Lucid Brands Home
©Lucid Brands 2026
AI Search & Marketing Automation resource article: 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.

PartWhat it doesExample
InstructionsTells the GPT its role, tone, rules and steps"You are a customer FAQ assistant for a dental office. Answer only from the attached files."
KnowledgeFiles the GPT can search when answeringPrice list, policies, service descriptions, brand guide
CapabilitiesBuilt in tools you switch onWeb search, image generation, data analysis
ActionsConnections to outside apps through an APILook 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.

UseInstructions focusKnowledge files
Brand voice writerTone, words to use and avoid, formatsBrand guide, example posts
Proposal or quote drafterStructure, pricing rules, required sectionsService list, past proposals with client details removed
Internal policy helperAnswer only from documents, say "I don't know" otherwiseEmployee handbook, SOPs
Social media plannerPlatforms, posting rhythm, content pillarsContent calendar, past top posts
Ad copy helperCharacter limits, offer rules, compliance notesOffers, landing pages, ad policies
Customer FAQ trainerAnswer style, escalation rulesFAQs, 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

OptionWhat it isGood for
Custom GPT (being retired)Shareable assistant with instructions, files and actionsExisting workflows until they move
ChatGPT ProjectsA workspace in ChatGPT that keeps files, chats and instructions togetherOngoing personal or team work
Claude Projects (Anthropic)Project workspace with custom instructions and filesLong documents and writing work
Gemini Gems (Google)Custom assistants inside GeminiTeams using Google Workspace
Automation tools with AI stepsWorkflows that run AI on a trigger, such as a new form or emailRepeatable tasks without a person prompting. See marketing automation
A custom app or API integrationYour own assistant built on an AI modelCustomer 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

  1. List every custom GPT your team uses and who relies on it.
  2. Export or copy each one's instructions and knowledge files.
  3. Decide for each: retire it, recreate it in a project workspace, or turn it into an automation.
  4. Rebuild and test with real tasks before the retirement date for your account type.
  5. 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.

  1. Open the GPT builder from the GPTs area in ChatGPT.
  2. Write instructions in the Configure tab: role, goal, steps, rules and an example of a good answer.
  3. Upload knowledge files (up to 20 per GPT, each up to 512 MB, according to OpenAI).
  4. Choose a recommended model if the option is available.
  5. Turn capabilities on or off: web search, image generation and data analysis.
  6. Add actions if it needs to connect to another app through an API.
  7. Test in the preview panel with real requests.
  8. Save and set sharing: only you, anyone with the link or your workspace.
  9. 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 taskReusable assistant idea
ContentA brand voice editor that rewrites drafts to your style
AdsA headline generator that knows your offers and character limits
EmailA newsletter drafter that follows your monthly structure
SalesA proposal helper that uses your service descriptions
ReportingA 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.