MagicSchool gives K-12 teachers 80+ task-specific AI tools for lesson planning, assessment and feedback, plus a supervised student workspace, wrapped in FERPA, COPPA and SOC 2 documentation that districts can actually sign.
Used this tool? Rate it
Used this tool? Rate it
MagicSchool is a K-12 AI platform that packages generative AI into dozens of narrow, task-shaped tools for teachers and, through a separate student-facing surface, for the learners in their classrooms. Rather than handing an educator a blank chat box, it presents a catalogue: one tool writes a lesson plan, another builds a multiple-choice quiz, another drafts report card comments, another produces a rubric. The company's own framing of the product puts safety and institutional alignment ahead of raw capability — the site's opening promise is to "Save time, spark creativity, and support every student with safe, district-aligned AI", and the navigation is organised by buyer role (For Schools, For Teachers, For Students) rather than by feature category. That structure tells you something important about the product before you use it: the primary sales motion is district procurement, not individual subscription.
The scale of the catalogue is the headline number. The vendor states the platform carries 80+ teacher tools alongside a separate set of 50-plus student tools. According to a case study published by one of its model providers, the platform reports roughly 7 million educators, 13,000+ schools and districts and over 100 million AI-powered interactions — figures that come from the vendor side and have not been independently audited. What distinguishes it from asking a general chatbot the same questions is not the underlying model but the packaging: ISTE's EdTech Index summarises the distinction plainly, noting that "Unlike general-purpose large language models, MagicSchool is designed for education" — meaning the prompts, the guardrails, the output formats and, critically, the legal paperwork are pre-built for a school context.
That last element deserves emphasis up front, because it is where most of the real decision-making happens. A teacher can evaluate a lesson plan generator in ten minutes. A district technology officer cannot sign anything until the FERPA posture, the COPPA position, the SOC 2 attestation and the data processing agreement have all been reviewed. MagicSchool's product design is visibly shaped by that second audience, and this page treats both perspectives as equally important.
A task-shaped teacher tool catalogue rather than a chat interface: The teacher side is a browsable library of named generators — an AI lesson plan generator, an AI worksheet generator, an AI presentation generator, a Multiple Choice Quiz Maker, an AI rubric generator and an Academic Content tool. Each one collects structured inputs (grade level, standard, topic, reading level) instead of asking the teacher to compose a prompt from scratch. The practical benefit is that the quality floor does not depend on prompt-writing skill, which is the single biggest barrier to AI adoption among educators who have no interest in becoming prompt engineers.
Administrative and special-education paperwork tools: Beyond instruction, the catalogue reaches into the documentation work that consumes teacher evenings. Named tools include Writing Feedback, Report Card Comments, Text Rewriter, Multi-Step Assignment, Professional Email and an IEP Generator for individualised education programme drafting. The IEP tool is a notable inclusion because IEP documentation is both extremely time-consuming and legally consequential, which makes it simultaneously the highest-value and highest-risk use of the platform.
A supervised student workspace, not an open student chatbot: The learner-facing side ships as its own product surface with named tools including Quiz Me!, Research Assistant, Character Chatbot, AI Learning Assistant, Writing Feedback and a Custom Chatbot that a teacher configures with their own instructions, boundaries and goals. The design principle is that students meet AI inside a container their teacher built, not on an open frontier.
Real-time teacher visibility into student AI use: Through a feature the vendor calls Student Room Insights, teachers can "see how students are using AI" and identify "where extra support may help". This turns student AI usage from an invisible risk into observable classroom data — a meaningful reframing for schools whose current alternative is students quietly using consumer chatbots at home.
Layered content moderation built for the K-12 age range: The vendor describes "a multi-layered moderation approach designed for K–12 settings" with built-in guardrails and content filtering, and offers enhanced moderation at the enterprise tier. Moderation is a prerequisite rather than a feature in this category — a product that cannot keep age-inappropriate output away from a twelve-year-old is unusable in a school regardless of how good its lesson plans are.
Multi-model architecture rather than a single vendor lock: The company states that "MagicSchool uses multiple large language models, including models from OpenAI, Anthropic, and Google, depending on the task". A published case study confirms the routing is actively revised — the company "transitioned many of its tools from OpenAI to Claude" for a substantial portion of its catalogue. For a buyer, the significance is that model quality can improve without any migration effort on the school's part, and no single provider outage takes the whole platform down.
Automatic de-identification as an operational backstop: The same case study describes a concrete safeguard for the most common human error in classroom AI use: "If you accidentally include the student's name, the AI removes it in the background". This matters because the realistic privacy threat in a school is not a sophisticated breach; it is a tired teacher pasting a student's name into a text box at 9pm.
