InnerVault

Your cart

Your cart is empty. Add a skill and it waits here. See skills you own

Subtotal$0.00

One payment, secure checkout by Whop. Prices in USD.

My skills

Master OpenAI: Turn ChatGPT Into Your Productivity Engine / Creating and Customising

Custom GPTs and Reusable Assistants

About 15 minutes

A GPT is a custom version of ChatGPT with its own instructions, knowledge files and capabilities. This lesson shows how to build one, and what to use instead if your plan cannot create GPTs.

Check whether you can create GPTs

OpenAI's help pages are not fully consistent on which plans can create GPTs. A current troubleshooting article says personal accounts (Free, Go, Plus and Pro) cannot create or publish new GPTs, while workspace plans such as Business can. Check the "Creating and editing GPTs" and "Troubleshooting GPTs" articles for your plan before you start.

If you cannot create a GPT, use a Project with project instructions and files (Lesson 4). It gives you most of the same benefit for your own work.

What a GPT is made of

According to OpenAI's help centre:

  • Instructions: rules, tone and steps the GPT follows.
  • Knowledge: files it can draw on. The help article gives a limit of 20 files per GPT, each up to 512 MB. It advises putting reference material in knowledge, and rules and tone in instructions.
  • Capabilities: such as web search, image generation, canvas, and Code Interpreter & Data Analysis. Some file types only work if Code Interpreter & Data Analysis is turned on.
  • Apps: connections to other services, where available.

You can build by chatting with the builder, or edit the fields directly in the configure view.

Step 1: Choose one narrow job

Good GPTs do one job well. Examples:

  • "Quote writer": turns job notes into a formatted quote.
  • "Brand voice checker": reviews copy against your style guide.
  • "Onboarding helper": answers new staff questions from your handbook.

Step 2: Write the instructions

Use this structure:

Role: You help [who] do [job].
Inputs: The user will give you [what].
Steps:
1. Check the input has [required items]. If not, ask for them.
2. [Do the main task.]
3. Format the output as [format].
Rules:
- Use only facts from the knowledge files for prices and policies.
- If the answer is not in the files, say so. Do not guess.
- Tone: [tone].

Example for a quote writer:

Role: You turn a tradesperson's job notes into a client quote.
Inputs: Job notes with tasks, materials and hours.
Steps:
1. If hours or materials are missing, ask for them.
2. Price labour and materials using rates-2026.pdf only.
3. Output: client name, itemised table, subtotal, tax line, total, 30-day validity note.
Rules:
- Never invent a rate. If an item is not in the rate card, mark it "price to confirm".
- Plain, friendly language.

Step 3: Add knowledge files

Upload clean, current reference files: rate card, terms, style guide. Remove old versions.

Step 4: Choose capabilities

Turn on only what the job needs. A quote writer may need Code Interpreter & Data Analysis for totals. It probably does not need image generation.

Step 5: Test before sharing

Run at least five test cases, including awkward ones:

Quote for: replace 2 taps, 3 hours. (No materials given.)
Quote for a job type that is not in the rate card.

Check it asks for missing items and refuses to invent prices. Adjust instructions and test again.

Step 6: Share carefully

Sharing options depend on your plan and workspace settings. Share only with people who should see the knowledge files. Treat files in a GPT as visible to its users.

Common mistakes

  • A GPT that does everything. Narrow jobs give reliable results.
  • Rules hidden in knowledge files. Put rules in instructions.
  • No test cases. Test edge cases before anyone relies on it.
  • Sensitive files in a shared GPT. Assume users can see what the GPT knows.

Exercise

  1. Pick one repeated task.
  2. Write instructions using the template.
  3. Build it as a GPT, or as a Project if your plan cannot create GPTs.
  4. Run five test cases and fix one weakness.