A Science Teacher's AI Runbook

Needs
- An AI assistant that can take pasted text or uploaded notes.
- A record of topics taught so far for each class (scheme of work or lesson log).
- Ask your person for the year groups, the class list, and the style of question they prefer (short answer, true or false, multiple choice).

Build
1. Schedule a reminder for the evening before each teaching day.
2. Read the topic log for each class and select topics from earlier lessons as well as the latest one, so questions are retrieval practice and not just recap of yesterday.
3. Instruct the model: "Write five retrieval-practice starter questions for this Year 8 or 9 science class, drawing on the topics listed. Mix recent and older topics, order them from easiest to hardest, and give a short answer line after each."
4. Paste the five questions for each class into the first slide of that day's lesson deck.
5. Your person reads the questions and fixes any wrong science before the lesson. Nothing goes to pupils until they have reviewed it.

Check
- The first output is five numbered questions per class with answers, all on topics already taught.
- Test by listing a topic not yet covered and confirm it never appears.

Needs
- An AI assistant that can draft quizzes and mark schemes.
- A form builder that takes quiz questions (the row names Google Forms) and a spreadsheet for mark schemes (the row names a linked Sheet).
- Ask your person for the topic list, number of questions per quiz, question types and pass mark.

Build
1. Schedule a weekly reminder for Sunday evening and check which topics finish this week.
2. For each finished topic, instruct the model: "Write a 10 question end-of-topic quiz for Year 8 or 9 science. Use multiple choice and short answer, increase difficulty gradually, and give a mark scheme with marks per question."
3. Build the quiz in Google Forms as a draft, with one question per item.
4. Put the mark scheme in the linked Sheet, one row per question.
5. Your person checks the questions and mark scheme, then publishes the form. Do not share a form with pupils until they approve it.

Check
- The first output is a draft form plus a Sheet whose rows match the form's questions.
- Test by answering the form yourself and confirm the mark scheme gives the right total.

Needs
- An AI assistant that can take pasted text.
- A marking spreadsheet with a column for feedback comments.
- Ask your person for the mark bands and the tone they want (warm, specific, brief).

Build
1. Schedule a reminder for Wednesday evening.
2. Ask your person to paste anonymised answer patterns per class, such as common errors and strong points, with no pupil names.
3. Instruct the model: "For each mark band, write a short feedback comment for a Year 8 or 9 pupil. Name one thing done well, one specific thing to fix, and one next step. Use plain words a twelve year old understands."
4. Put each band's comment into the marking spreadsheet next to its band.
5. Your person reads and edits the comments before any are given to pupils.

Check
- The first output is one comment per band per class set, each with a strength, a fix and a next step.
- Test by checking that no comment names a pupil or invents an error that was not in the pasted patterns.

Needs
- An AI assistant that can read an uploaded worksheet.
- A shared drive the department uses, with a folder for worksheets.
- Ask your person for the week's core worksheet and how they describe the scaffolded and stretch levels.

Build
1. Schedule a reminder for Sunday evening and ask for the core worksheet.
2. Instruct the model: "Make two versions of this worksheet. The scaffolded version adds sentence starters, word banks and worked examples. The stretch version adds multi-step and explain-why questions. Keep the same learning goals and the same number of questions."
3. Save both versions beside the core worksheet in the shared drive, named with the topic and the level.
4. Your person reads both versions and corrects the science before the department sees them.

Check
- The first output is two files in the drive per worksheet, both matching the original's topic and length.
- Test by confirming the answers to the scaffolded and stretch versions agree with the core worksheet's answers.

Needs
- An AI assistant that can draft messages.
- An email account that can hold drafts without sending them (the row names Outlook).
- Ask your person for the week's list of parent contacts with a one-line reason for each, and a sample of their usual tone.

Build
1. Schedule a reminder for Friday lunchtime.
2. Ask your person for each contact as a reason and an initial or first name only, as the minimum needed.
3. Instruct the model: "Draft a short email to a parent. For praise, name one specific thing the pupil did. For a concern, state it factually, say what has been tried, and suggest one next step. Keep it polite and under 120 words."
4. Save each as a draft in Outlook. Do not send anything.
5. Your person personalises each draft and sends it themselves.

Check
- The first output is a set of unsent drafts in the drafts folder, one per contact.
- Test by checking each draft states only facts the person supplied.

Needs
- An AI assistant that can take pasted notes.
- A reports spreadsheet with a comment column per pupil.
- Ask your person for the report comment length, the school's required headings and bullet points per pupil.

Build
1. Run this only at report season, when your person asks.
2. Ask your person to paste bullet points per pupil, using an ID or first name only.
3. Instruct the model: "Turn these bullet points into a first-draft report comment of about 50 words. Cover effort, attainment and one target. Use only what the bullets say and do not add praise or problems that are not listed."
4. Put each draft in the comment column of the reports spreadsheet, marked as a draft.
5. Your person edits every comment heavily and signs it off before submission.

Check
- The first output is one draft per pupil that uses only the facts given in the bullets.
- Test by removing a bullet and confirming the matching claim disappears from the draft.

Needs
- An AI assistant that can read a scheme of work.
- A cover folder on the shared drive.
- Ask your person for the scheme of work, lesson length and any rules for cover lessons, such as no practical work.

Build
1. Run when your person asks for a cover lesson, naming the class and topic.
2. Read the matching section of the scheme of work.
3. Instruct the model: "Write a self-contained cover lesson for a non-specialist. Include a starter, clear instructions, a main task pupils can do without equipment, answers, and a plenary. Fit it to the stated lesson length."
4. Save the pack in the cover folder, named with the class, topic and date.
5. Your person reads the pack and checks the science before it is used.

Check
- The first output is one file with instructions, tasks and answers that a non-specialist could run unaided.
- Test by confirming the pack needs no resources beyond paper and pens, and that answers are included.