What MeritForge shares with AI assistants
People can connect the AI assistant they already use (ChatGPT, Claude, or Gemini) to MeritForge. They can then ask it what needs them this week, or have it draft a Spark, a check-in, or an objective update. This page is for the People team, IT, and security reviewers. It explains what the assistant receives, what it never receives, and what your organization controls.
How it works
- Each person connects their own assistant with their MeritForge sign-in, and then chooses what it may do in each organization in Assistant connections. Nothing is shared until they do.
- The assistant only asks; MeritForge decides. Every request is checked again against the person’s current role, your organization’s settings, and their choices. Someone who leaves the company or loses a role loses that access immediately.
- There’s a step-by-step guide in MeritForge for each assistant: open Guides & tutorials and choose Connect your AI assistant.
- Nothing is sent or saved from a chat message. The assistant can only prepare a draft. The person reviews it on an approval card or a MeritForge page and approves it there. Drafts expire after 30 minutes.
What the assistant can receive
Only for the organizations and permissions the person turned on, and only if your organization allows them:
| Permission | What the assistant receives |
|---|---|
| Read context (always needed) | The person’s organizations, their own Sparks balance, and your company values, so it can draft valid Sparks. |
| Prepare Sparks | Names of the colleagues or teams the person mentions, to find the right people. The Spark itself is sent only after the person approves it. |
| Read my Growth | The person’s own Your week items, their own objectives and progress, their next 1:1 (shared agenda and open commitments), and their current check-in period. |
| Read my team’s Growth | For managers: the open items that need them for their team, with people’s names and how long each has waited. Off by default, because it shares colleagues’ information. |
| Submit check-ins | Drafts a check-in from what the person tells the assistant. Mood is used only when the person states it; it’s never guessed. |
| Update objective progress | The objective’s current and proposed values, to show the change before the person approves it. |
What the assistant never receives
- Mood history, mood notes, or past check-ins, the person’s own or anyone else’s.
- 1:1 notes, including the person’s own private notes.
- Pulse answers, feedback text, or anything from anonymous surveys.
- Other people’s balances, check-ins, or objectives they don’t share.
- Organizations the person didn’t connect, or in which your organization doesn’t allow assistants.
Bear in mind that whatever a person types into their assistant is seen by that assistant’s provider. That includes a mood or a win they dictate for a check-in. MeritForge doesn’t control that part.
What your organization controls
Owners and admins can do all of this in Admin → Integrations → AI assistants:
- Allow or turn off AI assistant connections for the whole organization.
- Choose the permissions people may give their assistant. Team Growth is off until you turn it on.
- Choose the assistants people may use: ChatGPT, Claude, Gemini, or only some of them.
- See who is connected, with which assistant and permissions, and revoke a connection for your organization.
- See the activity log: every request an assistant made in your organization, with who, which assistant, which action, and how it ended. What was asked and what came back are never stored. The log is kept for one year.
Changes apply on the assistant’s next request. Turning a permission off doesn’t erase people’s choices, so turning it back on restores them.
The People team can view the connected list but can’t change anything.
Who processes the data
The assistant is a service the person (or your company) chooses and signs in to under their own agreement with OpenAI (ChatGPT), Anthropic (Claude), or Google (Gemini). MeritForge doesn’t hire these providers to run MeritForge, so they are not MeritForge subprocessors. What MeritForge returns goes to them only at the person’s request, within the limits above. If your company has its own agreement with one of these providers (for example ChatGPT Enterprise, Claude for Work, or Gemini Enterprise), that agreement governs how they handle it.
Gemini
Gemini connects through custom connectors in Gemini Enterprise and Gemini Business. Your Google Workspace admin adds MeritForge once for the company. Write to support@meritforge.io for the connection details.
Questions from a security review? Write to support@meritforge.io.