What to Share and What Not To: A Teacher's Data-Hygiene Checklist

The most useful thing Claude for Teachers can do is also the one that makes teachers hesitate: hand it a folder of class data — roster, diagnostics, attendance, notes — and let it build a picture of where each student is, so it can help you plan. That hesitation is healthy. It should be a deliberate act to put real student information in front of any tool. The good news is that Claude for Teachers is built for exactly this, and two facts turn the anxiety into a routine.

First: you control what's shared. Nothing is synced or scraped behind your back — data goes in because you put it in. Second: nothing you share is used to train the model. That's a commitment of the K-12 privacy terms, not a hope. Together they mean the risk isn't "the AI will learn my students" — it's the ordinary, manageable risk of handling records, which teachers already do every day. This piece turns those two facts into a checklist you can actually run before you upload.

Start from the principle: share the minimum the task needs

Data hygiene isn't a vibe; it's a discipline with one governing rule — data minimization. Ask what the task genuinely requires, share that, and withhold the rest. If you want Claude to group students by readiness for a differentiated lesson, it needs proficiency signals; it does not need home addresses, family circumstances, or medical flags. The instinct to "give it everything so it has context" is the instinct to resist.

Every field you don't need to share is a field you don't have to protect. Minimization isn't a limit on what Claude can do for you — it's what keeps the powerful use safe.

Because Claude for Teachers doesn't train on your inputs, minimization here isn't about starving a hungry model. It's about limiting exposure in the ordinary sense — fewer sensitive fields in play means less to think about, less to explain to a privacy officer, and less that could ever be misplaced. Treat it the way you'd treat a printout you're carrying out of the building: bring what the lesson needs, leave the rest in the filing cabinet.

The pre-upload checklist

Run this before you hand Claude a folder or paste a dataset. It takes a minute and it's the whole game:

• Confirm you actually need identifiable data. Can the task work with de-identified inputs — "Student A / B / C," or performance patterns without names? If yes, strip the names first. Claude can differentiate three tiers without knowing who's in each one.
• Share proficiency and progress, not the whole cumulative file. Diagnostics, exit tickets, and mastery signals are the fuel for planning. A student's full history, discipline record, or health notes usually aren't — leave them out unless the specific task truly requires them.
• Quarantine the sensitive categories. Special-education status, health conditions, family or custody situations, free-lunch eligibility, immigration details — these are the fields to withhold by default. If a task seems to need one, that's a signal to pause and check your district's policy, not to paste it in.
• Check what your district and state allow. The FERPA-oriented K-12 Data Processing Addendum makes compliant use possible, but it doesn't override your institution's own rules. Know them before you upload student records to any tool.
• Prefer patterns over paperwork. Instead of the raw gradebook export, consider sharing the summary you'd actually reason from: "these six students are below proficiency on fractions, these four are above." You get the same instructional help with far less exposed.
• Name your files like someone might read them. Keep uploads scoped and clearly labeled so you always know what's in play. A folder called "period-3-fractions-diagnostic" is easier to reason about than a grab-bag of exports.
• Do a final read before you send. Glance at what you're about to share as if it were a document leaving your hands. If a field would make you uncomfortable on a printout in the hallway, it doesn't belong in the prompt either.

That's the list. It isn't long because it doesn't need to be — the platform's no-training commitment and educator-only access do the heavy lifting, and this checklist just keeps your side of the arrangement clean.

Why this works: the platform is already doing half the job

It's worth being explicit about what you don't have to worry about, because knowing the guarantees is what lets you use the tool without second-guessing every keystroke.

Claude for Teachers is gated to verified educators, consistent with the 18-and-over policy — there's no student account and no unsupervised-minor channel. It runs on teacher terms built for K-12 privacy, and student information falls under a data processing addendum written to comply with FERPA. And the anchor fact: your shared data is not used for model training. So the classic consumer-AI fear — "my inputs become part of the model" — is off the table by design.

What remains is the same professional discretion you already exercise with a gradebook or a parent email: share deliberately, minimize by habit, follow your policies. The checklist above is just those habits, ported to a new tool. You're not learning a new risk model. You're applying the one you already have.

When the task genuinely needs the full picture

Sometimes real instruction needs real detail. Analyzing class data to plan a re-teach, or setting up the 4pm exit-ticket review that adapts tomorrow's plan, can legitimately call for identifiable, per-student information — that's the feature, not a misuse. The point of data hygiene isn't to never share; it's to share on purpose.

When you do need the fuller picture, lean on three things. Lean on the platform guarantees — no training, FERPA-oriented handling, adult-only access. Lean on your control — you decide the scope of the folder, and you can keep the most sensitive categories out even while sharing rich proficiency data. And lean on your judgment — the same judgment that already tells you which details belong in a parent conference versus a hallway conversation. Used together, they let you do the high-value work (a genuinely data-informed plan) without carrying more exposure than the work requires.

The mistake to avoid runs in both directions. Over-share, and you're carrying sensitive records into a workflow that didn't need them. Under-share out of vague fear, and you deny yourself the very help — a real picture of where each student is — that makes the tool worth using. Hygiene is the middle path: everything the task needs, nothing it doesn't.

The one-sentence version

If you remember nothing else: you decide what goes in, and what goes in doesn't train the model — so share the minimum the task needs and keep the sensitive categories out by default. That single habit, applied every time, is the whole discipline. It's what lets you say yes to the powerful use — handing Claude your class data to plan better instruction — while keeping your students' information exactly as protected as your professional obligations require.

Pair this with the FERPA and K-12 Data Processing Addendum explainer so you know the terms behind the guarantees, then put it to work in the analyze-class-data workflow. Browse the AI tutoring skills that run on this teacher-controlled model, or explore education agent skills built for the classroom.

---

Part of the Claude for Teachers series. Related: Hand Claude a Folder: Analyzing Class Data to Plan Instruction · FERPA and the K-12 Data Processing Addendum. Browse AI tutoring skills or more builder insights.