Using Your Mentoring Data in an AI Assistant: Five Questions Worth Asking

A program owner can ask an AI assistant plain questions about live mentoring data, such as which matches have gone quiet or what a mentor should cover next week. The questions that work are narrow and tied to one decision. The data that goes into the chat should be limited to what the person asking is already allowed to see.

Most mentoring programs keep their data in two places: the platform and a spreadsheet. Questions about the program then wait for someone to export, filter and read. A growing number of teams now put an AI assistant in front of that data instead. This post covers five questions that suit that setup, and three things to keep out of it.

What "mentoring data in an assistant" means

An AI assistant that supports the Model Context Protocol can connect to a tool and read from it on request. In MentorStack, the connection is a key that a person generates in their own settings. The help article on connecting AI assistants has the steps. The key carries the same view as its owner. A participant key sees that participant's own matches, meetings and goals. An admin key sees the organization's program data.

That scoping matters more than any single question. The assistant does not get a wider view than the person who connected it.

Five questions worth asking

1. "Which active matches have had no meeting in the last 30 days?"

This is the question a coordinator asks every week and answers by hand. A direct answer saves the export step. The list is a starting point for a nudge, not a verdict. A quiet match can be a paused match, a holiday or a pair that meets without logging it.

2. "Summarize where the spring cohort stands: how many pairs are active, how many have a goal set, how many are waiting on a first meeting."

A cohort summary is a count and a short read. It suits a weekly update to a sponsor. Ask for the numbers first, then ask for a draft of the update. Check the draft against the dashboard before it goes anywhere.

3. "Draft a prep agenda for my next meeting with this mentee, using their goals and open action items."

A mentor can use this the evening before a session. The assistant has the goals and action items the pair already recorded. The draft is a prompt for the conversation. The mentor edits it, and the mentee does not need to see the draft.

4. "Which of my open action items are due this week?"

This one is for a mentee or a mentor. It is a small question, and it is the one people repeat. It replaces opening three pages to find one list.

5. "Suggest three goals that fit this profile and this role."

A mentee can ask for goal ideas based on their own profile. Treat the answer as a list of options. A goal belongs to the mentee, and the mentor should agree to it before it is set.

Three things to keep out of the chat

Free-text notes about a person's performance or health. Mentoring conversations often include private detail. A question about a pair's status needs counts and dates, not the content of what they said.

Decisions about people. Ask the assistant to find, count and draft. Do not ask it to rank people or to decide who stays in a program. A match is a human choice that an admin reviews.

Keys shared across a team. An admin key can change program data. Each person should generate their own key, name it, and revoke it when they stop using it. Treat it like a password.

Check your organization's policy on sending employee data to an AI service before you connect anything. Policies differ, and that choice belongs to your team and your counsel.

What this does not replace

A program still needs a coordinator who reads the signals and decides. The assistant shortens the path from a question to a list. It does not decide who needs a call, and it does not run the program. If your pilot is growing and the weekly questions take more time than the weekly answers, the fix starts with the operating process, then with the tooling.

Where to start

Pick one question from the list above. Run it once a week for a month and note how long the answer used to take. If it saves time, add a second question. If it does not, stop.

If you want to try it on your own program data, the Free Pilot supports up to 10 participants. Reports and export start on the Starter plan, and the connection steps are in the help article.