ChatGPT Prompts for Teachers (2026): 18 Prompts, Sorted by the Student Data They Require

Here is the answer up front: the prompt was never the hard part — the student data inside it is. Every prompt below is written to be pasted as-is, and the eighteen are sorted by how much student information each requires: nothing about a student, student work with names stripped, and student-level records that belong in a district workspace or nowhere.

That sort is the whole point: a prompt library is worthless if half the entries put you on the wrong side of a federal privacy rule, and most published collections do not say which half is which. The rule that decides it is FERPA’s school-official exception, quoted below with its three conditions.

This is not a record of teachers trying these prompts — this site does not run tools. We read the regulation and the vendor’s own pages, stamp the date, and mark the lines. Every price below was checked in October 2026.

The rule that decides which prompts you can use

The default is consent. 34 CFR 99.30 requires a school to obtain prior written consent from a parent before disclosing personally identifiable information from an education record. Pastes into a chatbot are disclosures to a third party. So the question is never “is this prompt good” — it is “which exception am I standing on”.

The exception most schools use. 34 CFR 99.31(a)(1)(i)(B) lets an outside party be treated as a school official, but only if all three of these hold:

• it “performs an institutional service or function for which the agency or institution would otherwise use employees”;
• it “is under the direct control of the agency or institution with respect to the use and maintenance of education records”; and
• it “is subject to the requirements of § 99.33(a) governing the use and redisclosure” of that information.

A consumer subscription in your own name fails the second condition outright. Your district has no control over the account, no contract governing retention, and no way to audit what happens to the text. A district-managed workspace is built to satisfy all three, which is why the free teacher plan is scoped to a single district rather than open to any teacher with an email address.

What “de-identified” actually means. Deleting names does not solve this. 34 CFR 99.3 defines personally identifiable information to include “other information that, alone or in combination, is linked or linkable to a specific student that 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.” A distinctive essay about a family event is identifiable with the name removed. So is a grade list with the names deleted but the rows in alphabetical order.

The de-identification exception in 99.31(b)(1) has a second clause that matters more in practice: the school must make a reasonable determination “whether through single or multiple releases, and taking into account other reasonably available information.” Multiple releases. Pasting the same class’s essays every week re-identifies a group that was anonymous the first time.

Under-13 data has its own floor. COPPA, 16 CFR 312.5(a)(1), requires an operator to obtain verifiable parental consent before any collection of personal information from a child. You are not the operator and cannot consent on a vendor’s behalf, so the only question is whether the tool your school chose has already answered it.

The four questions to ask before you paste

1. Does this text relate to a specific student and is it maintained by the school? If yes, it is an education record under 99.3.
2. Is the tool a school official under 99.31(a)(1)(i)(B) — all three conditions, not just a signed district contract you have heard about?
3. If you stripped the identifiers, could a colleague who knows the class still name the student?
4. Is the student under 13, and has the district already handled consent for this tool?

Any yes-or-unsure answer means the prompt does not go in a personal account. The prompts below are labelled so you know which you never have to ask about.

The 18 prompts at a glance

Set	What it takes off your desk	Student data	Prompts
Planning	Unit outlines, lesson structure, misconception maps	None	5
Materials	Leveled texts, vocabulary, exit tickets, item rewrites	None	4
Feedback	Rubrics, written feedback, error patterns	De-identified or workspace	4
Communication	Parent emails, accommodation language, progress notes	Workspace only	3
Professional	Agendas, PLC notes, funding blurbs	None	2

Bracketed values are deliberate. Leave one blank and a careful model asks; a careless one invents a number, and an invented class size or reading level is harder to catch than a blank.

1. Planning prompts — no student data

The unit outline. “Build a [NUMBER]-day unit on [TOPIC] for [GRADE] [SUBJECT]. For each day give the objective, a 10-minute opening, the core activity, and the formative check. Sequence it so each day’s check predicts the next day’s prerequisite. Flag any day that depends on a skill from an earlier grade.” The last sentence surfaces the prerequisite gap before you are three days in.

The 5E lesson. “Turn this objective into a 5E lesson: [OBJECTIVE]. For Engage, give me a phenomenon that contradicts what students will predict. For Explore, a task with no vocabulary introduced yet. For Explain, the minimum correct explanation. For Elaborate, one transfer task. For Evaluate, three questions ranked by difficulty.” Forcing a contradicting phenomenon is what stops the lesson from opening with a definition.