Vendor-reported outcome metrics, presented with the appropriate caveat: The site advertises time savings of seven to ten hours per week, that 88% of teachers say it helps them reach every learner, and a 28% improvement in students meeting literacy grade-level expectations. These are vendor-published figures with no independent auditor named on the page, and this page treats them as marketing claims rather than established findings. The Limitations section explains why that distinction is not pedantic.
Weekly lesson and materials preparation: The most common entry point. A teacher supplies a topic, grade band and standard, generates a lesson plan, then chains the worksheet generator and presentation generator off the same content so that the slide deck, the handout and the plan actually agree with each other. The pain point being solved is not creativity — most teachers know what they want to teach — but the mechanical hours spent formatting the same content into three different artefacts.
Assessment design and differentiated practice: The quiz maker and rubric generator let a teacher produce an assessment plus its scoring guide in one sitting, then generate variant reading levels of the same passage for students who need them. This is where the differentiation promise is most credible, because producing four reading levels of one text by hand is a task teachers genuinely skip for lack of time.
Writing feedback at a volume humans cannot match: Thirty essays with substantive comments on each is a weekend of work. The Writing Feedback tool produces a first pass that a teacher then edits. The honest framing is that this changes the teacher's job from writing feedback to reviewing and correcting feedback — faster, but not free, and the Limitations section covers why the review step cannot be skipped.
Special education and behaviour documentation: IEP drafting, behaviour support planning and similar structured documents follow rigid formats that AI handles well as a starting draft. The value is highest here precisely because the paperwork burden is highest, but so is the requirement for professional review, since these documents carry legal weight.
Supervised student research and study support: Instead of banning AI and hoping, a teacher opens a Research Assistant or a purpose-built Custom Chatbot for a specific assignment, sets its boundaries, and observes usage. This reframes the academic integrity problem from prohibition — which has a poor track record — to structured, visible use.
District-wide rollout with central oversight: A technology director deploys through single sign-on, connects the existing LMS, negotiates a data privacy agreement, and monitors adoption across schools through dashboards. This is the use case the product's pricing and documentation are actually built around, even though most individual teachers arrive through the free tier.
Create an account on the free tier and check whether your district already has one. This step is easy to skip and worth doing first: if your school or district has an enterprise agreement, signing in through the district route gives you the negotiated privacy terms, SSO and any district-specific tool configuration. Signing up independently with a work email gives you the individual free tier instead, under different terms.
Start from the tool catalogue, not from a blank prompt. Browse to the named tool that matches your task rather than trying to describe the task in prose. The structured inputs are the product's main advantage over a general chatbot, and bypassing them gives up most of that advantage.
Fill in the context fields honestly and specifically. Grade level, subject, standard, reading level and class context all materially change the output. Vague inputs produce generic output that then requires more editing than it saved. Do not enter student names or other identifying details — the platform discourages this, and the enterprise documentation assumes you will not.
Generate, then read the output critically before it goes anywhere near a student. This is not optional advice; it is the vendor's own contractual requirement. Check factual claims, check for bias, check that the reading level actually matches the class, and check that any assessment items have unambiguous correct answers.
Iterate with adjustments rather than regenerating from scratch. Narrowing one input — dropping the reading level, adding a specific misconception you want addressed, changing the question format — usually produces a better second draft than regenerating with the same inputs and hoping for different luck.
Export into the workflow you already use. Push materials into Google Classroom, Canvas or Schoology rather than maintaining a parallel set of documents inside the platform. A tool that becomes a second place to keep things has added work rather than removed it.
For student-facing use, configure the container before opening the door. Set the chatbot's instructions and boundaries, tell students explicitly what use is expected and what is not, and plan to actually look at the usage insights rather than treating them as a compliance artefact.
Treat the first output as a draft from a fast but unreliable colleague. The most effective users report the biggest gains on tasks where a mediocre first draft saves real time — administrative writing, structured documents, variant reading levels — and smaller gains on tasks where the teacher's own judgement is most of the value.
Never paste student names or identifying details, even when it seems convenient. The platform removes names it detects, but relying on an automated backstop rather than not entering the data is the wrong order of operations, and district agreements are typically written on the assumption that PII does not enter the input field.
Cross-check every factual claim in generated content. History dates, scientific mechanisms, mathematical worked solutions and any figure that a student might repeat on an exam are the highest-risk categories. A confidently wrong worked example does more damage in a classroom than in almost any other setting, because students memorise it.
