How to Use AI to Draft a Tutoring Lesson Plan Without Replacing the Tutor
Direct answer: Use AI as a structured drafting assistant, not as the teacher. Give it a clear learning objective, the lesson length, the learner's non-identifying needs, and the materials you have already chosen. Then check every suggestion against the objective, verify explanations and examples, adapt the sequence to the learner, and approve the final plan yourself. The useful output is a reviewable first draft—not an automatic diagnosis, curriculum decision, or guarantee of learning.
This distinction matters because lesson planning is more than filling a template. A tutor must notice misconceptions, decide what evidence will show understanding, choose an appropriate pace, and respond to what happens in the session. Current U.S. Department of Education guidance describes AI as potentially useful for high-quality instructional materials and high-impact tutoring while also emphasizing privacy, responsible adoption, and engagement with affected stakeholders, including parents.[1] The workflow below keeps those human decisions visible.
What AI should and should not do
AI is well suited to transforming information you supply into alternatives: a sequence of activities, question variations, a short practice set, or a checklist of possible misconceptions. It can also help you compare two lesson structures or make an explanation more concrete. These are drafting and organizing tasks.
AI should not be treated as the authority on what a learner needs. It may misunderstand the objective, invent a fact, use an example that is culturally or developmentally unsuitable, or recommend an activity that looks polished but does not produce useful evidence of understanding. A generated plan also cannot replace your observation of the learner or your responsibility to make a professional instructional judgment. Treat each output as a proposal that must earn its place in the lesson.
The five-part workflow
1. Define the lesson before opening the AI tool
Start with a compact planning brief. Write down the subject, specific objective, approximate learner level, session length, available materials, and what the learner should be able to say, solve, create, or demonstrate by the end. Include known instructional constraints such as “needs frequent retrieval practice” or “benefits from visual models,” but avoid unnecessary personal detail.
A strong objective is observable. “Understand fractions” is broad; “compare two fractions with unlike denominators using a visual model and explain the comparison” gives you something to teach and check. If the objective cannot be assessed in a few minutes, narrow it before asking for a plan.
2. Ask for a draft with explicit boundaries
Give the model a role that is limited and inspectable. For example:
“Act as a lesson-planning assistant. Draft a 45-minute tutoring session for the objective below. Include a brief warm-up, one worked example, guided practice, independent practice, two checks for understanding, likely misconceptions, and an exit task. Do not diagnose the learner, claim that the plan will produce a particular result, or assume access to materials not listed. Mark any uncertain suggestion for tutor review. Use plain language and explain why each activity supports the objective.”
Then add your planning brief. Request a table or labeled sections so that the output is easy to audit. Ask for multiple options only where a genuine choice exists—for example, a paper-based and a manipulatives-based version. More output is not automatically better; the goal is a manageable draft that exposes its assumptions.
3. Check alignment, accuracy, and feasibility
Read the draft in three passes. First, check alignment: does every major activity help the learner practice the stated objective? Remove attractive activities that are merely adjacent to the topic. Second, verify accuracy: solve examples yourself, check definitions, inspect answer keys, and confirm that the suggested progression does not smuggle in an unintroduced concept. Third, check feasibility: can the lesson fit the time, materials, learner level, and setting?
Pay special attention to generated “common misconceptions.” They are hypotheses, not observations about your learner. Convert them into questions you can use to investigate understanding: “What makes these denominators different?” or “Show how you know.” Do not present a model's guess as a diagnosis.
4. Add differentiation by changing support, not the objective by accident
Once the core plan is sound, ask for adaptations that preserve the objective while changing the route. Useful dimensions include representation, amount of scaffolding, response format, and practice spacing. For example, a learner might first compare fractions with area models, then move to a number line, and finally explain the comparison symbolically.
Have the AI label each adaptation and explain its purpose. This makes it easier to reject suggestions that lower the intellectual target without a reason, introduce a new prerequisite, or create an accessibility barrier. You remain the person who decides whether an adaptation fits this learner and this session.
