This guide is for educators and school teams deciding whether an AI tool fits a defined learning task. It focuses on the review process, not on listing services that may open on a school network. For student permission and access steps, see the school AI access and approval guide.
Start with the learning taskDefine what students should learn, what they will do themselves, and what work would show understanding. Then assess whether a particular tool supports that objective safely and accessibly.
Match the tool type to the task
“AI tool” covers services with very different purposes. A writing assistant, research tool, tutor, and coding environment should not be judged by the same output. Start with the task and ask whether AI adds a useful form of practice, feedback, or creation.
| Tool type | Possible learning task | What to review |
|---|---|---|
| Tutor or explainer | Offer another explanation, example, or practice question. | Can students check the explanation against course materials and solve a new problem independently? |
| Research assistant | Organize questions or summarize a teacher-assigned source. | Can students open the original source and verify each claim and citation? |
| Writing feedback | Point out unclear reasoning, grammar, or places needing evidence. | Does the student make and explain revisions, and do assignment rules allow this kind of feedback? |
| Coding or creation tool | Prototype, simulate a system, or explore creative coding. | Can students inspect, revise, and explain the result instead of submitting unexplained output? |
A tool may fit one lesson and not another. A school-wide “approved” label should still be paired with clear boundaries for grade level, data, account, and task.
Build a review record before adoption
The U.S. Department of Education’s 2026 classroom technology guidance frames responsible use around five practical questions: what learning problem a product addresses; when and for whom it should be used; for how long; and what evidence shows a learning benefit. Convert those questions into a short review record that can be revisited:
- Learning objective: State the specific knowledge or skill. Describe what students should be able to do without the tool.
- Intended users: Specify grade band, subject, language needs, accessibility requirements, and whether the service is student-facing or teacher-facing.
- Use boundary: Define permitted tasks, prohibited tasks, teacher oversight, student disclosure, and what work students must complete themselves.
- Access and support: Record account requirements, device compatibility, age restrictions, support needs, and what happens if the service is unavailable.
- Data and privacy: Identify what prompts, files, and account details are collected; who can access them; how long they are retained; and who in the school reviews those terms.
- Evidence and review date: Choose the work sample or assessment that can inform the decision, name the reviewer, and set a date to continue, change, or stop the trial.
The Institute of Education Sciences’ AI Planning Protocols add school-level prompts about decision rights, classroom alignment, privacy and tool risk, readiness, and communication with families. These are planning aids, not endorsements of any product.
Plan a bounded classroom trial
A short local trial can reveal usability and implementation issues. It cannot by itself prove that a product works across subjects, age groups, or schools. Make the trial narrow enough that its result is interpretable:
- Choose one objective. Avoid testing a tool against a vague goal such as “engagement.” Name the lesson outcome.
- Choose a permitted low-risk task. Do not use real student records or sensitive information in an unreviewed service.
- Set the measure first. Use a rubric, student explanation, or work sample that directly reflects the objective. Do not treat logins or time-on-tool as proof of learning.
- Compare thoughtfully. Where practical, compare work completed with and without the tool under similar conditions. Consider differences in prior knowledge and support.
- Collect implementation feedback. Review accuracy, accessibility, teacher workload, time, student understanding, and privacy handling.
- Record limits. Note the subject, age group, duration, task, and anything that makes the result hard to generalize.
Decide beforehand what would justify continuing, changing, or stopping the trial. If students produce faster work but cannot explain it or transfer the skill, that may not meet the learning objective.
Use a decision checklist
Before adoption, a teacher or review team should be able to answer:
- What learning objective does this tool support, and why is it preferable to an available alternative?
- What must students still think through, create, or demonstrate independently?
- Can students with different language, accessibility, or assistive technology needs use it?
- What data does it collect, store, or share, and who has reviewed those practices?
- What teacher support, account setup, or classroom time does it require?
- What evidence would trigger a change or stop decision?
For a student-facing guide to approved chat workflows and answer verification, see the AI chat study guide. This page stays focused on evaluating tools; it does not recommend a particular service or promise access on a school network.
Frequently asked questions
How should a school choose an AI learning tool?
Start with a defined learning task, then review educational fit, student responsibilities, accessibility, data practices, and evidence. Make the decision specific to the intended users and use.
Do usage or engagement numbers prove learning?
No. They can describe how a tool was used, but learning claims need evidence connected to the stated objective, such as student work or assessment results.
Can a short classroom trial prove that a tool works?
A bounded trial can inform a local decision and surface practical benefits or risks. It cannot establish general effectiveness across grades, subjects, or schools.
Should availability determine whether a tool is educationally suitable?
No. Network availability is an access matter. Review educational fit, privacy, accessibility, and assignment rules separately.
