PRO 25 min read

Lesson 1 — Pick the Right Job

The hardest part of building a Custom GPT is picking the right thing to build for. Most people pick too broad — here's how to pick well.

Course Custom Gpt · Lesson 1 of 4

The single most common mistake when building a Custom GPT: picking something too broad.

“My personal assistant” — too broad. It tries to do everything, ends up doing nothing well.

“My Tuesday morning content writer that drafts LinkedIn posts in my voice from a topic” — perfect.

This lesson teaches you how to pick a target narrow enough to actually be useful.

The 4 traits of a good Custom GPT target

A great Custom GPT is:

Specific. It does one job, not five.
Recurring. You do this work weekly or daily — not once a year.
Contextual. The same background info applies every time.
Bounded. The output has a predictable shape (a draft email, a 5-bullet plan, a meal idea).

If your idea fails any of these, narrow it.

Examples

Bad target	Why it’s bad	Better version
”My personal assistant”	Too broad	”My morning planner — turns my to-do list into a 6-block day"
"Marketing helper”	Too vague	”LinkedIn post drafter in my voice based on a 1-line topic"
"Code writer”	Too generic	”TypeScript test-writer for my project (knows our patterns)"
"Therapist”	Out of scope + too broad	”Journal companion that asks me 3 reflection questions each evening”

The pattern: the second column has a specific output shape, specific context, and specific frequency.

Step 1 — Brainstorm 5 candidates

Open AI and run this:

I want to build a Custom GPT to save myself time on a recurring task. Help me brainstorm.

Below is what I do most days/weeks. Suggest 5 specific Custom GPT ideas — each one should be a single, narrow job I do repeatedly. For each, name:

- What it does (in 1 sentence) - The shape of the output - The context it would need - How often I’d use it - Estimated time saved per use

Avoid: generic “personal assistant,” “writer,” “summarizer.” Be specific.

MY TYPICAL WORK/LIFE: [describe your job, your responsibilities, your hobbies, recurring tasks — be specific]

Read the 5 candidates. Mark the one that:

You’d use this week
You can clearly describe what “good output” looks like
Has obvious context that applies every time

That’s the one. If two are tied, pick the one that saves you the most time per use.

Step 2 — Stress-test the candidate

Before building, sanity-check:

I’m planning to build a Custom GPT that does [your idea]. Before I invest 30 minutes building it, tell me:

- Is this narrow enough? If not, how should I narrow it? - What context would it need that I’d have to provide upfront? - What’s the most likely failure mode (where would it produce bad output)? - Is this even a Custom GPT-shaped problem, or could a saved prompt work just as well?

If AI suggests narrowing further — narrow further. If it says a saved prompt would work — start there before building anything fancy.

Step 3 — Define “great output”

Before building, you need a target. Write 3 example outputs the GPT would produce.

For my Custom GPT idea ([paste idea]), write 3 example sample outputs of what I’d want it to produce in 3 different scenarios. Make them realistic — what a great real output would look like, not a generic template.

Read them. If you wouldn’t be happy with these as actual output, your idea isn’t ready. Refine.

Step 4 — Inventory the context it needs

What does the GPT need to always know to do the job well?

For a “LinkedIn post drafter”:

Your voice (1-page voice doc)
5 examples of posts you’ve written that performed well
Your audience (1 paragraph)
Topics you talk about / topics you avoid

For a “morning planner”:

Your typical work hours
Your energy patterns
The categories of work you do
Hard commitments you have most weeks

Make a list of what your GPT would need. Most of it is stuff you know but you’d otherwise have to retype every conversation. Fixing that is the whole point.

Step 5 — Decide: Custom GPT or saved prompt?

Sometimes you don’t need a full Custom GPT. A saved prompt (in a Notes app or in ChatGPT’s “Saved prompts” feature) does the job.

Use a saved prompt if…	Use a Custom GPT if…
You use it a few times a month	You use it weekly+
Context fits in one paste	Context is more than 500 words
You’d send it to a friend	It’s specific to you
The output shape varies a lot	The output has a consistent shape

Custom GPTs are for things you’ll use enough that the upfront 30 minutes pays back many times over. Don’t overinvest in Custom GPTs you’ll use once.

Common failure modes

“None of my ideas feel important enough.” → They don’t have to be important. They have to be recurring. Drafting Slack messages to your team is mundane; if you do it daily, a GPT for it pays off massively.

“I keep wanting to make it too broad.” → That’s normal. Broader feels safer (“what if I want to do X too?”). Narrow wins. You can always build a second one for X.

“I can’t decide between two ideas.” → Build the one you’d use tomorrow. The other one will still be there next week.