How to Create a Custom GPT With Reusable Instructions

The custom GPT maker is the editor inside ChatGPT where you configure a named assistant: give it instructions, upload knowledge files, and set which capabilities it has. The result is a reusable assistant you can share or use privately. Availability depends on your ChatGPT plan; check OpenAI's help center↗ (opens in a new tab) for which plans include the maker.

The instructions you develop in the maker are worth keeping somewhere portable. fenbs stores your approved instruction sets as Skills and makes them available to Claude, Cursor, and other connected AI apps without rebuilding.

Get to the maker

Log in to ChatGPT and navigate to chatgpt.com/gpts/editor. If you do not see the option, your current plan may not include the GPT maker. OpenAI has announced changes to how custom GPTs work; see their help center for the current availability and what changes are coming. The interface may look slightly different depending on when you visit, but the core sections (name, instructions, knowledge, capabilities) have remained consistent.

Name and describe your GPT

Give the GPT a name that describes the task it does, not a generic name like "My Helper." A name like "Contract Reviewer" or "Meeting Notes Formatter" tells users and reminds you what this assistant is for. The description appears in the GPT Store if you publish there, and as the subtitle in the chat interface.

Write the instructions

The instructions field is the most important part. Write these in plain sentences as commands to the assistant. A good set of instructions covers:

1. The main job in one sentence: what the assistant does and does not do.
2. The rules for how it should behave: format, length, tone, and any restrictions.
3. One or two worked examples showing a realistic input and the exact output you want.
4. A fallback instruction for inputs the assistant cannot handle.

Keep the total length as short as possible. A GPT with ten sentences of clear instructions is more reliable than one with ten paragraphs of loosely written guidance. Add more only when testing shows you need it.

Add knowledge and choose capabilities

If the assistant needs to search specific documents (a style guide, a product catalog, a policy document), upload them in the Knowledge section. Plain text and clean PDFs work best. Check each file for content that contradicts your instructions before uploading.

In the Capabilities section, turn on only what the task requires. Web search gives the assistant access to current information but may override your knowledge files. Code interpreter is useful for data work. Image generation is useful for creative tasks. Extra capabilities add scope without benefit if the task does not need them.

Test and publish

Use the Preview panel to do a quick check, then save the GPT as a draft and test it in a real conversation. Send the five to ten inputs you expect most often. If the output is wrong, change one thing at a time: the instructions, an example, or a knowledge file, never all three at once.

When the assistant passes all your tests, open the publish settings. Choose "Only me" for private use, "Anyone with the link" to share with a team, or "Everyone" to list in the GPT Store.

When your GPT needs context that changes often

A custom GPT is a fixed configuration. The instructions and knowledge files are set when you save; they do not update on their own. If the work the GPT supports changes regularly, the GPT will lag behind the real situation.

Say you build a GPT to help write client proposals and the client's requirements shift after each meeting. The GPT does not know about the meeting. It only knows what was in the instructions when you last saved it. You either update the GPT constantly or accept that each draft will need extra editing.

For context that changes week to week, a better approach is to supply the current state at the start of each conversation rather than baking it into the GPT. Paste the relevant update before your first request, or keep the project context in fenbs where a connected assistant reads it automatically at the start of a session without a manual paste.

The parts of your work that change belong outside the GPT. The parts that stay stable, such as format rules, voice guidelines, and required output structure, are exactly what a custom GPT is designed to hold. A well-maintained GPT handles what is consistent; a project record in fenbs handles what is current. Used together, they cover the full range of what most workflows need.

How to do this with fenbs

1. Create a free account and add a project for the task this GPT covers.
2. Write and refine your instructions in a plain text note before pasting them into the builder.
3. Add the finished instruction text as a Skill in fenbs with a note on what each rule is for.
4. After testing, add a Lessons learned entry for any problem you found and fixed.
5. When you update the GPT, update the Skill so the approved version is always in one place.

For a comparison with other platforms, see Claude Projects vs. custom GPTs or ChatGPT Projects vs. custom GPTs.