Microsoft Copilot and Student Data Privacy: What Teachers Should Know

A teacher weighing Microsoft Copilot for classroom use is really weighing several different products at once: Copilot Chat, the Teach lesson-planning assistant, Reading Coach, and the Study and Learn Agent, and each one touches a different slice of data under a different set of rules. Treating "Copilot" as one single tool with one single privacy posture is the most common way an otherwise careful rollout gets more complicated than it needs to be.

Quick Answer: Different Microsoft 365 Education surfaces touch different data. Copilot Chat requires a student's Entra ID ageGroup to be set to 13-or-older; the Teach lesson-planning assistant needs only a teacher's topic and grade level, no student data at all; and Reading Coach uses a separate, on-device voice-processing model available to younger students under its own privacy terms. This guide maps each surface to what it actually touches and the day-to-day habits worth building around it. The age-eligibility and Data Protection Addendum question is covered in Is Microsoft Copilot COPPA Compliant?; the content-filtering question is covered in Is Microsoft Copilot Safe for Students?. This guide focuses on the practical workflow those two leave open.

That guide-to-guide structure exists for a reason. A district's legal and safety review only has to happen once, but a teacher's day-to-day decisions, which surface to open for which task, what to paste into a prompt, whether a specific feature needs a signed-in student account, repeat every week. This is the guide for that second layer.

What This Guide Covers and What It Doesn't

If your question is "is this legal" or "is this content-filtered," start with the two companion guides linked above; they cover the Entra ID age-gate mechanics, the Data Protection Addendum, and the content-safety filtering layer in full. This guide picks up from there and stays on a narrower, more practical question: once a school has cleared Copilot for use, what does a genuinely privacy-conscious weekly workflow actually look like, surface by surface.

Neither companion guide walks through the day-to-day mechanics of Teach, Reading Coach, or the Study and Learn Agent in this kind of surface-by-surface detail, since their focus is the age-gate and content-filtering questions that apply mainly to Copilot Chat itself. Those three products are exactly where this guide spends most of its time.

Context worth keeping in mind while reading: RAND's 2025 survey found roughly a quarter of U.S. teachers already used some form of AI tool for instructional planning during the 2023-2024 school year (RAND, 2025). A district that has already enabled Microsoft 365 Education is often the fastest on-ramp into that habit, since the tools sit inside software teachers already open every day.

Mapping the Surfaces to What They Actually Touch

Microsoft 365 Education bundles several distinct AI-assisted tools under the "Copilot" umbrella, and each one has a different data footprint and a different eligibility rule. Knowing which is which is the single most useful thing a teacher can learn before a first rollout.

That table is worth returning to before enabling any new feature tenant-wide, since assuming one eligibility rule applies to every Copilot-branded surface is the most common source of confusion in an otherwise careful rollout.

The Teach Feature: Lesson Planning Without Student Data in the Loop

Teach launched in October 2025 and rolled out through early 2026 as a single interface inside the Microsoft 365 Copilot app for building a standards-aligned lesson plan from a topic and grade level, supporting standards from more than 35 countries (Microsoft Education Blog, 2025). From a privacy standpoint, it is the lowest-exposure entry point on the list above, because it never needs a student's name, work, or account at all.

A teacher types in a subject, a grade band, and a standard; the tool returns a draft lesson plan that can be adjusted for reading level, tone, or difficulty, and differentiated for a specific class's ability range, entirely through prompts inside the same interface. That workflow sits closer to a planning tool than to a student-data tool, which is exactly why it is worth introducing first in a new rollout, before any feature that touches an individual student's work or identity.

For teachers who want that same no-student-data approach to extend past lesson-plan text into actual handouts, EduGenius offers a comparable capability across a wider set of formats: a teacher configures a class profile with grade level and subject, then generates worksheets, quizzes, or flashcards without any student ever needing an account.

A Quick Practical Habit Worth Adopting

Even though Teach does not require student data, avoid pasting a specific student's IEP goals, behavior notes, or other identifiable details into a prompt when asking for differentiation. Describe the ability range or accommodation category generally, such as "a small group reading two grade levels behind," instead of naming the student, since the prompt itself becomes something worth minimizing regardless of what the underlying data-protection terms promise.

Reading Coach: A Different Privacy Model for Younger Students

Reading Coach is a separate Learning Accelerator, not the same product as Copilot Chat, and it uses a meaningfully different privacy model that makes it a more natural fit for elementary grades. Understanding that distinction matters, since it is easy to assume every Microsoft AI-branded feature follows the same 13-and-up Entra ID gate that governs Copilot Chat.

According to Microsoft's own published description of the product, Reading Coach processes a student's read-aloud voice data on-device, specifically to check pronunciation, and that voice data is not used to train any underlying model. Three categories of data get created during a session: the voice data itself, the AI-generated story content, and general usage data, with school-account data stored against a student's Entra ID and protected under the tenant's own policies. Microsoft's own guidance recommends using the app "with guardian or educator supervision," language worth taking literally rather than treating supervision as optional once the on-device processing claim is in place.

