AI for teachers: lesson plans, assignments and checking student work
How a teacher can prepare materials with a model: tiered assignments, criteria, draft feedback. What to check, what not to delegate and what it costs.
5 min · · бизнес · автоматизация
An AI model helps a teacher not by running the lesson but by taking over part of the routine: three versions of an assignment at different difficulty, grading criteria, draft comments on forty essays. The key condition for it to help rather than harm: the teacher remains the author and the checker. Below: which tasks to give a model, which not to, how to verify facts and what such work costs through the API.
AI for teachers: what the model does well
Tasks work best when you see the result immediately and can fix it:
- Lesson skeleton. Goal, timed stages, discussion questions, homework. You set the topic, grade, duration and what has already been covered.
- Three-level assignments. Basic, intermediate and advanced versions of one topic with different depth.
- Grading criteria. A rubric with four or five criteria and level descriptions for an essay, project or lab.
- Explaining in other words. The same material simpler, with a real-life example or an analogy.
- Draft comments on student work. What is good, what to fix, what to pay attention to. You decide the final comment and grade.
- Text adaptation. Simplify a paragraph, make a short summary, compile a glossary.
What must not be delegated
- Final grades without your review. The model can make mistakes in analysis, especially in math and unconventional solutions.
- Factual material from memory. Dates, quotes, names, formulas and source references can be invented. Check against the textbook and the primary source.
- Curriculum compliance. The model may know standards and curricula inaccurately or out of date. Pass excerpts from your own curriculum as text in the request.
- Decisions about a specific student. Conclusions about progress and behaviour belong to the educator.
How to phrase a request for assignments
A good request has five parts: grade and subject, topic, what has been studied, answer format and constraints. Example:
Grade: 7, subject: physics. Topic: liquid pressure.
Already covered: density, force, pressure of solids.
Make 3 sets of 4 problems: basic, intermediate, advanced.
For each problem: statement, short solution, answer.
Use only formulas from the grade 7 curriculum.
Pick numbers so the answer comes out without rounding.
At the end: a list of typical student mistakes on this topic.Ask for the answer as a table or JSON if you plan to paste it into a document. And be sure to solve a few problems yourself: checking answers in assignments you hand to children cannot be delegated.
Automation: a materials kit in one run
If you need materials regularly, a small script is convenient. A Python example with the OpenAI SDK:
import os
from openai import OpenAI
client = OpenAI(base_url="https://api.zukko.pro/v1",
api_key=os.environ["ZUKKO_API_KEY"])
SYSTEM = (
"You are a methodology assistant for a school teacher. Write clearly, "
"no filler. Do not invent dates, quotes or sources. "
"If unsure about a fact, write: 'check against the textbook'."
)
def lesson_kit(subject, grade, topic, minutes, covered):
task = (For history and other factual topics, use a separate request asking the model to list every date and name from the material. That lets you check them against the textbook in five minutes.
Checking work without losing quality
For essays and short answers a "rubric, then draft comment" scheme works well:
- Give the model the grading criteria and the text of the work.
- Ask for a conclusion on each criterion with a quote from the work.
- Ask for a short, kind comment to the student.
- You read the draft, correct it and set the grade.
Before sending, remove surnames and other personal data from the work: only the text is needed, so replace the student's name with "Work 17". Children's personal data demands extra care, so agree on the use with the school administration and consider the requirements of the personal data law (152-FZ). We do not describe data retention terms of model providers; read them separately.
For oral answers and problems with worked solutions the model is less reliable. When checking math, ask it not to "grade" but to solve the problem independently and then compare the steps with the student's. Comparing with your reference solution is more reliable.
Which model to choose
A cheap model is enough for assignment drafts; for methodology work and complex subjects try a stronger one. Test one topic on several models. Prices at October 18, 2026 (Zukko, per 1M tokens, input / output):
| Model | Price | Strength index |
|---|---|---|
glm-5.3-flash | $0.04 / $0.04 | 42 |
claude-sonnet-5-5 | $0.4 / $2 | 48 |
claude-opus-5-5 | $0.8 / $4 | 58 |
The Intelligence Index does not reflect knowledge of the school curriculum in your subject; it is only a reference point. A test on your topics decides.
What it costs
Formula: (input × input price + output × output price) / 1,000,000, with prices per 1M tokens. Assumptions: a lesson kit is 800 input and 2,000 output tokens, 20 kits per month.
claude-sonnet-5-5: 20 × 800 = 16,000 input tokens × 0.4 / 1,000,000 = 0.0064; output 20 × 2,000 = 40,000 × 2 / 1,000,000 = 0.08. Total about $0.09.claude-opus-5-5: 0.0128 + 0.16, about $0.17.glm-5.3-flash: (16,000 + 40,000) × 0.04 / 1,000,000, about $0.0022.
Draft comments for 30 essays: 1,000 input tokens each (the work and criteria) and 300 output. On claude-sonnet-5-5: 30,000 × 0.4 / 1,000,000 + 9,000 × 2 / 1,000,000 = 0.012 + 0.018 = $0.03. The amounts are small, so choose the model by how useful its drafts are on your students' work rather than by the price difference. Check current prices via GET https://api.zukko.pro/v1/models.
Connecting
Create a key in the dashboard, put a small monthly dollar limit on it and connect following the chat completions reference. If you do not want scripts, you can use the browser chat, and the choice between chat and API is covered in the article browser chat or API. The balance is topped up from $1: SBP (the payment page adds 8.5%), cryptocurrency or a MasterCard.
FAQ
Can AI grade notebooks instead of the teacher?
As an assistant yes, as a judge no. It prepares draft comments and the teacher sets the grade after reviewing.
How do I avoid mistakes in assignments?
Ask for solutions together with answers, solve some yourself and check factual material against the textbook. Read everything you hand to children in full.
Can I send students' work to the model?
Only without personal data and with the school's agreement. Replace names with neutral numbers.
Is AI suitable for creating texts and documents such as work programmes?
For a draft, yes, but you know the formatting and content requirements better. Provide a document sample and check compliance.
zukko
Pay 25×
less
for Claude Code & ChatGPT
One fetch_* key and base URL api.zukko.pro — for Cursor, Claude Code, Codex, and any SDK.
Related articles
- AI for testers: test cases, checklists and bug reportsHow a QA engineer can use a model for test cases, checklists and bug reports, check coverage, and avoid requirements the model made up.
- Anonymizing data before sending it to a neural network: a practical wayYou do not have to send personal data to a model for it to process a document. How to replace it with labels, restore it in the answer and check that nothing leaked.
- A neural network for HR: job posts, candidate emails and resume reviewA model can draft job posts and emails and do a first-pass resume review. What to hand over, how to treat candidates' personal data and how to prevent biased decisions.
- Generating an FAQ with a neural network from tickets and a knowledge baseA good FAQ is built on what customers really ask. A pipeline: group the tickets, answer from the knowledge base with a source reference, and verify the result.