Short answer: Instinct's current Privacy Policy says certain user information may be used to evaluate, fine-tune, and train AI models. It also describes a settings-based opt-out, but that opt-out is prospective and has exceptions.
Google Workspace data is treated differently: Instinct says information received directly from Google Workspace APIs is not used to evaluate, fine-tune, train, or improve AI models.
What does the Privacy Policy say?
Instinct says some information collected through use of the service may be used to develop and improve its products and technologies, including model evaluation, fine-tuning, and training.
The Terms contain similar language for user materials, subject to the privacy policy and opt-out rights.
That means the accurate answer is not "Instinct never trains on user data."
How does the training opt-out work?
Instinct says users can opt out through settings.
The policy describes the opt-out as applying on a go-forward basis. That means it affects future model-training use rather than retroactively reversing every use that may already have occurred.
The policy also says models previously trained, fine-tuned, or improved on information before the opt-out may continue to be used.
Is there a safety-review exception?
Yes.
Instinct says information flagged for safety review may still be used for model training even after the user opts out, for purposes such as harmful-content detection, policy enforcement, or safety research.
That is an important limitation and should be part of any complete explanation of the opt-out.
What about Google Workspace data?
Instinct makes a specific exception for data received directly from Google Workspace APIs.
The policy says that Workspace API information is not used to:
- evaluate AI models;
- fine-tune AI models;
- train AI models;
- improve AI models;
- serve ads.
That includes Workspace services such as Gmail, Calendar, Drive, Docs, Sheets, Slides, and Tasks when the data comes through the Workspace APIs.
See Instinct AI Google Workspace for the full integration breakdown.
Does disconnecting an integration remove training data?
Disconnecting and deleting are separate controls.
Instinct's policy says disconnecting a third-party integration does not automatically delete data already collected from that integration.
A user who wants removal should follow the applicable deletion process as well.
See How to Delete Instinct Data for the documented steps.
Does the opt-out erase previous models?
No such guarantee appears in the current policy.
Instinct explicitly says previously trained, fine-tuned, or improved models may continue to be used.
That is why the most precise wording is:
the opt-out restricts future training use, subject to the safety-review exception; it is not a retroactive model erasure promise.
What should privacy-conscious users do?
Review:
- which integrations are connected;
- what information each integration exposes;
- the model-training setting;
- whether the task needs the data being shared;
- deletion controls for indexed data.
Also remember that Google Workspace has more restrictive training rules than the broader service.
"Uses data for training" needs a more precise reading
The important question is not simply whether the word "training" appears in a privacy policy.
Users should separate at least four issues:
- what information the service can access;
- what information may be used to evaluate or improve models;
- whether a specific data source has a special restriction;
- whether the user can opt out or delete previously collected information.
Those questions can have different answers.
Google Workspace data is a special case
Instinct's privacy policy describes separate treatment for information received through Google Workspace APIs.
That means a blanket statement such as "Instinct trains on all of your Gmail data" is too broad. The source and route by which data enters the service matter.
If Gmail, Calendar, Drive, or another Workspace service is important to your decision, read the current first-party policy rather than extrapolating from the rules for general user information.
Opting out, disconnecting, and deleting are different actions
These controls solve different problems.
Opting out of model improvement or training changes how eligible information may be used going forward under the current policy.
Disconnecting an integration removes or limits future access to that connected service.
Deleting data or an account concerns information that has already been stored or associated with the service.
Do not assume that performing one automatically performs the others.
See How to Delete Instinct Data for the deletion distinction and Instinct AI Permissions for access control.
What should privacy-conscious users do?
Use a task-by-task approach.
Before connecting a service, ask what the task actually requires. If the task only needs a date and location, broad mailbox or payment access may be unnecessary. If it needs an order receipt, share the minimum relevant record rather than treating full access as the default.
Also recheck the policy when the product changes. Instinct is developing rapidly, and privacy guidance can become stale faster than a generic AI explainer.
What this page can and cannot conclude
We can summarize the current policy and distinguish documented categories. We cannot prove how every internal model-training pipeline operates beyond what reliable public evidence supports.
That is why this page uses dated policy language rather than claims such as "your data is never used" or "all your data is used."
Frequently asked questions
Does Instinct train its AI on user data?
Its policy says some user information may be used for evaluation, fine-tuning, and model training.
Can I opt out?
Yes. Instinct describes a settings-based opt-out.
Is the opt-out retroactive?
The policy says it applies on a go-forward basis and previously trained models may continue to be used.
Is Gmail data used for training?
Instinct says data received directly from Google Workspace APIs is not used to train or improve AI models.
Can safety-reviewed data still be used?
Yes. The policy describes a safety-review exception.