Verify assessment items have exactly one defensible correct answer. Auto-generated multiple choice questions frequently produce two arguably correct options or a distractor that is technically true. This is the single most common editing task reported for quiz generation.
Use the free tier to establish whether the tools you personally use are worth paying for. Because the free tier includes the full catalogue and limits generation volume rather than features, a few weeks of real use tells you exactly which tools you reach for — and whether that number justifies a subscription.
Set an explicit classroom AI policy before deploying student tools, not after. Students need to know what use is expected, what is prohibited and what is being observed. Deploying a monitored tool without disclosing the monitoring is both an integrity problem and a trust problem.
Do not use generated material as a substitute for reading the source. If you would not teach from a colleague's lesson plan on a topic you had not studied, do not teach from a generated one either. The tool compresses preparation time; it does not substitute for subject knowledge.
Keep the human review step visible in your own workflow. Teachers who build the review step into their process — for instance, always editing in a separate document rather than exporting directly — report fewer errors reaching students than those who treat generation as the final step.
Classroom teachers in K-12 schools: The core audience for teacher AI tools of this kind. The tool catalogue maps directly onto the recurring tasks of a teaching week, and the free tier removes the procurement barrier for individual trial.
Special education and support staff: The IEP generator, behaviour documentation and differentiation tools address the paperwork-heaviest roles in a school, where the ratio of documentation to direct student contact is most lopsided.
Instructional coaches and curriculum leads: People responsible for materials quality across multiple classrooms can use the platform to produce consistent, standards-aligned resources and to model responsible AI practice for the teachers they support.
District technology and data officers: The audience the compliance documentation is written for. Custom data privacy agreements, SSO, rostering integrations, VPAT accessibility documentation and district dashboards exist specifically for this role.
School and district administrators: Principals and superintendents evaluating whether to sanction AI use at all, and who need visibility into adoption and a defensible answer to parent questions about student data.
Students in supervised classroom settings: The learner-facing tools are genuinely designed for students, but only within a teacher-configured context — this is not a product a student adopts independently.
Teacher educators and professional development leads: The platform's structure makes it a usable teaching object for AI literacy work, where the point is to show educators what these systems do well and badly.
Web application: The platform is delivered through the browser as its primary and universal surface, which is the correct architectural choice for schools where device fleets are mixed and installation rights are restricted.
LMS integrations: Canvas, Schoology and Google Classroom are supported, allowing generated materials to land inside the course structure teachers already use rather than in a parallel system.
Single sign-on and rostering: Google, Clever, ClassLink and Microsoft are supported. The inclusion of Clever and ClassLink is a meaningful signal — these are the de facto rostering standards in US K-12, and supporting them indicates the product was built to district IT expectations rather than retrofitted.
Language coverage: The company reports 24 interface languages with translation into a much wider set, which matters for multilingual classrooms and for communication with families who do not read English.
Enterprise administrative surface: District dashboards providing visibility into adoption and usage patterns across schools are part of the enterprise tier rather than the individual product.
The free tier is genuinely usable rather than a demo. It is offered at no cost with no expiry and includes the full teacher and student tool catalogue, with the constraint applied to generation volume rather than to feature access. Remaining usage is visible in the account dashboard, and the platform warns as the limit approaches. This structure is worth understanding because it inverts the usual freemium logic: you are not locked out of tools, you are metered on how much you generate. For a teacher whose peak usage is concentrated in planning weeks, that distinction determines whether the free tier is workable.
Paid access comes in two shapes. The individual Plus plan is listed at $8.33 USD/user/month (billed annually) or $12.99 USD/month (billed monthly), and what it adds is principally the removal of limits — unlimited generations, unlimited output history, unlimited editing of AI output — plus advanced tool features and access to the MagicSchool Labs experimental channel. The Enterprise tier is quoted rather than listed, and its additions are institutional rather than personal: custom data privacy agreements, SIS/LMS integrations, Single Sign-On, curriculum alignment, tool customisation, enhanced moderation, advanced dashboards and dedicated onboarding support. The gap between Plus and Enterprise is therefore not about better AI; it is about governance. A district buying Enterprise is buying the paperwork and the controls, and should evaluate it on those terms.
The teacher-AI category has become crowded, and the honest comparison depends on which problem you are solving. Other platforms in the same K-12 space — the same independent reporting that covers MagicSchool names SchoolAI and Buddy in the same breath — compete on overlapping catalogues of teacher tools, and Brisk Teaching takes a different architectural approach by living as a browser extension inside the documents teachers already work in, which suits teachers who do not want to move their workflow to a new destination.
