AI settings and what data is used

Two questions come up about AI in a CRM, and both deserve a direct answer: what does it read, and what is switched on by default?

This covers both, including where the answer is 'off unless you turn it on'.

Who this is for

Admins deciding what to enable, and anyone asking what the AI has access to.

Before you start

• Admin access to AI settings.
• Your own organisation's position on AI use — this describes the controls, it does not make the policy.

AI answers on the help center are off by default
The help center can answer visitor questions with AI, and that feature is deliberately opt-in and switched off unless you enable it. Nothing starts answering your customers automatically. The same posture runs through the AI surfaces generally: capabilities that act on their own are opt-in rather than assumed.

Understand and control it

1. Know what a request reads
   AI features work from the thing you are on — the email thread, the record, the document you supplied. That context is what makes the output useful rather than generic.
2. Control the agent's knowledge explicitly
   The AI agent answers from sources you connect: uploaded files, question-and-answer pairs, your help center, or a website crawl. It is not quietly reading everything.
3. Understand the permission boundary
   An agent acting through the tool surface inherits the permissions of the account it is connected as, checked on every call. It cannot reach data that account could not.
4. Leave help-center AI answers off until you mean it
   It is off by default. Turn it on when you are satisfied your content is good enough to answer customers unsupervised.
5. Correct wrong answers explicitly
   Adding a question-and-answer pair with the right answer is the direct way to fix a specific mistake, rather than hoping it improves.
6. Confirm what is enabled in your workspace
   Availability differs. Check your own settings rather than assuming a feature described anywhere is switched on for you.

What you get

• A clear account of what AI reads when you use it.
• Explicit control of the agent's knowledge sources.
• A permission boundary the agent cannot exceed.
• Customer-facing AI that is off until you deliberately enable it.

Frequently asked questions

Does the AI read my whole CRM?

No. Assistant features work from the record or thread you are on, and the agent answers from the knowledge sources you connect. A CRM-wide knowledge kind exists in the schema but has no implementation and is refused.

Will AI answer my customers without me enabling it?

No. AI answers on the help center are opt-in and off by default.

Can an AI agent see data I cannot?

No. It inherits the permissions of the account it is connected as, and that is re-checked on every call rather than assumed at connection time.

How do I fix a wrong AI answer?

Add a question-and-answer pair with the correct answer. That is the most direct correction, and it works for things not written down anywhere else.