
What Is a Custom GPT?
A custom GPT is a tailored version of ChatGPT that you set up for one job, without writing code. You give it instructions (how to behave), knowledge files (documents it can reference) and capabilities (such as web search or image creation), and it then answers in that role every time. Important: OpenAI has announced it is retiring custom GPTs, with standard retirement on December 11, 2026 according to its help center, so anyone relying on one should plan a move now.
How a custom GPT works
A custom GPT has four building blocks.
| Part | What it does | Example |
|---|---|---|
| Instructions | Tells the GPT its role, tone, rules and steps | "You are a customer FAQ assistant for a dental office. Answer only from the attached files." |
| Knowledge | Files the GPT can search when answering | Price list, policies, service descriptions, brand guide |
| Capabilities | Built in tools you switch on | Web search, image generation, data analysis |
| Actions | Connections to outside apps through an API | Look up an order, check a calendar |
OpenAI's help center says a GPT can hold up to 20 knowledge files, each up to 512 MB. Creators could keep a GPT private, share it by link or publish it in the GPT Store.
What is the difference between GPT and AI?
AI is the broad field. GPT (generative pretrained transformer) is a type of AI model that OpenAI uses to power ChatGPT. A custom GPT is a configured version of ChatGPT built on those models.
What is changing in 2026
OpenAI's help center describes a staged retirement of custom GPTs. As of November 2026 it states:
- Personal accounts (Free, Go, Plus and Pro) can no longer create or publish new GPTs.
- New GPT creation ended on October 26, 2026 for affected Enterprise workspaces.
- Standard retirement is December 11, 2026.
- Some workspaces (FedRAMP or those with a deferral) have until February 11, 2027.
Google's AI Overview for this topic describes the change as a shift toward plugins. Dates and replacement features can change, so check OpenAI's help center for the current status before you make decisions.
What this means for you: if your team uses custom GPTs every day, write down what each one does, save its instructions and knowledge files, and plan where that workflow will live next.
Business examples
Even with the retirement, the idea behind a custom GPT (a reusable assistant with fixed instructions and reference files) is useful and transfers to other tools.
| Use | Instructions focus | Knowledge files |
|---|---|---|
| Brand voice writer | Tone, words to use and avoid, formats | Brand guide, example posts |
| Proposal or quote drafter | Structure, pricing rules, required sections | Service list, past proposals with client details removed |
| Internal policy helper | Answer only from documents, say "I don't know" otherwise | Employee handbook, SOPs |
| Social media planner | Platforms, posting rhythm, content pillars | Content calendar, past top posts |
| Ad copy helper | Character limits, offer rules, compliance notes | Offers, landing pages, ad policies |
| Customer FAQ trainer | Answer style, escalation rules | FAQs, service details |
For customer facing chat on your website, a dedicated chatbot for your website is usually a better fit than a ChatGPT based assistant.
How to write good instructions
These principles work in any assistant tool.
Role: You are [role] for [business], serving [audience].
Goal: Help with [task]. A good result looks like [description].
Steps:
1. Ask for [inputs] if they are missing.
2. Produce [output] in [format and length].
3. End with [next step].
Rules:
- Use only the attached files for facts about our business.
- If you do not know, say so. Never invent prices, dates or policies.
- Write in [tone]. Avoid [words or styles].
Example of a great answer: [paste one]
Test with ten real requests, note where it goes wrong, and tighten the rules.
Privacy: what not to upload
Anyone who can use a shared GPT may be able to get it to reveal parts of its instructions or knowledge files. MIT Sloan's guide also cautions about sharing protected information. Do not upload:
- Customer names, contact details or payment information
- Health, legal or financial records
- Passwords, API keys or login details
- Employee personal information
- Confidential contracts or pricing you would not show a competitor
What are the 5 things you shouldn't tell ChatGPT?
A good rule: never share passwords, personal identity numbers, financial account details, private health information or confidential business data with any public AI tool, and check your plan's data settings.
Custom GPTs vs alternatives
| Option | What it is | Good for |
|---|---|---|
| Custom GPT (being retired) | Shareable assistant with instructions, files and actions | Existing workflows until they move |
| ChatGPT Projects | A workspace in ChatGPT that keeps files, chats and instructions together | Ongoing personal or team work |
| Claude Projects (Anthropic) | Project workspace with custom instructions and files | Long documents and writing work |
| Gemini Gems (Google) | Custom assistants inside Gemini | Teams using Google Workspace |
| Automation tools with AI steps | Workflows that run AI on a trigger, such as a new form or email | Repeatable tasks without a person prompting. See marketing automation |
| A custom app or API integration | Your own assistant built on an AI model | Customer facing use, deeper control, CRM connections |
Features differ by plan and change often. Compare current options before moving a critical workflow.
