Custom GPT for beginners: build a small AI helper for repeated tasks

Do you paste the same ChatGPT prompt every week? Then you may already have the seed of your first Custom GPT.

This guide continues from ChatGPT Projects or Custom GPTs: which should you start with?. The simple rule there was: use a Project while the work is still changing, and build a Custom GPT once the routine is clear enough to reuse.

Here we do that. We will build a small private GPT for one task: turning messy weekly notes into a short status update. The goal is simpler than an AI employee or a public product: a helper that makes the same kind of work more consistent.

What you will learn in 10–20 minutes

You will learn:

• what a Custom GPT is in practical terms
• when it is better than a normal chat or a Project
• where to find the GPT builder in ChatGPT
• how to fill in name, description, instructions, and conversation starters
• how to test in Preview before sharing anything
• what to leave for later, such as Actions, apps, and GPT Store publishing

We will keep this to one first workflow. If you mainly work with PDFs, course material, notes, and source-grounded study questions, start with the NotebookLM guide for students instead.

What is a Custom GPT?

OpenAI describes GPTs, often called Custom GPTs, as versions of ChatGPT configured for a specific purpose. In the builder, you can work with instructions, conversation starters, uploaded knowledge, and selected capabilities.

Sources: OpenAI Help Center – GPTs in ChatGPT and Creating and editing GPTs

Think of a Custom GPT as a saved work routine. You do not have to write the same long instruction every time, and other people can start from a more consistent setup.

It works best when the task is:

• repeated often
• narrow enough to describe clearly
• easy to test with realistic examples
• dependent on the same tone, structure, or review questions each time

If the work is still messy, start in a Project or a normal chat. A Custom GPT gets better once you know what the helper should actually do.

Before you build: choose one tiny helper

Do not start with "help me with everything in my business." That is too broad.

Choose something smaller:

• reviewing a weekly update before it is sent
• turning meeting notes into a fixed status format
• checking a proposal draft against a short service checklist
• creating five revision questions from notes you are allowed to use

In this guide we use a weekly update. A team lead has a few messy bullets every Friday: what changed, what is blocked, who needs to decide, and what happens next week. The GPT's job is to turn that into a short update and end with a review checklist.

Where to find the GPT builder

1. Go to chatgpt.com and sign in.
2. Open Explore GPTs in the sidebar, or go straight to chatgpt.com/gpts.
3. Choose Create.
4. Build conversationally, or open the configuration view if you want to fill in the fields yourself.

OpenAI Help Center says creating or editing GPTs requires a paid ChatGPT subscription. Building and editing happens on the web; mobile apps can use GPTs but cannot build or edit them.

Source: OpenAI Help Center – Creating and editing GPTs

The workflow: build a weekly update helper

1. Name and description

Give the helper a name you will understand next month. For example: Weekly Update Helper.

The description can be simple:

Helps me turn rough weekly notes into a short status update with decisions, next actions, and a review checklist.

That is enough. A clear name is usually better than a clever one.

2. Instructions

Instructions are the core of the GPT. OpenAI describes them as the place where you guide behavior, tone, goals, format, and boundaries. Put the work rules here, not inside an uploaded file.

Source: OpenAI Help Center – Creating and editing GPTs

Copy and adapt this:

Role:
You help turn rough weekly notes into a short, clear status update.

Input you expect:
- bullet notes
- decisions or blockers
- optional audience or context

How to work:
1. If dates, owners, decisions, or audience are missing, ask up to three questions first.
2. If enough information is present, draft the update directly.
3. Use calm, plain language.
4. Structure the answer as:
   - Summary
   - What changed
   - Decisions needed
   - Next actions
   - Review checklist
5. Do not invent dates, names, promises, or decisions.

Review habit:
Always end by reminding me to check names, dates, numbers, and commitments before sharing.

3. Conversation starters

Add a few starters that show how the helper should be used:

• "Turn these rough notes into a weekly update."
• "Ask me what is missing before drafting."
• "Review this update for clarity and missing decisions."
• "Make this update shorter and easier to scan."

They are not decoration. Good starters help you and other users begin in the right way.

4. Knowledge, only when useful

Knowledge is for reference material: an approved template, a public service description, a cleaned checklist, or class material you are allowed to use. OpenAI says rules, tone, and workflow guidance belong in Instructions.

Source: OpenAI Help Center – Creating and editing GPTs

Do not upload secrets, API keys, customer-sensitive material, or private school material unless you have the right account, policy, and permission.

If a later version connects to apps or APIs, build it with scoped permissions and approval steps. Use logs and clear ownership where your organization’s tooling supports it. That does not mean avoiding integrations, just making them possible to review.

5. Capabilities: start with fewer

Most first GPTs do not need Actions, apps, GPT Store publishing, or every capability turned on.

Use web search only if the task needs current outside information. Use Code Interpreter & Data Analysis only when the helper actually works with files or tables. Use image generation only if the helper creates images.

OpenAI Academy describes Custom GPTs as useful when you often reuse instructions, context, or files. That does not mean you need to connect everything on day one.

Source: OpenAI Academy – Custom GPTs

6. Test in Preview

Run a messy but harmless example:

Here are my rough notes:
- supplier delayed delivery one week
- Anna needs decision on budget by Tuesday
- customer asked for clearer timeline
- send update Friday morning

Make a weekly update for the team.

Check:

• does the GPT ask when important information is missing?
• does it invent facts?
• does it follow the format?
• is the output short enough?
• does it end with a useful review checklist?

Tighten the instructions before you add more capabilities. That is usually where the improvement is.

7. Save privately first

Once the test looks reasonable, keep the GPT unshared or invite-only first. Share by link or workspace only after a few real tests.

OpenAI describes several sharing levels and permissions, but they vary by plan, workspace, and role. The first version does not need to be public. It needs to work.

Source: OpenAI Help Center – Sharing and publishing GPTs

Copy the prompt: plan your first Custom GPT

I want to build a small Custom GPT for this repeated task: [describe the task].

Help me write a first configuration.
Ask up to five questions first if important information is missing.

Then give me:
1. a short name
2. a one-sentence description
3. instructions with clear steps
4. four conversation starters
5. which capabilities I should leave off in the first version
6. three test prompts to run in Preview

Rules:
- Build for one small clear task, not a general assistant.
- Put behavior and format in Instructions.
- Suggest Knowledge only if the GPT truly needs stable reference material.
- Do not suggest Actions, APIs, or GPT Store publishing for the first version.

Common beginner mistakes

• Building a broad assistant instead of a tiny helper.
• Writing vague instructions and expecting consistent output.
• Uploading many files before the task is clear.
• Putting behavior rules in Knowledge instead of Instructions.
• Turning on every capability too early.
• Sharing by link before realistic tests are done.
• Forgetting to edit the GPT when the workflow changes.

Next step

If you are still unsure whether the routine is ready, go back to the Projects vs Custom GPTs guide. If your material is mostly sources, PDFs, and study notes, start with the NotebookLM guide.

Want the next simple AI workflows when they are ready? Subscribe to Hammer Automation's newsletter. If your team wants to do this properly, Hammer can help you choose the first clear task, write the instructions, and set sensible review steps.