Against general-purpose assistants such as ChatGPT, Claude or Gemini, the trade is capability breadth versus fit. A frontier chatbot will often produce a better single output for a skilled prompter, and costs about the same for an individual. What it will not produce is a signed data processing agreement, a FERPA posture, rostering integration, or teacher visibility into what students did with it. For an individual teacher experimenting privately, the general assistant may be the better tool. For anything institutional, the compliance layer is the entire point, and that is precisely where a general assistant cannot compete without the school assuming the risk itself.
The vendor's own terms of service tell you not to trust the output. This is the single most important thing on this page. The contract states plainly that "AI can occasionally produce biased, fictitious, or otherwise incorrect Content" and that users should always double-check for bias and accuracy before sharing output with students. It further states that output "by no means replaces your professional skills and judgment" and that reviewing and revising it is the educator's obligation. Read alongside the marketing claim of seven to ten hours saved per week, this creates a tension every buyer should reason about explicitly: the time saving is real only to the extent that reviewing generated content is faster than producing it, and for content where errors are costly, that margin is thinner than the headline suggests.
Independent evidence for the learning outcomes is thin. The vendor advertises a 28% improvement in literacy grade-level attainment. Independent reporting on the category is markedly more cautious, noting that the effectiveness of personalised learning chatbots "has yet to be independently established by researchers". The same reporting raises a finding that should give any school pause: students using AI show immediate performance gains, but "their performance plummets once the tool is removed" — a pattern consistent with the tool substituting for cognitive work rather than building capability. None of this is specific to MagicSchool, but all of it applies to it.
AI grading is not reproducible. The same independent reporting documents that "When the AI chatbot grades the same paper hundreds of times, it does not always provide the same grade". This has direct implications for the Writing Feedback and rubric tools: they are defensible as a first-pass drafting aid and indefensible as an unreviewed source of marks. Any school using them for anything that affects a student's record needs a human in the loop by policy, not just by habit.
Generated output carries no intellectual property guarantee. The terms state that MagicSchool "does not represent or warrant that the Outputs are protectable by any intellectual property rights". Districts that plan to compile generated materials into a shared curriculum bank and distribute them should factor this in.
Independent scrutiny of the category is not uniformly positive. Mainstream reporting frames the sector with visible scepticism, describing the beneficiaries of the current situation as "a growing list of companies with names like MagicSchool, SchoolAI, and Buddy". Independent commentators have also raised structural criticisms of this specific product, arguing that "there is misalignment between what MagicSchool says their company does, and the features it is building", and questioning whether adding another tool to schools already running on the order of 1,500 digital tools meaningfully reduces teacher burnout at all. These are arguments rather than findings, but they are the kind of arguments a procurement committee should hear.
Independent review coverage is shrinking. Common Sense Education, historically the most-cited independent editorial reviewer in K-12 edtech, has paused its review programme, stating "We are not completing new reviews or review updates at this time". Buyers who relied on that channel now have fewer independent signals to work with, and should weight their own pilot evidence more heavily as a result.
Review-platform sample sizes remain small. Ratings for this product on the major B2B review platforms rest on a limited number of reviews, which is not a sufficient basis for a statistical quality conclusion. This page therefore does not quote a score. Treat any single aggregate rating for this product as directional at best.
The free tier and the district deployment are different products in practice. A teacher's positive experience on the free tier does not transfer automatically to a district rollout, where moderation configuration, rostering behaviour, model routing and support responsiveness all become material. Pilot the deployment you intend to buy.
This section exists because in K-12 procurement it is usually the deciding factor, and because the details here are specific enough to be checkable rather than reassuring generalities.
The legal architecture. Under FERPA, the platform positions itself as the processor and the school as the controller: "MagicSchool acts as a data processor when it processes personal data on behalf of its customers, including schools", acting solely on the documented instructions of the educational institution. This is the standard school-official-exception structure, and it means the school retains legal responsibility for the data — a point district counsel will want to confirm they understand rather than assume.
Minors cannot self-enrol. The privacy policy states that "Students under the age of 18 may only access the Service through MagicSchool's agreement with a Customer". Consent obligations therefore sit with the school under its agreement, not with the student. Practically, this means a district cannot avoid the consent question by letting students sign up individually.
Training use is explicitly excluded. The policy states the company "does not use personal information to train artificial intelligence or machine learning models", and separately prohibits using student data to train, fine-tune or improve models. This is the question parents ask most often and the one where a clear negative commitment in the policy text matters more than a marketing line.
