Free to read · no account needed

16 min · chatgpt

Custom GPTs for repeat work (without turning into spam)

Goal: Build a small custom assistant for one recurring job—with clear instructions, limited knowledge files, and honesty rules.

Best for: Weekly status, FAQ drafts, study quizzer—from *your* materials.

Custom GPTs (name may evolve) package instructions + optional files. Keep scope narrow. Do not upload secrets. Prefer your own docs over scraping the web into a megabot.

A custom GPT is a reusable job hat—not a second brain for everything. One job, short instructions, one test prompt, clear never-do list.

Why beginners skip this

A vague custom GPT becomes a spam machine. Narrow scene, output format, and never-do—trial three real tasks before you trust it. Mid-week cost shows up as rework, not as a missing bookmark.

Step 1 — One job definition

Do this step with a real artifact open—not from memory.

Custom GPT job: [one sentence]
Instructions:
- Audience / tone
- Output shape
- Unknown rule
- Never request secrets
Knowledge: [1–3 redacted files max]

Step 2 — Example

Example

Filled prompt

Live task this week (not a toy): apply this guide's Goal — Build a small custom assistant for one recurring job—with clear instructions, limited knowledge file
Fill with your real artifact names/files. Constraints: no invented facts; label Unknowns; stop before send/delete/pay.
Return the working prompt + one pass/fail check I can mark in a minute.

What good output looks like

  • Instructions ≤15 lines with audience, format, and never-invent-metrics
  • Includes a concrete weekly test prompt you will actually re-run
  • Job is singular (e.g. weekly status)—not therapist+coder
  • Knowledge files listed are redacted and on-job only
  • You could publish the card to a teammate safely

Step 3 — Retire or version

If the GPT drifts, update instructions like a prompt card version bump—or delete and remake narrower.

Scene: Monday in five minutes

You open your status-update GPT, paste bullets, get 150 words, paint the traffic light yourself, send. No mega-bot—just one job that respects your metrics ban.

Keep it boring on purpose

Resist adding a second job to the same GPT (“also write LinkedIn”). Split instead. Boring single-purpose GPTs compound; clever mega-bots rot.

Propose whether to split my Custom GPT into two. Current job: [ ]. Tempting add-on: [ ]. Recommend split vs keep with 5 bullets.
Example

Filled prompt

Custom GPT job: weekly status polish for eng manager.
Instructions: Done 4 boxes, never-do (invent %, soft asks), one golden example.
Test twice on real bullets. Fail if a new metric appears.

What good output looks like

  • Split into two named GPTs with one-line jobs each
  • No leftover kitchen-sink capabilities from the old mega-GPT
  • Each has its own ≤15-line instruction stub
  • Migration note names which chats move to which new GPT
  • Never-invent-metrics line appears in both new instruction stubs

Save this card

Card: One GPT per repeat job; short instructions; weekly test prompt; never invent metrics—check it once this week, or it stays a bookmark. Keep this beside the tool; read before you start.

1. What I shipped: ___ 2. What I refused: ___ 3. What I will never-do again: ___ (date: Asia/Shanghai)

  • [ ] Tools/writes paused
  • [ ] Refuse-line dated
  • [ ] Artifact path saved
  • [ ] Next link chosen (or stop)

Before you start

Open ChatGPT on the account you keep. Pick the live thread or Project for “Build a small custom assistant for one.” Have the messy input in front of you—not a blank box. Guide: chatgpt-custom-gpts-for-repeat-work. If you only have a blank practice box, stop and fetch the real job first.

  • Goal in one line: ___
  • Allowed tools: ___
  • Forbidden: ___
  • Human approve needed for: ___

How to use this week

This week on ChatGPT: build or trim one Custom GPT for a single recurring job; short instructions; weekly smoke test; never invent metrics on a live artifact—not a blank demo. Write one refuse-line and date the note (Asia/Shanghai). If you only skimmed chatgpt-custom-gpts-for-repeat-work, put 25 minutes on the calendar and redo once.

One job hat

A GPT that is marketer+coder+therapist is a kitchen sink. Split.

Test prompt is part of the product

If you cannot name a weekly test ask, the GPT’s job is fuzzy. Store the test prompt next to the instructions. When output drifts, fix instructions—not your memory of “how it used to answer.”

Example

Filled prompt

Custom GPT for weekly status polish drifting into "strategy advice."
Rewrite instructions: one job, Done 4 boxes, never-do invent %, golden example.
Retest twice.

What good output looks like

  • Uses only facts you typed for “Build a small custom assistant for one recurring j” — no invented owners
  • Format matches your asked shape (bullets/table/length) on first try
  • Unknowns marked [?]; nothing filled to sound confident
  • You can mark pass/fail in ≤60s for chatgpt-custom-gpts-for-repeat-work drill #1
Example

Filled prompt

GPT for FAQ tone drifts into legal advice.
Rewrite: one job, Done 4 boxes, never-do legal claims, golden test twice.
Context: real workstream this week (Asia/Shanghai). Name actual files/people.
Constraints: no invented metrics/owners/prices; label Unknowns; stop before send/delete/pay.
Done check: I can mark pass/fail in one minute on the output.

What good output looks like

  • Format matches your asked shape (bullets/table/length) on first try
  • Unknowns marked [?]; nothing filled to sound confident
  • You can mark pass/fail in ≤60s for chatgpt-custom-gpts-for-repeat-work drill #2
  • Uses only facts you typed for “Build a small custom assistant for one recurring j” — no invented owners

Red flag

Instructions full of adjectives, no format; one GPT owning ten unrelated jobs; output reads like spam—narrow or kill.

Pitfalls

PitfallFix
One GPT for every life taskSplit by job; kitchen-sink fails
GPT instructions full of secretsPoint to sanitized examples only
Mega-GPT for all lifeOne job
Uploading customer exportsRedact / don’t
No test conversationRun a golden example

Practice now

Live card: one ChatGPT project/GPT/canvas action for “Custom GPTs for repeat work (without turning into spam)” with refuse written first (Asia/Shanghai).

Write standing instructions for one narrow repeat job (inputs, output format, never-do). Trial it on three real tasks today. Date keep/kill (Asia/Shanghai).

Today card (copy) — chatgpt-custom-gpts-for-repeat-work: 1. Artifact I will produce: ___ 2. Pass/fail check: ___ 3. Refuse-line if tempted to skip verify/send: ___ (date: Asia/Shanghai) 4. One never-do from this guide: ___

  • [ ] Artifact path written
  • [ ] Refuse example would say "yes"
  • [ ] Dated Asia/Shanghai

One sentence takeaway

One GPT per repeat job; short instructions; weekly test prompt; never invent metrics—check it once this week, or it stays a bookmark

  • You can point to a dated refuse-line
  • You have one artifact path from tonight
  • You can teach the red flag in one sentence

Next

Version templates · Projects as workspaces · Skills vs GPTs · Clear everyday prompts

Pair with today: open the first Next link (prompt-templates-you-can-version) only after you dated a refuse-line for this guide (Asia/Shanghai).

← Back to home · Learn · Tools · Compare · Start