Level 9 · 3 to 4 hours

Build your Gemini Power User system

You will assemble a personal Gemini system: a small set of reusable prompts for research, writing, productivity, and review, plus a quality-control step you run before you trust a draft. This is not a pile of clever one-liners. Each prompt needs a job, an input you will paste, an output shape, and a check. You will chain two prompts on one real task so the second step receives only what you approved from the first. Secrets stay out of the system. The result is a short operating note you can open on Monday and run without rereading the whole course or trusting a chat to remember you.

Outcomes

A master instruction plus separate prompts for research, writing, productivity, and review.
A two-step chain on a real task, with you checking the handoff between the steps.
A quality-control prompt you can run before you keep, send, or publish a draft.

Steps

1. Write the master instruction
Write a short standing instruction: you are the decider, the model is the drafter, and missing briefs get one clarifying question. Ban invented facts, citations, and numbers. Ban secrets. Tell it to stop before any step that would send, publish, book, or delete something. Save this in a note you control. A fresh chat does not remember it unless you paste it.
2. Add four job prompts
Create one prompt each for research, writing, productivity, and review. For every prompt, name the input you will paste, the output headings you want, and the check you will run. Research must leave room for sources you open yourself. Writing must leave room for your voice. Productivity must not invent your calendar. Review must look for unsupported claims before it praises the draft.
3. Chain two steps on real work
Pick a real task that needs two stages, such as notes into an outline, then the approved outline into a draft. Run the first prompt. Delete anything invented or private. Paste only the approved artifact into the second prompt, together with the master instruction. Do not ask one prompt to do the whole job and also check itself without a pause. The pause is the power-user habit.
4. Run the quality-control step
After the draft, run a review prompt that lists unsupported claims, privacy problems, and mismatches with the format you requested. It should not rewrite yet. You decide what to change, then change at least one thing. If the review finds nothing, look again for a fluent line you did not verify. A system that always says the draft is ready is not a quality control.
5. Package the system
Put the master instruction, the four prompts, the chain example, and the review prompt in one note with titles you can scan. Add a line on what never goes into the chat. Test the package by starting a new chat and pasting only what the note says to paste. If you cannot run it from the note, it is not a system yet. Trim until you can.

Starter prompts

Master instruction

You are my drafting partner inside a reusable system. I will paste this instruction at the start of a fresh chat. If the reader, the task, or the output shape is missing, ask one clarifying question and stop. Do not invent facts, citations, quotations, or numbers. Do not ask me to paste secrets, passwords, API keys, or private customer data. Stop before any action that would send, publish, book, delete, or purchase something. When a draft contains a claim I did not supply, label it needs a source instead of smoothing it into the paragraph. Prefer short structured output I can check over a long essay.

Handoff and review

We are on step two of a chain. I will paste only the artifact I approved from step one, plus the job of this step. Use that artifact as the whole source of facts. Do not add examples, numbers, or citations that are not in it. Produce the output in the headings I name. Then stop. After I say review, list unsupported claims, private details, and any mismatch with the requested format. Do not rewrite during the review. Do not say the work is finished or ready to publish. Wait for me to revise one item and show you the change.

Checklist

The master instruction bans invented facts and stops before sending or publishing.
I have separate prompts for research, writing, productivity, and review.
Each prompt names its input, its output shape, and the check I run.
I chained two steps and passed forward only an artifact I approved.
The review listed problems before any rewrite, and I changed at least one thing.
A new chat can run the system from my note, with secrets kept out.

Rubric

The system is a note you can paste, not a memory you hope the model kept.
Each prompt has one job, a defined output, and a human check.
The chain does not smuggle an unverified claim into the next step.
Quality control can fail a draft, and you used it on real work.