Update, November 2026: OpenAI is retiring custom GPTs, and its help center lists December 11, 2026 as the standard retirement date. Personal ChatGPT accounts (Free, Go, Plus and Pro) can’t create new ones anymore. The steps below still help if you’re editing a GPT you already have before then. I moved all of mine to skills and wrote up what changes and how to move yours.
Ever wish you could build your own version of ChatGPT, one that actually sounds like you, gets your niche, and doesn’t need constant babysitting? You pretty much can, as long as your account still lets you make one. Creating a custom GPT is way easier than it sounds, no code required, and once you have one set up it saves you so much back and forth.
Whether you want it to help with writing, brainstorming, product listings, or just making your daily workflow a little smoother, it all starts with knowing what to ask and how to shape it into something repeatable.
What Is a Custom GPT Good For?
Think of a custom GPT like a very fast learner you’ve trained to work exactly the way you do. Want AI to write like you? It can. Generate YouTube descriptions in your format? Sure. Write listings for print on demand products? Yep. They’re flexible, they don’t require any coding, and once the setup is done you stop starting from scratch every single time.
Quick note: Personal ChatGPT accounts can't create new GPTs, and OpenAI ended new GPT creation for affected Enterprise workspaces on October 26, 2026. Business and Edu workspaces depend on their own settings. Existing GPTs keep working until they retire, and editing one depends on your plan.
How to Build Your First Custom GPT
If your account can still make GPTs, the setup is pretty straightforward. On the ChatGPT website, go to OpenAI’s GPT builder, click Create, and type in what you want your GPT to do. (The mobile app lets you use GPTs, not build them.) Think of it like giving a smart new hire their first set of instructions. Then test it in the preview, tweak it, and save when it feels right. That’s genuinely it for the basics.
I skip the Create tab and go straight to Configure. That’s where you can paste in your system instructions, add conversation starters, and upload files, all in one place without the back and forth. Once you’ve got your instructions written out, Configure is faster.
The Settings That Actually Matter
System instructions are where you set the tone for everything. This is where you tell your GPT how to behave, how formal or casual to be, what to avoid, what format you want, all of it. Be specific here because vague instructions give you vague results. Examples help too. OpenAI’s own tips suggest putting a couple of short examples of good and bad answers right in the instructions, so the GPT sees what you mean instead of guessing.
The Knowledge section lets you upload files your GPT can reference, things like your brand voice guide, product details, writing samples, or FAQs. It’s a good way to give your GPT reference material without cramming everything into the instructions field. OpenAI caps it at 20 files, up to 512 MB each, and says rules and tone belong in the instructions, not the files. Conversation starters are optional but useful if you’re sharing your GPT with someone else or just want to give yourself a quick launch point when you open it.
For capabilities, just turn on what you actually need. Web search if it needs current information, image generation if visuals are part of the job, Code Interpreter if you’re working with files or data, and Canvas if you want to edit longer drafts with it. You don’t need everything on.
How to Build a GPT from a Chat That Already Worked
This is one of my favorite things to do. If you’ve had a conversation where the AI was totally on point, the responses felt right, the tone was perfect, don’t let that disappear. Go back to that chat, scroll to the beginning, and look at what you actually said that worked. Then ask the AI to help you turn it into something reusable.
You can literally say: “Can you help me turn this conversation into a reusable prompt so I can get similar results next time?” And then if you want to go further: “Can you help me turn this into custom GPT system instructions based on what worked in this conversation?” You’re using the AI to help you use the AI better. When it works, it works well.
Prompts That Get Better Results
A few things I’ve found actually make a difference. Telling your GPT to ask clarifying questions before answering cuts down on those confidently wrong responses that come out of nowhere. Adding “answer clearly and concisely” sounds obvious but it genuinely helps. Asking for step by step breakdowns when things get complicated keeps the output usable instead of overwhelming.
Setting the tone explicitly is underrated too. “Keep it casual and conversational” or “sound like a professional coach” gives the GPT a direction that shows up in every response. And using a scenario, like “pretend you’re helping someone who’s never used n8n before,” adds context that changes the output in a good way.
How Do Claude Projects Compare to Custom GPTs?
