Chat Picker

Choosing Education AI Tools: Curriculum Design and Student Assessment

This afternoon test plan compares curriculum and assessment tasks using vendor-documented price, limit, and data-control differences rather than outside quality scores.

Sources checked 2 Oct 2026

Choosing education AI tools for curriculum design and student assessment starts with what vendors document about access, limits, data controls, and accuracy. The ChatGPT pricing page, Claude pricing page, and Google AI plans page show different plan structures and usage allowances, but do not establish instructional quality or assessment validity for your materials. Chat Picker has not tested ChatGPT, Claude, or Gemini for this use, so run a controlled comparison instead of treating feature lists as a quality ranking.

Know these limits before you start

As read on October 1, 2026, OpenAI’s ChatGPT pricing page lists Plus at $20 per month, says exact message counts are not published, and describes Free uploads as limited. Anthropic’s Claude pricing page says Pro provides more usage than Free, while Max offers 5x or 20x Pro usage subject to rolling five-hour and weekly limits. Google’s US AI plans page says usage depends on prompt complexity, features, and chat length, and refreshes every five hours until a weekly limit is reached.

Context capacity can affect a test with lengthy documents. ChatGPT’s pricing page lists 27K for Free, 54K for Go and Plus, and 128K for Pro with instant models; its reasoning-model figures are 256K for Go and Plus and 400K for Pro. Claude’s pricing page says context can reach 1M on every plan, depending on the model. The Google plans page does not provide an app token figure, so do not assume one from its storage allowance.

Data settings also differ. ChatGPT’s pricing page lists an opt-out from training on chats for Free, Go, Plus, and Pro. Claude’s page lists an opt-out for Free, Pro, and Max, and says Team is not trained on by default. The Google plans page does not state a chat-training setting, so verify the current controls before submitting student information.

  • OpenAI: ChatGPT can produce incorrect or misleading responses and may sound confident, according to OpenAI’s usage policies and accompanying guidance. Those policies prohibit automated high-stakes decisions in sensitive areas without human review. They also prohibit inferring emotions in educational settings except when necessary for medical or safety reasons.
  • Anthropic: Claude’s Usage Policy directs users to report potentially inaccurate outputs and says a qualified professional must review covered advice, recommendations, and subjective decisions before dissemination or finalization.
  • Google: Gemini’s safety and policy guidelines prohibit factually inaccurate outputs that could cause significant real-world harm, acknowledge limits in training data, and say context—including educational use—matters when evaluating outputs.

The test plan

Run these five tasks in fresh chats, using the plan you would actually consider buying. Use the same device and connection where practical. Keep one log row for each run with: plan, task, input, visible sources, errors or omissions, limit messages, review time, and your final decision. Do not use real student names or identifiable records.

  1. Build a curriculum artifact.
    Input: Your syllabus, one unit, stated learning outcomes, and approved reading list.
    Prompt: “Using only the attached syllabus and unit, create a draft 45-minute lesson plan. Map each activity to a stated learning outcome, preserve the required resources, and list missing information instead of inventing it.”
    Look for: Alignment with the supplied outcomes, invented requirements, unsupported material, and ignored constraints.
    Log: Omissions, unsupported additions, source use, and review time.

  2. Draft an assessment framework.
    Input: Six de-identified student responses and an existing assessment rubric.
    Prompt: “Draft a scoring guide from this rubric and the six responses. Quote the response text behind each proposed judgment, separate evidence from inference, and flag ambiguity instead of assigning a final grade.”
    Look for: Observable criteria, consistent treatment of similar answers, unsupported inferences, and recognition of ambiguous responses.
    Log: Weak criteria, disputed interpretations, missing evidence, and required revisions.

  3. Check source fidelity.
    Input: One dated educational source and a short summary of it.
    Prompt: “Compare the summary with the source. List every factual claim, quote the supporting passage, and mark any claim that the source does not support.”
    Look for: Accurate quotations, correct attribution, preserved qualifications, and unsupported additions.
    Log: Each mismatch and whether the response makes verification easy.

  4. Carry constraints through a workflow.
    Input: The lesson plan, assessment rubric, and three required revisions.
    Prompt: “Revise both documents to meet the three requirements. Preserve everything else unless a change is necessary, and provide a change log for each modification.”
    Look for: Lost requirements, contradictions between documents, unnecessary rewrites, and usable change tracking.
    Log: Manual repairs, duplicated work, and whether the change log was complete.

  5. Check controls, limits, and cost.
    Input: Six fictional student records containing no real names, plus your intended subscription plan.
    Prompt: “Using only these fictional records, suggest three differentiated extension activities and identify the record fields supporting each suggestion. Do not infer emotions, motivation, or family circumstances.”
    Look for: Use of the supplied fields, unsupported assumptions, visible privacy notices, and any usage or file limits shown in the interface.
    Log: Displayed price, billing basis, data setting, limit messages, review time, and your own calculated cost for the intended number of users.

How to read your results

Set pass, revise, or fail conditions before reviewing the outputs. Use four gates:

  • Curriculum traceability: Every major element maps to a supplied outcome or source, with missing information identified rather than invented.
  • Assessment evidence: Judgments connect to the student work, ambiguity is acknowledged, and the assistant does not make the final consequential decision.
  • Source integrity: Quotes, dates, attributions, and factual claims match the original material.
  • Operating fit: The plan accepts the input, the data setting meets your requirements, and the cost and review burden remain acceptable.

Do not average a source or policy problem away with good writing. Compare runs only when they use the same input, task, and plan. If several pass, use your logged review time and actual cost to choose the next pilot. If none passes, revise the test or the workflow before using the output with students.

Where the plans differ

Open Chat Picker’s Claude vs ChatGPT vs Gemini comparison matrix and inspect these rows:

  • Free plan
  • Low-cost tier
  • Main paid plan
  • Heavy-use plans
  • Team plan
  • Context window in the app
  • How usage limits are described
  • Training on your chats
  • Ads
  • API prices

The matrix leaves an unconfirmed figure blank and marks it “Not verified.” Every displayed figure carries its vendor source and read date. Before subscribing, follow the source links: prices vary by country, and plan features and limits can change.

Sources

For current prices and limits, see the dated comparison pages.

Compare assistants