELL ABOVE IS SOURCED FROM A QUOTED CLAUSE OR A DOCUMENTED SEARCH · SEE THE EVIDENCE ABOVE

EXPOSURE IS SCORED 0–100 ON ONE AXIS, HOW HARD IT IS TO STOP THIS PLAN TRAINING ON YOU. THE GRADE IS THE BAND IT FALLS IN: A 0–10 · B 11–25 · C 26–45 · D 46–60 · E 61–80 · F 81–100. HOW THE SCORE IS SET

Free, Plus & Pro (ChatGPT consumer)TRAINS: YES

OPT-OUT: Self-serve toggle
RETENTION: Not stated
EXPOSURE: 51 / 100
DE-IDENTIFIED: Not stated
HUMAN REVIEW: Unknown

Business, Enterprise & APITRAINS: NO

OPT-OUT: Nothing to switch off
RETENTION: Not stated
EXPOSURE: 18 / 100
DE-IDENTIFIED: Not stated
HUMAN REVIEW: Unknown

DeepSeek: every plan

DeepSeek · what each plan can refuse · checked 13 Aug 2026

1 PLAN

Plan

Exposure

Trains

Opt-out path

Retention

De-identified

Human review

All accounts

57 / 100

YES

Contact support

Not stated

Unknown

How the score is made57 / 100 · the rubric

Default32 / 50

On by default, with a way out

Way out18 / 33

Refusing means writing to the company

Paywall0 / 4

No paywall on the control

Scope3 / 4

4 kinds of data are in scope

Human review2 / 4

The policy does not say whether people read it

De-identified2 / 2

Nothing is said about de-identification

EVERY CELL ABOVE IS SOURCED FROM A QUOTED CLAUSE OR A DOCUMENTED SEARCH · SEE THE EVIDENCE ABOVE

All accountsTRAINS: YES

OPT-OUT: Contact support
RETENTION: Not stated
EXPOSURE: 57 / 100
DE-IDENTIFIED: Not stated
HUMAN REVIEW: Unknown

Every grade on this page is the one on the company's own entry, read from a dated, archived copy of its policy. Nothing here is computed separately. How grades are set · All comparisons