Check what the assistant reads and sends

Numbered sources, what each module contributes, your organization's AI settings, and reused answers.

In short

The assistant reads only what you could open yourself, in the workspace shown in the header, plus the Help guides. Every answer lists its sources by number so you can check them. Your organization's admins decide whether AI is used there, what it may read and whose key answers.

Before you start

• Who: everyone sees their own sources. Only an organization admin changes the organization's AI settings.
• Plan: AI is part of every plan, Free included. On Free the platform's key answers questions about how {{brand}} works. For questions about your own work, add your own key (it is free to add) or choose a paid plan (see /help/ai/answers-left). An organization's own AI key comes with the Business and Enterprise plans.
• First: nothing. Open the record the question is about before you ask.

Steps

1. Read the numbers in the answer, such as [1] and [2].
2. Under the answer, open "Sources" and press a number to open that record or Help guide. Guides are marked Help.
3. If the record you expected is missing, the assistant did not see it: open that record and ask again from there.

When your records and the guides do not hold the answer, the assistant says the sources don't cover this instead of guessing, and lists no sources. Open the record the question is about and ask again, or press "Ask about my workspace" when the answer came from Help alone.

A module switched off for you adds nothing.

Your organization's AI settings

In an organization, its admins decide how AI is used there, under the organization's settings, "AI in" the organization. The settings apply to everyone working in that organization, from the next question. Everyone else can read them but not change them.

• "Use AI in this organization": when it is off, the assistant and every AI action are unavailable to everyone working there.
• "Data AI may read": a module that is unticked adds nothing to any answer, and its own AI actions are unavailable.
• "Hide emails and phone numbers": they are taken out before anything is sent to a model, unless a question asks for them.
• "Keep AI in the European Union": everything is processed there, and personal keys and the organization's key are not used while it is on. It can be turned on only where the platform has a deployment in the European Union.
• "Whose key answers": by default a member's own key first, then the organization's own key (Business and Enterprise), then the platform's AI. Admins may choose the organization's key only, or the platform's AI only, so the organization's data does not reach members' own AI accounts; everyone working there is told once. Jobs' candidate features always use the platform's AI.

The same tab lists every AI call made in the organization for its admins: who made it, what for, the model, whose key answered and what it cost. It never shows the question or the answer.

Reused answers

A question about how {{brand}} works is answered from the guides alone, and that answer can be reused for everyone. A question about your workspace can be reused until the end of the day (UTC) by someone in the same workspace asking the same about the same records. A reused answer says when it was first given and that no question was used. "Ask again" gets a fresh one. An answer written with your own key is never kept for anyone else.

Good to know

• Answering a question never changes data. Only Do makes anything, and only the steps you confirm: see /help/ai/ask-or-do.
• It never answers from general knowledge. If your records and the guides do not hold the answer, it says so.
• It never stores the conversation. Only the fact that a call happened, and its size, is recorded.
• It never sees your key. The key reaches the provider; the model is not told it.
• Answers are short unless you ask for detail: a count is one sentence, a list is the list.
• In a long conversation the latest turns are sent in full and the earlier ones only as the questions you asked, so follow-ups stay quick and cheap.
• Each module has a few seconds to say what it holds; one that is slow is left out of that answer rather than holding it up.
• A follow-up to a question about how something works keeps reading the Help guides, until you ask about your own work.
• Asking for the same draft again - acceptance criteria, estimation questions, a vacancy, an interview kit, your Home brief - within a week, with nothing changed, shows the kept draft and costs no AI.

Your question and the records it reads go to your provider Each question sends the question, the earlier turns and the records and guide sections the assistant selected to the provider that answers, under that vendor's terms. Never paste an API key, a password or a token into a question.

If something goes wrong

• "Only an organization admin can change its AI settings.": ask an admin of the organization.
• "Open an organization first - its AI settings belong to the organization.": switch to the organization in the workspace menu.
• "AI in the European Union is not available on this deployment yet, so it cannot be switched on.": leave it off, or ask the platform's operator.
• "An organization's own AI key comes with the Business and Enterprise plans.": the organization's plan has no organization key.
• "Your organization answers AI only with its own key, and none is set up": an organization admin can add it in Organization settings → AI.

Related

• /help/ai/asking-the-assistant
• /help/ai/saved-and-session-keys
• /help/organizations/settings-and-governance