FERPA and ChatGPT: Does the Law Allow US School Staff to Use It?

FERPA, the US student-privacy law, does not ban school staff from using ChatGPT. It limits sharing personally identifiable student information from education records without consent, unless an exception applies. Staff may use ChatGPT freely on de-identified or non-student material, and should paste student records only into a tool the school has a proper agreement with.

The common error runs the other way from what people expect. Schools worry about "ChatGPT" as a brand, when the legal question is about the data and the contract. A teacher who pastes a pupil's name and a safeguarding note into a free personal account has a problem. A teacher who pastes an anonymised paragraph of student writing into the same account may not. For a private school owner, the useful result is a rule staff can follow on a Monday morning, backed by a redaction routine that takes two minutes.

Follow me on Instagram@sagnikteaches

Does FERPA cover your school at all?

The regulation, 34 CFR Part 99, applies to an educational agency or institution to which funds have been made available under any programme administered by the US Department of Education. Funds count whether they reach the school by grant, cooperative agreement, contract, subgrant or subcontract, or reach students who then pay them to the school. Public schools are covered. For private schools, the Department's Student Privacy Policy Office says that private and parochial schools at the elementary and secondary level generally do not receive such funding and are therefore not subject to FERPA. "Generally" is doing real work in that sentence: a school that takes money directly under a Department programme should not assume it is outside. I'm not a lawyer, so confirm your own position with an education-law adviser, and do not read this as an exemption.

Connect on LinkedInSagnik Bhattacharya

Treat the answer as less important than it sounds, for three reasons. Your enrolment contract or privacy notice probably promises parents confidentiality. State privacy and consumer laws can apply. And an insurer or a parent complaint will judge you on what you did with the child's information, not on a technical status. So a sensible private school follows the FERPA pattern whether or not FERPA is binding.

Subscribe on YouTube@sagnikteaches

The four FERPA ideas that decide every case

Education records and personally identifiable information

An education record is a record directly related to a student and kept by the school. Personally identifiable information (PII) includes the obvious items, such as a name, address, student number and date of birth, and also, in the regulation's words, information that alone or in combination would allow "a reasonable person in the school community, who does not have personal knowledge of the relevant circumstances, to identify the student with reasonable certainty" (34 CFR 99.3). In a class of eighteen, "the sixth-grader who moved from a sister school in March and has a nut allergy" is PII with no name in it.

Consent

The default is that a school needs written consent from a parent (or an adult student) before disclosing PII from education records. Consent for every AI use is clumsy, so most schools rely on an exception or on de-identification instead.

The school official exception

The regulation lets a school treat an outside party as a "school official" with a legitimate educational interest if that party performs a service the school would otherwise do with its own staff, works under the school's direct control over how the records are used and maintained, and is subject to the rules on using and redisclosing the information (34 CFR 99.31(a)(1)(i)(B)). In practice that means a written agreement: the vendor uses the data only for the school's purposes, does not pass it on, and the school can tell the vendor what to do with it. Without that, you do not have this exception, however friendly the product looks.

De-identified information

A school may release records once all PII is removed, if it has made a reasonable determination that a student's identity is not personally identifiable, "whether through single or multiple releases, and taking into account other reasonably available information" (34 CFR 99.31(b)(1)). That second clause is the one that bites in small schools. Removing a name is not enough where the context points to one child.

Directory information

Schools may designate categories such as name, photograph, grade level, dates of attendance or participation in sports and activities as directory information. Under 34 CFR 99.37 the school must give public notice of the types it has designated, of a parent's right to refuse, and of the time allowed to say so. That lets a school publish a team list. It does not make directory information safe to feed into any AI tool. A notice can limit directory disclosures to specific parties or purposes, and if yours does, the school must stay within those limits; your notice almost certainly did not mention AI vendors. Check your own notice before assuming otherwise.

What the plans mean for a school

The plan your staff use matters more than the chatbot's name. What follows reflects vendor information I could check; terms change, so confirm on the vendor's own pages before you rely on any row.

Two cautions. A vendor's statement that a product "supports FERPA" or "helps schools meet FERPA requirements" is the vendor's own description, not a ruling from the Department of Education, so it never replaces your own reading of the contract. And plan terms change often, so describe settings in your policy in general terms ("training switched off") and have one named person recheck them each term.

ChatGPT for Teachers versus a teacher's free account

For a private school owner, the difference between these two is mostly about control, not about the chatbot. On a free personal account, the teacher owns the account, sets the training switch, decides what to keep in chat history and takes everything with them when they leave. In a ChatGPT for Teachers workspace, OpenAI says school leaders can claim the school's email domain, bring staff into one workspace with role-based access, and secure sign-in with single sign-on, and the free period runs through June 2028 with advance notice promised before any price change. Verification is done by a third-party partner, so check early whether your staff qualify, because a private school that cannot verify cannot use it.