What This Means for a K-9 Rollout

• Reading Coach can reasonably be introduced to younger grades that would not clear the Copilot Chat 13-and-up gate, since it operates under its own separate eligibility terms rather than the ageGroup restriction covered in the companion legal guide.
• On-device voice processing is a meaningfully different privacy posture than a cloud-processed chat prompt, worth explaining to families in plain language rather than assuming "AI-powered" alone tells them enough.
• Supervision is still explicitly recommended, not an optional nicety, even with the on-device processing model in place.

Prompt Hygiene: What Not to Paste In

A signed-in work or school account's prompts and responses are encrypted during the session, isolated within the school's own tenant, and excluded from foundation-model training, according to Microsoft's own documentation, but none of that changes what belongs in a prompt in the first place. A privacy-conscious habit is still worth building deliberately, the same way a teacher would think twice before emailing sensitive information even on a secure system.

None of this is a workaround for a gap in the stated protections; it is a habit that holds up regardless of which AI tool a district adopts next, since minimizing what goes into any prompt is a durable practice, not a tool-specific one.

The Permission-Inheritance Trap Worth Knowing About

Full Microsoft 365 Copilot only surfaces content a signed-in user could already see through existing SharePoint and file permissions, which sounds reassuring until a school's own permission hygiene turns out to be the actual privacy boundary. A teacher who has not audited a shared class folder's sharing settings in a while can be surprised by what a search actually pulls up.

A practical example: a shared drive folder set up years ago for a grade-level team, with a sharing link left broader than intended, does not become a new privacy problem the moment Copilot arrives, but Copilot can make an old, forgotten permission gap far more visible than it was before, simply by surfacing a document in response to a search that a manual folder browse never would.

A quick audit worth doing before a first rollout: open the sharing settings on the folders and files a class actually uses, confirm they are scoped to the intended group rather than "anyone in the organization," and repeat that check on a schedule, not just once at setup.

This matters more in a K-9 building than it sounds, because a shared folder often accumulates a broad "anyone with the link" setting precisely because a teacher needed to send something quickly to a substitute or a parent volunteer years ago and never revisited it. A permission audit costs a fraction of the time a full rollout review does, and it protects every other tool that touches the same shared drive, not just Copilot.

Study and Learn Agent: What to Configure Before Turning It On

The Study and Learn Agent, available for students 13 and up under the same eligibility rule as Copilot Chat, is designed as an interactive tutor that guides practice and gives feedback rather than simply answering a question outright. For an upper-grade K-9 classroom, roughly Grade 7 through 9, that tutoring format raises a slightly different practical question than a general chat interface does: what happens during an extended back-and-forth, not just a single exchange.

A few things worth reviewing before assigning it to a class:

1. Preview a full practice session yourself first, not just a single sample question, since a tutoring agent's value and risk both live in the back-and-forth, not a single response.
2. Check whether your building's IT team has restricted Agent Builder, the feature that lets a student or staff member construct a custom AI agent, since a custom-built agent inherits whatever configuration its creator gave it rather than the platform's own defaults.
3. Decide which subject and skill the tutoring session should stay scoped to, and communicate that boundary to students before the first session, the same expectation-setting habit that applies to any new AI tool.

A Grade 7 Scenario, Walked Through

Say you teach Grade 7 social studies and your district has just enabled Teach, Copilot Chat, and the Study and Learn Agent tenant-wide. Rolling all three out at once is tempting; a staged approach protects more data than it costs in convenience.

• Start with Teach for your own lesson planning, since it needs no student data at all and lets you get comfortable with the interface before any student account is involved.
• Introduce Copilot Chat for teacher-only tasks first, drafting rubric language or summarizing a reading, before opening it to student use.
• Confirm Entra ID ageGroup accuracy with IT before enabling the Study and Learn Agent for the class, since a Grade 7 roster commonly spans both sides of the 13-year line.
• Hold a short conversation with students about what the Study and Learn Agent is for before their first session, mirroring the same disclosure habit worth building around any AI tool that interacts with a student directly.
• Document which grade levels and subjects actually use each surface, so next year's rollout planning starts from real usage data instead of guessing at adoption the way the first year did.

Related Reading

• AI, FERPA & COPPA: The Complete School Data-Privacy Guide (2026)
• Is Microsoft Copilot COPPA Compliant?
• Is Microsoft Copilot Safe for Students?
• Diffit and Student Data Privacy: What Teachers Should Know
• Is Perplexity Safe for Students?
• Is EduGenius COPPA Compliant?
• Is Curipod COPPA Compliant?
• SchoolAI vs Khanmigo: Which Is Better for Teachers?