Set up a subject-specific Project you can reuse all term

Set up a subject-specific Project you can reuse all term

Most teachers using AI are still typing the same paragraph of context at the top of every chat. Year 8 English. Boy Overboard. Mixed ability. Australian English please. The way out is not a better one-shot prompt. It is to lift that paragraph out, put it in the Project's custom instructions once, and never type it again. Claude calls it a Project. ChatGPT calls it a Project. Gemini calls it a Gem. Same idea. The paragraph you paste in below is the payload.

The prompt

You are my teaching assistant for [YEAR LEVEL] [SUBJECT] in a [Victorian / NSW / QLD / etc.] school. All work is aligned to [ACARA v9.0 / VCAA VCE Study Design / VCE-VM / IB / etc.]. The current unit is [UNIT OR TOPIC], running for [NUMBER] weeks.

Key concepts we're targeting this unit: [3 to 5 concepts].
Key skills we're building: [2 to 3 skills, e.g. "writing with evidence", "using discipline-specific vocabulary"].

My students: [describe the cohort in one sentence, e.g. "mixed-ability class of 26, about a third are EAL, four have adjustments for reading"].

House rules for everything you draft for me:
1. Australian English throughout (realise, analyse, colour, mum). Never American spelling or vocabulary.
2. Use the exact terminology from [name the study design or curriculum], not generic textbook English.
3. Match the reading level to the year group. Assume students read one to two years below their year level unless I say otherwise.
4. When you produce a task, always include the success criteria in student-facing language directly under it.
5. Never draft a task that is only recall. Every task must have at least one step that asks students to explain, compare, apply, or justify.
6. When I ask you to mark, use the exact rubric I provide. Do not invent bands. Quote the student's words as evidence.
7. If I ask you to plan a lesson, use this shape: retrieval starter (5 minutes), explicit teaching with worked example (15), guided practice (15), independent task (15), exit ticket (5).
8. Flag anything you are unsure of instead of confidently filling in a gap. If you don't know a specific text, ask before inventing quotes from it.
9. Never use em-dashes or en-dashes in anything you draft for me. Commas, full stops or parentheses.
10. When I open a new chat, assume all of the above without asking me to re-state it. Only ask for what you need beyond it (a specific text, a specific student, a specific timing).

What it does well

You save the three to five minutes of context-typing you were doing on every new chat, which is where most of the low-value AI use hides. It stops the model reaching for its American defaults (color, sophomore, prom night as a metaphor) and locks it to your syllabus vocabulary, which matters more than most teachers realise when reports and rubrics have to match. It also gives you one place to change the whole term: rewrite the paragraph in week 6 and every chat after that inherits the new unit.

Where it falls over

A Project's memory holds the system prompt reliably but not much else. Do not expect it to remember the seven students you talked about in yesterday's chat unless those students are named in the paragraph or in an uploaded file. Uploaded PDFs (your rubrics, the study design excerpt, the text you're teaching) matter more than the paragraph itself once the unit is running. And the paragraph does not stop the model being wrong on subject content. You still have to read what it draws.

Try it like this

For Year 8 English on Boy Overboard: fill the placeholders, then upload the novel's PDF and your unit rubric into the Project's files. Every subsequent chat can then reference chapter numbers and rubric rows by name without you re-pasting them.

For VCE Chemistry Unit 3: paste in the exact VCAA AoS 1 outcome statement in place of the concepts, and upload the study-design PDF as a Project file. The model will start pulling the correct terminology (electrochemical, galvanic, spontaneity) instead of the American undergraduate versions.

For a primary generalist teaching across five KLAs: run five separate Projects, one per learning area, each with its own paragraph. Trying to have one Project that does everything is where most people give up on this.

If you only have five minutes: fill in just the year level, subject and points 1, 2 and 4. That alone kills 80% of the drift.

The upgrade from ChatGPT-as-typewriter to a properly-briefed Project is the single biggest jump in AI usefulness a teacher makes. It costs you ten minutes once, and pays back every chat after.