The misconception list — the most useful prompt here. “List the five most common wrong answers students give to [SPECIFIC SKILL], and for each one explain what the student is probably thinking and what question would expose it.” This is diagnostic work that usually happens after the quiz. Run it before and you know which wrong answer to expect.

Unpacking a standard. “Rewrite this standard as a list of things a student can do, ordered from easiest to hardest: [STANDARD TEXT]. For each, give one task that would prove it and one that would not, and say why.” The “would not” column is where standards documents go vague.

The assessment blueprint. “I have [N] minutes and [N] items on [TOPIC]. Give me a blueprint: how many items per objective, what cognitive level each should sit at, and which objective I should cut if I lose five minutes.”

2. Materials prompts — no student data

Leveled text from a source. “Rewrite this passage at three reading levels — below, on, and above [GRADE] — keeping every fact identical and no sentence longer than [N] words at the lowest level: [PASSAGE].” Keeping the facts identical stops a “simplified” version from quietly changing the science.

Vocabulary with context. “Pick the eight words from this passage a [GRADE] student is least likely to know: [PASSAGE]. For each, give a student-friendly definition, one sentence using it in a different context, and one common misuse.” The misuse column turns a word list into a lesson.

The exit ticket with a key. “Write four exit-ticket questions on [OBJECTIVE]: two recall, one application, one that only a student who understands the concept can answer. Give the answer key and, for each wrong option, the reasoning error it represents.” Distractor rationales are the part most generators skip.

The readability rewrite. “Rewrite these instructions so a student reading two grades below level can follow them without help, without removing any step: [INSTRUCTIONS].”

3. Feedback prompts — de-identified or workspace

A rubric from an exemplar. “Here is an exemplar response with all identifying details removed: [TEXT]. Build a four-level rubric for [ASSIGNMENT TYPE] that this exemplar would score at the top level, with descriptors specific enough that two teachers would agree on a mid-level paper.” Two-teacher agreement is the honest test of a rubric.

Feedback on anonymised work. “Here is a student response with names and identifying details removed: [TEXT]. Give me three comments: one on what works, one on the single highest-leverage fix, one question that pushes the thinking further. Do not rewrite it for them.” The prohibition keeps the feedback from becoming a replacement essay.

Error-pattern analysis. “Below are the wrong answers from [N] students, with names removed and order randomised: [LIST]. Group them by the misunderstanding they share, and tell me which group to reteach first.” Randomising the order stops an alphabetical list from re-identifying the class.

Feedback at a lower reading level. “Rewrite this feedback so a [GRADE] student can act on it independently, keeping the same three points: [FEEDBACK].”

4. Communication prompts — workspace only

These three need student-level records: a district workspace or a blank page.

The parent email. “Draft an email to a parent about [ISSUE] for a [GRADE] student. Lead with what the student did well, state the concern in one sentence without diagnosis, propose one specific thing the family can do this week, and offer two times to talk.” Leading with the positive keeps the message from reading as an accusation and getting forwarded to an administrator.

Accommodation language. “Draft accommodation language for a team meeting: the student needs [SUPPORT] to access [TASK]. Write it as an instructional accommodation, not a diagnosis, and note what the teacher will do differently.” An IEP or 504 is a team decision; drafting the wording of a support is the whole of a model’s role here.

The progress note. “Summarise these [N] observations into a progress note for [PERIOD]: what improved, what did not, and the evidence for each, in plain language a family can read: [NOTES].”

5. Professional prompts — no student data

The PLC agenda. “Build a 45-minute PLC agenda on [TOPIC] with three segments: a data discussion, a decision, and a task with an owner. Give each segment a time box and one question the facilitator asks to keep it moving.”

The funding blurb. “Write 150 words for a [PLATFORM] project page requesting [ITEMS] for [ACTIVITY] in a [GRADE] classroom. State the need in the first sentence, describe what students will do with it, and avoid words that imply the materials teach themselves.”

Where a prompt is the wrong tool

Three jobs look like prompt jobs and are not.

Grading a stack of essays. A prompt handles one response at a time and keeps no record of whether you applied the same standard to paper 3 and paper 27. Rubric-anchored tools exist for this, and our roundup of AI grading tools separates the ones that score against your rubric from the ones that summarise.

Building a term of lessons. A prompt forgets the unit it wrote yesterday; a lesson-plan generator that holds the sequence does not. The lesson plan generator comparison prices them.

