How to create a Custom GPT in ChatGPT — and what to actually use it for

ChatGPT is a generalist. Custom GPTs aren’t.

Every time you start a new ChatGPT chat, you’re starting from scratch. No context about your business, no knowledge of how you like to work, no sense of what good output looks like for you. The tool is capable, but it doesn’t know you.

Custom GPTs are different. They’re versions of ChatGPT you build for one specific job — properly briefed with your documents, your instructions, and your context — so every session starts from a much better position. You set them up once and use them repeatedly.

This walkthrough builds a tone of voice reviewer from scratch: a Custom GPT that checks written content against a brand guide and flags anything that’s drifted. It’s a practical example of what these tools can do, and a pattern you can follow for any task you find yourself repeating in ChatGPT.

What’s actually happening under the hood

Standard ChatGPT is designed to be useful to anyone, which means it’s optimised for generality. That’s useful for varied tasks, but it’s a limitation when you need something that consistently works the way your business works.

A Custom GPT solves this by letting you set the context in advance. You define what the tool is for, how it should behave, what constraints it should follow — and, importantly, what documents it should draw on. You can upload your methodology, your brand guidelines, examples of your best writing. The GPT isn’t just following instructions in the abstract; it’s working with your actual materials.

The result is closer to briefing a specialist than asking a general assistant. The difference is real.

When it’s worth building one

The test is simple: are you doing the same task in ChatGPT regularly? Do you find yourself copying the same prompt across sessions, or re-explaining your context every time you start a new chat?

If so, a Custom GPT is likely worth 10–15 minutes of setup time. Common use cases for expert businesses include:

• Tone of voice reviewers
• Proposal or tender reviewers
• Client onboarding guides
• Meeting preparation tools
• Diagnostic frameworks
• Business strategy guides built around your specific approach

The more clearly you can define what the tool should do — and what it shouldn’t — the more reliably it will perform.

Note: Custom GPTs require a ChatGPT Plus subscription or above. They’re not available on the free plan. If you use Gemini, the equivalent feature is called Gems. In Claude, Projects serve a similar purpose.

Building one: the setup process

Access the builder by expanding the GPTs section in the ChatGPT sidebar, choosing Explore GPTs, then Create in the top right.

There are two routes: the Create tab guides you through setup via a conversation, or you can configure everything manually in the Configure tab. For most people, starting with Create is the easiest way in. You describe what you want the GPT to do, it generates a name and instructions, and asks a few clarifying questions to refine the behaviour.

The Configure tab is where the real work happens. This is where you:

• Review and edit the instructions that have been generated
• Upload supporting documents (brand guides, writing examples, methodology)
• Set conversation starters — the prompts that appear when the GPT is opened
• Adjust the name and profile image if needed

The document upload: where Custom GPTs become genuinely useful

This is the step most tutorials skip over, and it’s where the real value lies for expert businesses.

You can upload files directly into the GPT — PDF, Word, plain text. The GPT will reference these when generating responses. For a tone of voice reviewer, the uploads might include:

• A tone of voice guide
• An example of writing that reflects how the business wants to sound
• A brief about the business and its clients

If you don’t have a tone of voice guide, you can create one in a standard ChatGPT chat: upload 20 or so examples of writing you’re proud of and ask the AI to analyse them and produce a codified guide. That document then becomes the foundation for your Custom GPT.

The quality of what you upload directly affects the quality of what comes out. The more precise your materials, the more precise the feedback.

Testing before you commit

Before saving, test on real content — not ideal examples. Paste in something you know is off-brand, or a draft that needs work. Check whether the feedback is specific and actionable, or vague and generic.

If it’s not working well, you don’t need to start over. Flip back to the Create tab and describe what’s missing, and the builder will adjust the instructions. You can also edit them directly in the Configure tab if you know exactly what needs to change.

Once you’re satisfied, click Create. The GPT appears in your sidebar for future use, and any team members with access can use it without needing to know how to prompt well themselves.

Privacy and sharing

When saving, you can keep a Custom GPT private, share it within your team (on Team or Enterprise plans), or publish it to the GPT store for public use.

For most professional contexts, private or team-shared is the right setting. Think carefully about what you upload — particularly if the GPT will be shared. Client-sensitive documents, commercially sensitive materials — these shouldn’t go in without a considered view of how they’re being handled.

And as with any AI output: it still needs your judgement. A Custom GPT is a well-briefed reviewer, not an autonomous decision-maker.

The pattern is worth understanding

The tone of voice reviewer in this video is one example. The underlying pattern — define the task clearly, load it with your materials, test it on real content — applies to a wide range of repeatable work in an expert business.

The question worth sitting with: which ChatGPT task are you doing regularly, in the same way, with the same context, every time? That’s the one worth turning into a Custom GPT.