Even if your staff qualify, treat the workspace as a candidate for your approved list, not as automatically approved. Read the services agreement, check whether it fits the school official conditions above, decide which connected apps (formerly connectors) staff may switch on, and only then let pupil information in. A sensible split for a small school: the Teachers workspace, once reviewed, for anything touching pupil work; personal free accounts only for lesson ideas with no pupil data.

A paste test every member of staff can run in five seconds

Staff will not read regulations. They will follow a short test. Use three questions in order.

1. Could a parent, colleague or older pupil work out which child this is? Consider the name, but also the year group, a rare incident, a family detail or a combination of small facts. If yes, stop.
2. Is this a school-approved tool with an agreement? Only the tools on the approved list count. If it is not on the list, do not paste anything identifying.
3. Is the task worth the risk? If the same job can be done with a made-up example, do that.

The data you never paste, and the data you can

A redaction routine for the school office

Redaction means replacing identifying details with neutral labels so the AI still has what it needs. Doing it the same way every time is what keeps it reliable. This routine takes about two minutes per document once learned.

1. Copy the text into a blank document, not into the AI tool.
2. Find and replace the names with codes: Student A, Student B, Parent A, Teacher A. Keep the key (Student A is [pupil's name]) in a separate locked note that never goes near the tool.
3. Remove dates and places that narrow the field: replace "on 14 March at the grade 4 trip" with "during a trip this term".
4. Remove rare facts: a named condition, a family event, a transfer from another school. Replace with a general phrase or delete.
5. Read it once as a stranger. Could you tell who it is about? If you could, cut more.
6. Paste the redacted version, then map the codes back to names in your own document afterwards.

The longer method, including using AI to catch what you missed, is in the tutorial on redacting personal data from documents.

Three examples, redacted properly

A behaviour note to parents. Before: "Please can you write to [parent's surname] about [child's first name]'s fight with [second child's first name] in the playground on Tuesday, and mention his ADHD plan." After (what the teacher pastes): "Draft a calm, factual letter to a parent about a playground altercation between two children earlier this week. Offer a meeting. Do not mention any diagnosis." The AI writes the letter with "[parent name]" and "[child's name]" left blank, and the teacher fills them in. No identifying detail left the school.

Feedback on a piece of student writing. Before: a scan with the child's name in the header and a sentence about their grandmother's illness. After: the teacher types the paragraph into a document, removes the header, replaces the grandmother sentence with "[personal detail removed]" and asks for feedback against a rubric. If the sentence matters to the feedback, the teacher adds it back by hand.

An end-of-term report comment. Rather than pasting a pupil's marks, the teacher asks the tool for six comment frames at different performance levels, "secure in number work, needs support with fractions", then selects and edits one per child. The AI never sees a name or a mark.

What staff may do on a personal or free account

Plenty. These uses involve no student information:

• Lesson outlines and starter activities from a curriculum objective.
• Differentiated versions of a text the teacher wrote, with no child's details.
• Generic parent-letter templates, newsletter wording and event notices.
• Quiz questions and model answers.
• Admin wording: a staff-meeting agenda, a policy summary, a job advertisement.

Staff should still check outputs, since AI invents facts and can write confident nonsense, but the privacy risk here is low. This is also where a school gets its quickest wins, with no legal complication.

A one-page staff rule you can copy and adapt

AI USE AT [SCHOOL NAME] - STAFF RULE (reviewed each term by [role])

1. Approved tools: [list]. Only these may receive any information about a pupil, parent or staff member, and only through a school account.
2. Never paste into any other AI tool: names, dates of birth, addresses, photographs, grades or attendance linked to a child, safeguarding, medical, special-needs or disciplinary notes, or fee account details.
3. Before pasting anything about a child, apply the paste test: could someone identify the child? If yes, redact first using the routine on the office shared drive.
4. Personal and free accounts are for material with no pupil information.
5. AI output is a draft. The member of staff who sends it is responsible for checking it.
6. If you think you pasted something you should not have, tell [role] the same day. You will not be in trouble for reporting; you will be if you hide it.
7. Do not enter pupil work into an AI tool to mark or grade without checking this rule with [role].

Rule 6 does more work than any other. The school can contain a mistake that is reported on the day. It cannot contain one that surfaces six months later in a parent's complaint. For advice on getting staff to follow a rule like this, see how to roll out an AI policy so staff actually follow it.

The owner's decisions, in order

1. Work out whether you are covered. Ask your adviser, then write down the answer and the date.
2. Update your privacy notice. Say that the school uses AI tools for certain tasks and under what protections. Parents who read it in advance complain less.
3. Pick one approved tool. Fewer tools mean fewer agreements to read. Get the terms in writing and file them. If you use a management system with built-in AI, ask the vendor what sub-processors handle the data.
4. Issue the staff rule and run a twenty-minute briefing with the three examples above, using your own school's situations.
5. Review each term. Plan terms change. Name one person to check settings and the approved list.

A parent-facing example is useful too. For admissions, fee letters and parent messages, the practical uses are covered in AI for private schools, and fee wording specifically in fee reminder letters with AI.