If you’re using Claude by Anthropic, their version of this is called Projects, and it’s on every Claude plan, including free accounts, which get up to five. Same general idea: you set up a custom space, give it instructions, and it knows what you’re working on. Claude handles longer content and more complex tasks really well, so if you’re working on something that spans multiple prompts or needs a lot of depth, it’s a solid option. You set it up the same way, start a project, add your instructions, upload context files if you have them, and build from there.
| Custom GPT (ChatGPT) | Claude Projects | |
|---|---|---|
| Setup | Instructions, example prompts, and file uploads in the Configure tab | Set up a space, add instructions, and upload context files |
| Best at | A bigger ecosystem of pre-built tools and more sharing options | Longer content and complex, multi-prompt tasks |
| Where to start | OpenAI’s GPT builder, Create or Configure | Start a new Project and add your instructions |
| Can you still make one? | Not on personal plans (Free, Go, Plus, Pro) | Yes, on every plan, up to five projects on a free account |
| Retirement | Standard retirement date is December 11, 2026 | None announced |
What Goes Wrong With Custom GPTs (and How to Fix It)?
Vague instructions are the first thing I’d check. If you tell your GPT to “help with writing” and nothing else, it has no idea what that actually means for you. The more specific you are upfront, the less you have to correct later. The other trap is expecting it to be perfect right away. Your first version will probably need some adjusting and that’s normal. Tiny changes in your prompt or instructions can lead to noticeably better output, so treat it like an ongoing thing you refine over time rather than a one-time setup. Even a well-built GPT can drift. The one I used for my resume handed me a different resume every time I gave it the same job posting, which is why I built a resume tool that cuts the same way every time.
Frequently Asked Questions About Custom GPTs
Do I need a paid ChatGPT account to create a custom GPT?
No. Paying for a personal plan won’t let you make one anymore. Personal ChatGPT accounts, including Free, Go, Plus and Pro, can’t create new custom GPTs, and new creation ended for affected Enterprise workspaces on October 26, 2026. Business and Edu workspaces depend on their settings, and OpenAI’s standard retirement date for custom GPTs is December 11, 2026.
What’s the difference between a custom GPT and a regular ChatGPT chat?
A regular chat starts fresh every time with no memory of your preferences. A custom GPT is pre-configured with your instructions, tone, and any files you’ve uploaded, so it behaves consistently without you having to re-explain yourself every session.
Can I use a custom GPT for my n8n workflows?
You can use what’s inside it. A custom GPT lives in ChatGPT, and the part that makes it yours is the instructions. Paste those into the System Message option on n8n’s AI Agent node and your workflow follows the same rules. If nodes and credentials are new words, the n8n vocabulary guide covers both.
How is Claude Projects different from a custom GPT?
They’re similar in concept: both give the AI standing instructions and files to work from. The big difference now is that custom GPTs are being retired, with OpenAI’s standard date set for December 11, 2026, while Claude Projects are on every Claude plan, including free accounts, which get up to five.
How do I make my custom GPT sound like me?
Upload writing samples in the Knowledge section and write out your voice guidelines in the system instructions. Be specific about tone, what words you avoid, how formal or casual you are, and what you never want it to sound like. The more examples and direction you give it, the less editing you’ll do on the back end. If you’ve moved to Claude, a Claude Code skill for writing does the same job without pasting your style guide every time.
The short version: a custom GPT is a reusable set of instructions, so write down how you want it to work once, test it, and refine it over time instead of re-explaining yourself in every chat. Keep a copy of those instructions outside ChatGPT, since that’s the part you can carry to a skill or a Claude Project when GPTs retire.
A tool I built
Build a Bio
Build a branded link in bio page in about a minute, then embed it on your own blog so your clicks land on your site, not someone else’s.
Build your bio →I run an affiliate program on the tools I build. Approved affiliates earn 20% on what their referrals pay, for up to a year. US only for now. See the program →
Related reading
- Automate Content Creation for the Shiny Object Chaser
- The n8n Vocabulary Guide for People Who Are Not Developers
Subscribe below if you want the next one. I write up what I build, including what broke and how I fixed it. My custom GPTs turned into skills, so what a Claude Code skill is and how to make one is a good place to start in the meantime.