Retention has a stated ceiling. Data transmitted through the AI interfaces is "deleted within thirty (30) days" unless a longer period is legally required. A concrete number is more useful than a promise of "limited retention", and it is the kind of figure that belongs in a district's own privacy documentation.
The directory-information loophole is closed. The policy states that the company "does not independently disclose Student Data as 'directory information' under FERPA" — worth noting because directory information is the mechanism through which student data is most often disclosed lawfully but unexpectedly.
Third-party verification exists and has a date. The independent Common Sense Privacy evaluation awarded the product a Common Sense Privacy Verified Seal at 95% as of 12 March 2026, assessed against a rubric of more than 200 questions and based on the published policy documents. Its specific findings include that "Personal information is not sold or rented to third parties", that personalised advertising is not displayed, and that third parties do not collect data for their own purposes. Two caveats belong with this: the evaluation assesses published policies rather than auditing runtime behaviour, and some third-party write-ups still quote an older 93% figure, so check the evaluator's own page for the current number rather than relying on a citation.
The documentation set is unusually complete for this category. The vendor publishes a data protection addendum, a student data policy, a sub-processor list and a VPAT 2.5 accessibility report, alongside claims of SOC 2, GDPR alignment, 1EdTech Trusted Apps certification, and FERPA and COPPA compliance. Enterprise customers get customizable Data Privacy Agreements aligned to local and state requirements. Access to student interaction records is scoped: "Only authorized educators and administrators in your school or district can view these interactions".
What to verify yourself. Certification claims are a floor, not a conclusion. Ask for the current SOC 2 report rather than the badge, read the sub-processor list against your state's requirements, confirm which model providers process your data and under what terms, and confirm the deletion mechanism at contract termination. A vendor with this much published documentation makes those questions easy to ask, which is itself a reasonable signal.
Yes, there is a genuinely free tier with no expiry that includes the full catalogue of teacher and student tools. The limit is on generation volume rather than on which tools you can open, and your remaining usage is shown in the account dashboard with a warning as you approach the cap. For many teachers the free tier is sufficient outside of intensive planning periods.
The individual Plus plan is priced at $8.33 per user per month billed annually, or $12.99 per month billed monthly. What it adds is principally the removal of limits — unlimited generations, unlimited output history and unlimited editing of AI output — along with advanced tool features and access to an experimental features channel. It does not unlock a different or better AI model.
No, according to the company's privacy policy, which states that it does not use personal information to train, fine-tune or improve AI or machine learning models, and separately prohibits the use of student data for that purpose. Data sent through the AI interfaces is stated to be deleted within thirty days unless a longer retention period is legally required.
The company publicly claims FERPA and COPPA compliance along with SOC 2 and GDPR alignment and 1EdTech Trusted Apps certification, and an independent Common Sense Privacy evaluation awarded it a verified seal at 95% as of March 2026. Under FERPA it operates as a processor acting on the school's documented instructions, meaning the school remains the controller and retains legal responsibility. Districts should still request the underlying reports rather than relying on the badges.
No. The privacy policy states that students under 18 may only access the service through the company's agreement with a customer — that is, through their school or district. Consent obligations sit with the institution under that agreement.
Multiple. The company states it uses large language models from OpenAI, Anthropic and Google, selected according to the task, and a published case study describes it having moved many of its tools from OpenAI to Claude. The routing is an implementation detail that can change without any action from the school.
Canvas, Schoology and Google Classroom are supported on the LMS side, and Google, Clever, ClassLink and Microsoft are supported for single sign-on. Clever and ClassLink support in particular indicates the product was designed around US K-12 rostering practice. Deeper SIS and LMS integration is an enterprise-tier capability.
No, and the vendor says so contractually. The terms of service state that AI can produce biased, fictitious or otherwise incorrect content, that output does not replace professional judgement, and that reviewing and revising it before sharing with students is the educator's responsibility. Treat every output as a draft.
Not robustly. The vendor publishes its own outcome figures, including a 28% improvement in literacy grade-level attainment, but these are not independently audited. Independent reporting on the category notes that the effectiveness of personalised learning chatbots has yet to be independently established, and that students' performance can fall sharply once AI tools are withdrawn. Treat the vendor metrics as claims and pilot with your own measures.
Enterprise deployments include a feature the vendor calls Student Room Insights, giving teachers real-time visibility into student AI activity so they can identify where extra support is needed. Access to these records is limited to authorised educators and administrators within the school or district. Schools should disclose this monitoring to students as a matter of policy.
