ChatGPT Custom GPTs: How They Work, Limits, and Migration
Custom GPTs bundle prompts, knowledge files, and tools into tailored bots. Here is how they work, their limits, and how to prepare for the 2026 sunset.
👁 118 reads
Custom GPTs are tailored versions of ChatGPT: you provide permanent instructions, upload reference files, and enable specific tools, launching a purpose-built assistant with a single click instead of pasting repetitive prompts. However, custom GPTs have an expiration date: OpenAI is deprecating GPTs on December 11, 2026, transitioning them into plugins, and new GPT creation is already restricted on personal accounts. Below is a breakdown of what GPTs offer today, how to build them in a workspace account, how to vet third-party bots, and how to prepare before December.
At a glance
Custom GPTs will remain functional until December 11, 2026, after which they will be replaced by plugins carrying over the same instructions and reference files. Since September 16, 2026, only Business, Enterprise, and Edu workspaces can create and publish new GPTs, and all new creation ends entirely on October 26. Free, Go, Plus, and Pro accounts can still run existing GPTs within their regular plan usage limits. If you need a customized ChatGPT setup on a personal account, use Projects instead: they offer system instructions and uploaded files without facing the December sunset.
What are GPTs, and how do they differ from Projects and agents?
A custom GPT is a saved ChatGPT configuration under a unique name: tailored instructions (how to respond, voice and tone, required steps), a knowledge base built from uploaded documents, toggled capabilities (web search, image generation, Canvas, Code Interpreter, connected apps), and optional actions connecting to external APIs. According to OpenAI help documentation, a GPT can hold up to 20 files of up to 512 MB each. Actions and connected apps are mutually exclusive.
Key differences compared to other ChatGPT features:
- Project - a private workspace folder containing custom instructions, reference files, and conversation threads. Available on personal plans, it cannot be shared publicly. Today, this is the closest alternative to a personal custom GPT.
- Agent (agent mode) - an operational mode where ChatGPT actively browses websites and completes multi-step tasks across browser tabs. It is an execution method, not a saved assistant preset.
- Plugin - the updated package format housing skills, reference files, and connected apps. In July 2026, OpenAI rebranded existing ChatGPT “apps” as plugins (distinct from the legacy 2023 plugin ecosystem). Custom GPTs migrate directly into this format.
For foundational usage tips, see our complete ChatGPT guide.
How to build a custom GPT step by step (workspace accounts only)
- Navigate to chatgpt.com/gpts (via “Explore GPTs” in the sidebar) and click Create. This button is only visible in Business, Enterprise, or Edu workspaces with creator permissions enabled.
- Choose your setup mode: the conversational Create builder (where you describe what you need and ChatGPT drafts instructions) or the Configure tab (manual setup).
- Instructions - define the workflow step by step: role, response formatting, negative constraints, and fallback behavior when information is missing. While instructions are hidden from the user interface, adversarial prompting can leak them; never store API keys, passwords, or confidential customer data here.
- Knowledge - upload reference materials (PDFs, spreadsheets, text files, code). Users can prompt the bot to extract or quote these files, so treat any uploaded data as public if the GPT is shared.
- Capabilities and actions - enable only the tools your workflow requires. Custom API actions require a valid OpenAPI schema and a publicly accessible privacy policy.
- Sharing - choose access levels: Only me, Anyone with a link, Workspace-only, or the public GPT Store (if your administrator permits). Finalize and publish before October 26, 2026, when new creation closes and unpublished drafts expire.
The GPT Store: how to find and evaluate third-party bots
When searching the GPT Store (chatgpt.com/gpts), review three critical details before chatting: the author’s identity (their name and verified domain below the title), whether the GPT uses external actions (ChatGPT will prompt for approval before sending data to an external endpoint; avoid simple utility tools requesting third-party server access), and the creator’s privacy policy. The creator cannot view your chat history, but any payload passed to an external action lands directly on their servers. To understand what data OpenAI retains and uses for model training, review our breakdown on AI data privacy.
Limits and privacy across plans
| Plan | Creating GPTs | Using GPTs |
|---|---|---|
| Free, Go | No | Yes, when logged in, subject to standard model caps; access pauses until the rate limit resets |
| Plus, Pro | No (since September 16, 2026) | Yes, with higher usage limits |
| Business, Enterprise, Edu | Yes, until October 26, 2026, subject to admin permissions | Yes; enterprise migration to plugins starts September 22, 2026 |
For detailed rate limits across tiers, consult our guide to ChatGPT limits, or compare subscription features in our breakdown of ChatGPT Plus vs. Pro. Privacy considerations: chats with GPTs on consumer tiers may be used for model training unless opted out in Data Controls; workspace accounts exclude chat data from training by default. If the underlying base model tied to a GPT is retired, OpenAI automatically maps the assistant to a comparable current model.
5 practical use cases for custom setups
You can build the following configurations either as custom GPTs within a workspace account or as private Projects on an individual subscription. Both approaches eliminate repetitive context pasting:
- Style and copy editing - instructions: correct grammar, punctuation, and typos without altering voice or tone; list all revisions in a comparison table. Upload your company style guide and brand glossary as reference files.
- Contract analysis - instructions: extract obligations, key deadlines, indemnities, and unusual liability clauses; flag ambiguities for legal review without providing formal legal advice. Reference file: your standard template terms for comparison.
- Resume and cover letter tailoring - upload your complete career history; set instructions to align phrasing with pasted job postings without inventing experience absent from the source document.
- Targeted language practice - instructions: converse in Spanish at an A2 proficiency level, provide brief error corrections after every response, and trigger a vocabulary review every 10 exchanges.
- Customer support drafting - reference files: pricing tiers, standard operating procedures, refund policy, and product FAQ; instructions: draft responses to customer tickets matching your brand tone, quoting exact policy terms for returns.
What to do before December 11, 2026
- Audit and bookmark the third-party GPTs you rely on, noting their creators. Public GPTs lose their sharing configurations during migration, and creators must re-submit them as plugins.
- If you own custom GPTs: export and back up your system prompts, reference documents, and benchmark test prompts. During migration, system instructions convert into skills, files become reference documents, and connected apps remain apps; custom API actions and hardcoded model selections do not carry over, nor do chat histories.
- Following migration, legacy GPTs remain accessible in read-only mode until their final sunset date.
- On personal plans, rebuild your primary workflows in Projects now to avoid service disruptions.
Frequently asked questions
Are custom GPTs free to use?
Using existing GPTs is supported on Free accounts when logged in, within standard model rate limits. Building new GPTs requires an active Business, Enterprise, or Edu workspace, and all creation ends permanently on October 26, 2026.
Do custom GPTs retain uploaded files permanently?
Files uploaded to a GPT’s knowledge base remain accessible to the bot until manually removed, and their contents can potentially be retrieved by anyone using that GPT. In contrast, documents attached inside an individual chat session remain confined to that specific conversation.
Can you monetize custom GPTs?
OpenAI launched a creator revenue-sharing pilot in 2024 in the US, but it was never rolled out globally. With custom GPTs scheduled for sunset in December 2026, direct GPT Store monetization is effectively closed; commercial value now lies in custom business integrations and consulting, not store payouts.
What is the best alternative to custom GPTs on a Plus account?
Projects are the recommended replacement: they support persistent custom instructions, uploaded reference files, and dedicated conversation histories. While Projects cannot be published publicly, they provide the same functionality for private personal and professional workflows.
Sources and verification date
More on this from QUANTSLANT
news·METR API Key Stolen: $600K in AI Credits DrainedAn attacker tricked an AI agent into leaking a METR API key, draining $600K in free credits. Here is what happened and what teams must learn.