Skip to content

Custom GPTs in ChatGPT: What Changed and How to Migrate to Plugins

OpenAI is replacing GPTs with plugins. See what changed, how to migrate an existing GPT, a reusable base prompt, and when a real customer service agent on WhatsApp makes sense.

Marlos Carmo

Founder of Tolky

Updated on November 15, 2026

6 min read
Custom GPTs in ChatGPT: What Changed and How to Migrate to Plugins

Share

Updated on November 15, 2026. This post started as a step-by-step guide to the GPT Builder. Since OpenAI is replacing GPTs with plugins, the focus is now what changed, how to migrate, and how to reuse what you already wrote. Always check the official OpenAI help, because dates and plans can change.

What changed for custom GPTs

Custom GPTs (or simply "GPTs") let you shape ChatGPT with instructions, knowledge files and Actions. That model is being retired:

  • Personal accounts (Free, Go, Plus and Pro) can no longer create or publish new GPTs, according to OpenAI's help center.
  • Business, Enterprise and Edu can create them depending on workspace configuration and permissions. The plan formerly called "Team" is now Business.
  • On September 11, 2026, OpenAI announced that GPTs are being replaced by plugins. For Enterprise, creating new GPTs ended on October 26, 2026 and retirement is scheduled for December 11, 2026, with an approved extension to February 11, 2027 for some customers. For other plans, OpenAI says dates "may differ".

Sources: OpenAI migration guide and Virtualization Review coverage.

Similar options exist in other ecosystems, each with its own change cycle: a Project in Claude, a Gem/skill in Gemini and an agent in Microsoft Copilot.

Migrating an existing GPT to a plugin

OpenAI documents what migrates and what does not. In short:

GPT itemWhat happens in migration
InstructionsBecome a skill in the plugin
Knowledge filesBecome reference files
Conversations, selected model and sharingDo not migrate
Custom ActionsDo not migrate; must be rebuilt with an app/connector or MCP server
Resulting pluginStarts private and may answer differently from the original GPT

Practical checklist:

  1. Export the instructions and files of your GPT before the retirement date for your plan.
  2. Rebuild the Actions (OpenAPI) as an app/connector or MCP server, if your flow depended on them.
  3. Retest your prompts: rerun your test questions, because the plugin may answer differently.
  4. Reconfigure access: GPT sharing is not inherited, and the plugin starts private.
  5. Check the interface: menu and tab names change often. If a label mentioned here does not appear in your account, follow the official help.

A portable base prompt: what stays valuable

The most durable part of a GPT is the instruction text. It is a starting point for a plugin, a skill, a Project or a customer service agent. Structure it in blocks:

# PROFILE & OBJECTIVE
You are Company X's support assistant, an expert in helping customers with orders, deadlines and the exchange policy.
 
# TONE OF VOICE
- Professional, direct, and empathetic.
- Never use informal slang or overly long responses.
- If the user asks a complex question, break the answer down into numbered steps.
 
# KNOWLEDGE GUIDELINES
- ALWAYS answer based on the documents in the knowledge base.
- If the answer is not in the documents, politely say: "I couldn't find that information in my manuals. Would you like me to connect you with a human agent?"
- NEVER fabricate links, support emails, or credentials.
 
# CONSTRAINTS
- Do not answer questions outside the company's context (recipes, unrelated code, political opinions).

Tips that work on any platform: consolidate knowledge into a few well-organized files with clear headings, and test failure. Ask something that is not in the base and check that the assistant says it does not know instead of making something up.

Where the GPT/plugin ends and customer service begins

A GPT or plugin lives inside ChatGPT. It works well for individual or team productivity, but it does not serve the end customer in the channel where they already are. The comparison below shows the difference in purpose:

FeatureGPT / plugin in ChatGPTCustomer service agent (e.g. Tolky)
Where it runsInside ChatGPTWhatsApp and webchat (website)
Who uses itYou and your teamThe end customer
Knowledge baseReference files sent to the pluginAgent knowledge base integrated with systems and CRMs
System integrationApps, connectors or MCP server, rebuilt on each migrationIntegrations configured in the platform (talk to the team about your case)
Human handoffNot part of the productTransfer to a human agent with the conversation history
Governance and metricsDepends on the plan and workspace settingsCustomer service operations dashboard

Other channels, such as Instagram, SMS and email, appear as "coming soon" on Tolky's website. See the pricing page and the changelog for what is active today.

When the topic is customer service, the point is not having the "best prompt" but having a stable channel, integration, rules and handoff. Read more in AI governance in customer service, AI without integration becomes a FAQ and how to create a corporate chatbot with generative AI.

The same prompt running as an agent on WhatsApp

The base prompt above can become the starting point of a Tolky agent. The difference shows up in the failure test. An example check:

  1. The customer asks something that is in the base (exchange deadline): the agent answers based on the document.
  2. The customer asks something outside the base (a subject the company does not cover): the agent says it could not find the information and hands off to a human agent with a summary of the conversation.

This repeatable, auditable behavior is what separates a productivity assistant from a customer service channel.

Next step

If you had a GPT that answered questions from customers or partners, now is the time to decide whether migrating to a plugin is enough (internal use) or whether the content should move to an agent that serves customers on WhatsApp and your website. Talk to the Tolky team, bring your prompt and your knowledge base, and watch the agent answer and hand off to a human with your own data.

Share

Tags

  • how to create a custom gpt
  • build custom chatgpt agent
  • gpt builder step by step
  • corporate custom gpt
  • configure chatgpt actions

About the author

Marlos Carmo

Founder of Tolky

Marlos Carmo is an AI entrepreneur and founder of Tolky, the conversational-era infrastructure and AI CRM that unifies intelligent service, multi-channel support (such as WhatsApp and voice), live CRM, and operational intelligence in a single ecosystem. He is a finalist for the SXSW Innovation Awards and a member of Francesco's Economy, a global network of young entrepreneurs focused on innovation and social impact. He works connecting Artificial Intelligence and digital transformation in projects for large organizations.

Next step

Want to bring this to your operation?

Tati, Tolky's AI assistant, understands your scenario in a few questions, suggests the best path and connects you with someone from the team, with full context.

Read also