Making a worksheet with a usable answer key. Here a general assistant loses to a purpose-built tool: the output is a wall of text you have to lay out, and the key is whatever the model decided the questions meant. The worksheet generator comparison marks which of the six state an answer key publicly.

What to run them in, and what it costs

The bottom line

Copy the prompts in sections 1, 2 and 5 into a document today; they need nothing from a student and work in any account, including the free one. Sections 3 and 4 are where the law enters, and the test is not “did I delete the name” but whether a colleague who knows the class could still name the student — the standard 34 CFR 99.3 sets, including the “multiple releases” clause.

If you are a verified U.S. K-12 educator, the teacher workspace is free through June 2028 and is the only option here that satisfies the school-official conditions without an argument. If you are not, the consumer tiers cover the prompts with no student data, and the honest answer for the rest is a blank page. For the wider paid stack, start from our best AI tools for teachers pillar.

Related guides

• The 7 Best AI Tools for Teachers in 2026 (Classroom-Ready)
• 6 Best AI Lesson Plan Generators for Teachers (2026): Compared, Not Sponsored
• Best AI Grading Tools for Teachers (2026): Picks for Essays, Exams, and Bubble Sheets
• The 8 Best Free AI Tools for Teachers in 2026 (Free Tiers, Checked)
• The 6 Best AI Worksheet Generators for Teachers in 2026 (Editable, With Answer Keys)

Frequently asked questions

Is ChatGPT for Teachers actually free, or is it a trial?

Free through June 2028 for verified U.S. K-12 educators, according to OpenAI's own help center page. You verify with a school or district email address at chatgpt.com/k12-verification, then either create a workspace or join your district's existing one. The plan is not available to students, and workspaces cannot mix educators from different districts. OpenAI says it may introduce or adjust pricing after June 2028 and will give advance notice.

Can I paste student work into ChatGPT to grade it?

Not into a personal account, and not without thinking about which exception you are relying on. FERPA requires prior written consent before a school discloses personally identifiable information from an education record, unless an exception applies. The exception most districts lean on is the school-official one at 34 CFR 99.31(a)(1)(i)(B), and it has three conditions: the vendor performs a service the school would otherwise staff itself, the school has direct control over how records are used, and the vendor is bound by the use-and-redisclosure rules. A personal ChatGPT account in your own name satisfies none of them.

Do I need ChatGPT Plus for prompt work?

For most teachers, no. Free and Go both include unlimited everyday text chats, and prompt drafting is everyday text chat. Plus is $20 a month and buys reasoning models, projects and custom GPTs; Pro starts at $100 a month across three tiers. All prices verified October 2026 on OpenAI's own pages. If you are a verified U.S. K-12 educator, the teacher workspace is the better answer at $0.

Which single prompt on this list is worth memorising?

The misconception list in section 1. Ask for the five most common wrong answers to a specific objective and why each one is attractive to a student, and you get a diagnostic you can use before you teach the lesson rather than after the quiz. It costs one prompt, needs no student data, and changes what you say in the first ten minutes of class.

Will my district's AI policy let me use these at all?

Ask before you paste, because the answer varies by district and the policy is usually newer than the tool. The question to put to your administrator is narrow and answerable: is there an approved-tools list, and does the district workspace cover this use? If the answer is no list exists, treat anything containing identifiable student information as off-limits in a personal account and stay inside the no-student-data prompts in sections 1, 2 and 5.

Keep reading

→

Best AI Grading Tools for Teachers (2026): Picks for Essays, Exams, and Bubble Sheets

CoGrader, EssayGrader, Class Companion, Gradescope, ZipGrade, and MagicSchool compared on the work each one is built to grade. Which AI grading tool fits your classroom in 2026 — with pricing verified at the source and honest limits.

→

6 Best AI Lesson Plan Generators for Teachers (2026): Compared, Not Sponsored

MagicSchool, Diffit, Curipod, Eduaide, Kuraplan, and ChatGPT compared on the job a teacher actually has — turning a standard into a full lesson — across standards alignment, differentiation, exports, and free-tier limits. Here's what saves prep time.

→

The 6 Best AI Math Tools for Teachers in 2026 (Computed, Authored, or Guessed)

Math tools for teachers sorted by where the answer key comes from: a computer algebra engine, a human-authored item bank, or a language model predicting the next token. Six picks, with prices read from each vendor's own page in October 2026.