What to do now
- List every custom GPT your team uses and who relies on it.
- Export or copy each one's instructions and knowledge files.
- Decide for each: retire it, recreate it in a project workspace, or turn it into an automation.
- Rebuild and test with real tasks before the retirement date for your account type.
- Document the new setup so it does not depend on one person.
How to create or edit a custom GPT
If your account still has access, OpenAI's help center describes the builder in these steps. Availability depends on your plan and the retirement timeline above.
- Open the GPT builder from the GPTs area in ChatGPT.
- Write instructions in the Configure tab: role, goal, steps, rules and an example of a good answer.
- Upload knowledge files (up to 20 per GPT, each up to 512 MB, according to OpenAI).
- Choose a recommended model if the option is available.
- Turn capabilities on or off: web search, image generation and data analysis.
- Add actions if it needs to connect to another app through an API.
- Test in the preview panel with real requests.
- Save and set sharing: only you, anyone with the link or your workspace.
- Use version history to roll back changes that make answers worse.
Moving a custom GPT to another tool
Use this worksheet for each GPT before it is retired.
GPT name:
Who uses it and how often:
What it does (one sentence):
Instructions (paste in full):
Knowledge files (list, and where the originals are stored):
Capabilities used (web, images, data analysis):
Actions or connected apps:
Example requests and good answers (3 to 5):
New home (project workspace, other AI tool, automation, custom app):
Owner of the new version:
Date tested and approved:
Recreate the instructions and files in the new tool, run the same example requests and compare answers before switching your team over.
Custom GPT FAQs from business owners
Do my customers need a ChatGPT account to use a GPT? Generally they needed access to ChatGPT, which is one reason customer facing assistants are usually better built as a website chatbot.
Can a custom GPT learn from conversations? No. It answers from its instructions and files. To improve it, you update the instructions or knowledge.
Does a custom GPT keep my files private? Treat anything you upload as potentially visible to anyone who can use the GPT.
Is a custom GPT the same as an AI agent? Not quite. A custom GPT mainly answers and drafts inside ChatGPT. Agents and automations can take actions across tools, such as updating a CRM or sending an email, usually with more setup.
Where custom GPT ideas still fit in your marketing
The thinking behind a good custom GPT is worth keeping even after the retirement:
| Marketing task | Reusable assistant idea |
|---|---|
| Content | A brand voice editor that rewrites drafts to your style |
| Ads | A headline generator that knows your offers and character limits |
| A newsletter drafter that follows your monthly structure | |
| Sales | A proposal helper that uses your service descriptions |
| Reporting | A summarizer that turns monthly numbers into plain language notes |
Each needs the same three things in any tool: clear instructions, good reference files and a person who checks the output. Our guide to AI marketing tools covers options.
Whatever tool you move to, keep a written copy of each assistant's instructions and files in a shared folder your team controls, so you are never locked into one provider again.
FAQ
What is a custom GPT?
A custom GPT is a version of ChatGPT set up for a specific task with its own instructions, knowledge files and capabilities, created without code.
Are custom GPTs going away?
OpenAI's help center describes a retirement, with personal accounts already unable to create new GPTs and standard retirement on December 11, 2026. Check OpenAI's current guidance, as timelines can change.
Which is the best GPT model?
It depends on the task and changes often as new models are released. For most marketing writing, the current default model in your AI tool is a sensible starting point; test a few on your own tasks.
Is there better AI than GPT?
Several providers, including Anthropic (Claude) and Google (Gemini), offer strong models. Each has strengths, and many businesses use more than one. Test them on your own work.
Build AI workflows that last
Lucid Brands' AI Process Integration service starts at $250 a month. We map the tasks your team repeats, move important assistants off tools that are being retired, and build AI workflows that connect to your CRM and marketing. Contact us or read our guide to AI for small business.
Your path to lucidity
Get Started With a Free Website Audit
Not sure where to begin? Start with a free website audit. Our team analyzes your website's SEO, speed, content and user experience to identify what to improve first.
Explore More
Related Resources

AI Video Generators for Marketing: What Works in 2026
What AI video generators do well and badly, the main tool types, how to compare cost per second, commercial use and copyright checks, prompt templates by use case and a test plan for marketing teams.
Read More
Answer Engine Optimization (AEO)
What answer engine optimization (AEO) is, how it differs from SEO and GEO, how AI tools choose answers, a before and after rewrite, a small business checklist and how to track it.
Read More
GEO vs. SEO: What's the Difference
GEO (generative engine optimization) aims to get your content cited in AI answers; SEO aims to rank in search results. See a side by side comparison, the overlap, a small business priority checklist and free ways to measure GEO.
Read More