5. Run the lesson and revise from evidence
Before the session, mark three tutor checkpoints: the opening diagnostic question, the mid-lesson check, and the exit task. During the lesson, record brief evidence such as “chose the correct model but reversed the comparison symbol.” Afterward, revise the next plan from what the learner actually did, not from the original AI assumptions.
A useful revision prompt is: “Here is the objective, the activities used, and anonymized evidence from the learner's responses. Suggest three possible next steps, explain the instructional reasoning for each, and identify what the tutor should verify before choosing.” The final sentence is important: it keeps the model in an advisory role.
A review checklist you can reuse
Use this decision tool before a generated plan enters your tutoring notes:
- Objective: Can I state the intended learner action in one sentence?
- Evidence: Does the plan include at least one task that would reveal whether the objective was met?
- Accuracy: Have I independently checked the explanations, examples, and answer keys?
- Fit: Is the sequence realistic for the time, materials, age or level, and learning context?
- Misconceptions: Are possible errors framed as questions to investigate rather than claims about the learner?
- Differentiation: Does each adaptation have a clear instructional purpose?
- Human approval: Have I edited the plan and decided what to keep, change, or discard?
- Data minimization: Did I leave out names, contact details, school identifiers, grades, disability information, and other unnecessary personal information?
If any answer is “no,” keep the draft in review rather than using it as-is. This checklist is an original planning aid, not a substitute for a school, district, platform, or professional policy.
Privacy and student information: use a conservative default
Do not paste identifiable student information into a consumer AI tool merely to make a prompt more specific. Use a neutral description such as “a middle-school learner who is confident with visual models but needs practice explaining proportional reasoning.” Remove names, email addresses, exact school details, student IDs, screenshots, and free-form notes that could identify a child.
Rules and institutional policies can depend on the service, the information collected, the child's age, and the educational setting. The FTC says COPPA applies to operators of online services directed to children under 13 and to services with actual knowledge that they collect personal information online from a child under 13.[2] Its compliance guidance explains that, as a general rule, covered operators must obtain verifiable parental consent before collecting personal information online from children under 13.[3] These are general information points, not a determination that a particular tutoring arrangement is covered or compliant.
If you work through a school, nonprofit, tutoring company, or platform, check its current approved-tool list and data-handling instructions before using AI. Ask what information may be entered, whether prompts are retained, who can access outputs, and how deletion works. When the answer is unclear, use no student-specific information and consult the responsible administrator or a qualified privacy professional about the applicable rules.
Prompt patterns that improve reviewability
Instead of asking “Make me a lesson plan,” ask for a plan with visible reasoning and limits. Useful additions include: “separate facts from assumptions,” “list what the tutor must verify,” “provide one simpler and one more challenging example,” “include a misconception check after each worked example,” and “do not use personal data.” You can also ask the model to critique its own draft, but treat that critique as another unverified output.
Keep a small library of prompts organized by task: first draft, differentiation, question generation, misconception probes, and post-session revision. Save the human-edited version separately from the raw output. This creates a clear distinction between machine suggestions and the plan you actually approved.
Common failure modes
The plan is too generic. Narrow the objective and provide the actual materials and time limit. The plan is overstuffed. Remove activities until the learner has enough time to practice and explain. The examples are wrong or misleading. Check every one manually and replace them. The “personalization” is invented. Strip out assumptions and gather evidence in the session. The plan quietly replaces tutor judgment. Restore explicit approval points and write down what you observed. The prompt contains sensitive information. Stop, remove it, and follow the relevant organization’s current guidance before continuing.
Bottom line
The safest and most useful mental model is “AI drafts; the tutor decides.” Start with a measurable objective, request a bounded and transparent first draft, verify content and feasibility, adapt from evidence, minimize student data, and approve the final plan yourself. Used this way, AI can reduce blank-page work without turning a learner into a data profile or a generated suggestion into an unquestioned teaching decision.