Questions to put to any vendor before staff use it with records

The school official route stands or falls on the contract, so put these to the vendor in writing and file the answers.

• Does the vendor use our content to train or improve models? If there is a switch, who controls it, and is it off by default for our plan?
• Will the vendor use student information only to provide the service to us, and not disclose it onward except to named sub-processors?
• Can we direct deletion, and how long does the vendor keep chats and uploaded files? What happens when a staff member leaves?
• Who at the school can see staff chats, and can an administrator turn features such as connected apps or memory off?
• If there is a security incident affecting our data, how and when will the vendor tell us?

An answer of "see our general terms" is not an answer. If the vendor cannot say plainly what happens to a child's information, the tool goes on the "no pupil data" list.

A safe prompt and what comes back

Here is a prompt a teacher can run on any account because it contains no pupil information. The output is illustrative.

Write four report-comment frames for a grade 5 maths report, one for each level: working towards, secure, strong, exceeding. Each is two sentences, positive in tone, mentions fractions, and leaves [name] and [pronoun] as placeholders.

The model returns something like: "[name] is beginning to compare simple fractions and benefits from using a number line. With regular practice [pronoun] will become more confident." What you would fix: the output often sounds generic and repeats "confident". The teacher edits for the child's real work after the placeholders are filled in on the school's own system, so the specific detail never enters the tool.

How a small school might run it day to day

Picture a school of 140 pupils with twelve teachers and two office staff. The owner picks one approved tool, writes the one-page rule, and holds a twenty-minute briefing in the first staff meeting of term. The office manager keeps a list of approved tools on the shared drive and a locked folder with the vendor's terms. Teachers use the free account at home for lesson ideas with no pupil data. For the one task that needs real records, drafting letters to parents, the office uses a template with placeholders and merges names through the school's existing system. The owner spends about two hours in total on the setup and ten minutes a term on the check. The saving is not the point here. The point is that the next parent who asks "do you put my child's information into ChatGPT?" gets a clear, true answer.

Three slips a school office can make with AI

A school secretary pastes a whole spreadsheet of fee arrears, with parent names, into a chatbot "to draft reminders". The names should have stayed behind, and the fix is to draft one template with placeholders and merge the real names in the school's own system. A teacher asks AI to "mark" a class set of essays by uploading photographs of handwritten pages, names included. The fix is to type or transcribe samples without names, or to wait for an approved marking tool. A dean of students drafts a sensitive email to a parent with a pupil's full circumstances in the prompt. The fix: describe the situation in general terms and add the specifics by hand afterwards. None of these people meant harm, which is exactly why a short rule and a quick test work better than a ban.

How to check the rule is working

After one term, ask three questions in a staff meeting: which AI tools did you use, did anyone hit a case where the rule was unclear, and did anyone paste something they now doubt? Anonymous answers on a slip of paper get more honest replies. Count the doubts. If the number is more than zero and nobody reported it, your culture needs attention more than your policy. Then check the approved tool's settings against what the policy says.

Questions school owners ask about FERPA and ChatGPT

Does FERPA apply to my private school?

The Department of Education's student privacy office says private and parochial K-12 schools generally receive no funds from its programmes and so are not subject to FERPA. Funding arrangements vary, so check with your adviser and your funding paperwork. Parents expect the same care either way, and your contracts may promise it.

Can a teacher paste a first name and a behaviour note into ChatGPT?

Treat that as personal data from an education record, because a first name in a small class identifies the child. Use a code such as Student A and remove dates, incidents and family details, or use a tool your school has an agreement with.

Is a teacher's free personal ChatGPT account acceptable for school work?

For anything that contains student information, no. A personal free account sits outside any school agreement, and the school cannot control retention or training settings. Keep personal accounts for lesson ideas with no student data.

Do I need parental consent to use AI with student work?

It depends on whether the tool receives personally identifiable information. Consent is one route, but an agreement that makes the vendor a school official, or removing identifiers first, are the usual alternatives. Ask your adviser which suits your circumstances.

Further reads

• AI Acceptable Use Policy for a Small Professional Firm — A longer policy structure to adapt.
• How to Anonymise Client Data Before You Paste It Into AI — Anonymising methods beyond find-and-replace.
• Private School CRM: Do You Need One, or Will a Sheet and AI Do? — Where pupil enquiry data should live.
• How to Increase Private School Enrolment: Faster Replies With AI — Enquiry replies without exposing records.
• AI Policy for a Small Library: One Page, Patron-Data Rules — A short policy for another record-holding organisation.
• Are ChatGPT, Claude, Gemini and Copilot GDPR-Compliant? — How vendor plans differ on data handling.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: 34 CFR Part 99 (sections 99.1, 99.3, 99.31 and 99.37 as read on the Legal Information Institute site); Student Privacy Policy Office FAQ on which institutions FERPA applies to; OpenAI's ChatGPT for Teachers announcement and business data page; OpenAI data controls FAQ (all checked November 2026). Check the vendor's current terms before relying on any plan.