Learning Center AI
Leadership & Operations Nov 4, 2026 · 6 min By

Custom GPTs vs. ChatGPT Plugins: What's Changing and How Centers Can Use Them

OpenAI is retiring custom GPTs on December 11 and moving everyone to plugins. What each one is, how a learning center might use plugins, what won't carry over, and three prompts to plan your move.

If your center built a custom GPT in the last couple of years, you're in good company. A BIO 101 review partner. A tutor training bot that plays a frustrated student. A front desk helper that knows your hours and drop-in policy. Custom GPTs made it easy to build a small AI tool without writing any code.

Now OpenAI is retiring them. According to OpenAI's help center, custom GPTs stop running on December 11, 2026, and creators are being moved to plugins. If "plugins" sounds familiar, that's because ChatGPT had plugins once before. They aren't the same thing.

A quick history

  • 2023: the first plugins. One-off connections to outside services like Expedia and Zapier. OpenAI shut them down in April 2024.
  • Late 2023: custom GPTs. Your own version of ChatGPT with instructions, uploaded files and, if you were technical, "actions" that called other systems.
  • 2025: apps and connectors. ChatGPT started connecting to services like Google Drive and Slack through MCP, an open standard for linking AI tools to other software.
  • 2026: the new plugins. In July, OpenAI renamed its App Directory the Plugin Directory. In September, it announced that custom GPTs would retire and move to plugins.

So when people say "plugins" today, they mean the 2026 version.

The difference in one sentence

A custom GPT is a separate chatbot you open and talk to. A plugin is a kit of parts that teaches regular ChatGPT new tricks.

Custom GPTPlugin (2026)
How you use itOpen it from a list and chat inside itInstall it once, then use it in any chat. ChatGPT uses it when your request fits, or you call it with @
What's insideInstructions, knowledge files and optional actionsSkills (instructions plus resources for one task), reference files, and apps that connect to other services
ConnectionsCustom actions, which took some technical setupApps built on MCP, including ready-made ones for services like Google Drive and Slack
SharingA link, your workspace or the GPT StoreShared through your workspace, within whatever your admin allows

The biggest shift is the skill. A skill is a set of instructions for one job, plus whatever it needs to do that job well: templates, examples, a rubric. We wrote about SKILL.md files a few weeks ago, and ChatGPT's skills work on the same idea. Instead of one GPT trying to do everything, you build small skills that each do one thing well, and a plugin bundles the ones that belong together.

How a learning center might use plugins

Here's how some common center GPTs would translate:

  • Course review partner becomes a skill. "Socratic review for BIO 101," with the syllabus and unit outlines as reference files. A tutor or student types @ and picks it, and it asks questions instead of handing out answers.
  • Tutor training scenarios become a skill. A role-play skill with your scenarios and debrief questions. Supervisors can use it right in the chat where they're drafting the training agenda.
  • Front desk FAQ becomes a plugin with reference files. Hours, drop-in rules, the referral list. Update a file once and everyone using the plugin gets the change.
  • Reporting helper becomes a plugin with an app. A skill that knows your report format, plus a connection to the shared drive where your exports live.
  • Ongoing work might belong in a Project instead. If your GPT was really a place to keep files and conversations about one effort, like accreditation prep, a ChatGPT Project may be a better home.

The upside is flexibility. A skill shows up wherever the work is already happening, instead of making people switch to a separate GPT. The downside is that a plugin is less "send students this link and they're set." More on that below.

What won't come over

OpenAI is adding a "Migrate to plugin" option in My GPTs. It turns your instructions into a skill, copies your knowledge files and adds any connected apps. Some things stay behind:

  • Conversations. Past chats stay with the old GPT and go away when it retires.
  • Your model choice. The model you picked for the GPT doesn't carry over.
  • Custom actions. Anything that called another system has to be rebuilt as an app connection.
  • Sharing. A migrated plugin starts out private. Everyone who used your GPT, including students you sent a link to, loses access until you share it again.

That last one matters most for centers. If a student-facing GPT is linked in a syllabus, a Canvas page or a tutor handout, those links stop working on December 11.

Only the latest published version of a GPT migrates, so publish any pending edits first. OpenAI has already stopped new GPT creation on at least some plans (October 26 for Enterprise workspaces), and dates can differ by account, so check the notice in your account or ask whoever runs your campus ChatGPT workspace.

Your checklist before December 11

  1. List every GPT your center made or relies on, who uses it and where it's linked.
  2. Back up each one's instructions and files. Copy the instructions into a document.
  3. Save five or so test prompts for each one, with the answers you expect, so you can compare after the move.
  4. Decide what each one becomes: a skill, a plugin with files, a Project, or retired.
  5. Migrate and test, then share it again and update every link you found in step 1.

One privacy reminder from our privacy post: build and share these in your institution's workspace, not a personal account, and check what any connected app can see before you turn it on.

The point

Custom GPTs showed a lot of us that we could build our own AI tools. Plugins keep that idea and break it into smaller, reusable parts. The deadline is real, though, and the sharing piece is easy to miss. Spend an hour this month taking inventory, and the switch will be quiet instead of a scramble during finals week.

Try It

These prompts work in ChatGPT or whatever AI tool your campus provides. Fill in the bracketed parts before you paste. Leave out student names, ID numbers and copies of student conversations.

1. Build your GPT migration inventory. For directors and coordinators who need to know what their center has built and where it's linked.

You are an operations consultant helping a college learning center prepare for OpenAI retiring custom GPTs on December 11, 2026.

Goal: turn my notes into a migration inventory I can act on.

Context:
- Center: [CENTER NAME] at [COLLEGE NAME]
- Our custom GPTs, one per line, with what each does, who uses it (staff, tutors, students), and where it is linked:
[PASTE YOUR LIST]

For each GPT, give me:
1. Its main job in a few words.
2. A recommended new home: a skill, a plugin with reference files, a plugin with a connected app, a ChatGPT Project, or retire it.
3. One sentence on why.
4. What might break: custom actions, a specific model, shared links, or student-facing access.
5. Every place I listed that will need a new link.

Format the output as a table, sorted most urgent first, with student-facing and widely linked tools at the top. Then list any GPT where you need more information before recommending a home, and what you need.

2. Turn a GPT's instructions into a skill. For anyone who built a custom GPT and wants to rebuild it cleanly instead of relying only on the automatic migration.

You are an expert at writing instructions for AI assistants. I am moving a custom GPT to a ChatGPT skill.

Goal: rewrite my GPT's instructions as one focused skill.

Context:
- What the GPT is for: [ONE OR TWO SENTENCES]
- Who uses it: [e.g., peer tutors, students in BIO 101, front desk staff]
- Files it uses: [LIST FILE NAMES AND WHAT EACH CONTAINS]
- Current instructions:
[PASTE YOUR GPT INSTRUCTIONS]

Give me:
1. A short skill name and a one-sentence description that says when the skill should be used. Write the description so an AI could tell from a request whether this skill fits.
2. The rewritten instructions as clear numbered steps, keeping everything that matters from the original and cutting repetition.
3. A list of the reference files the skill needs.
4. Anything in the original that depended on a custom action, a specific model, or web browsing, and what to do about it.
5. If the original tries to do more than one job, a suggestion for splitting it into separate skills.

Keep the language plain. Flag any new rule you add that was not in my original instructions.

3. Write a test set for your migrated tool. For coordinators and supervisors who want to know the new version works before students and staff rely on it.

You are a quality tester for AI tools used in a college learning center.

Goal: write a short test set I can run against my old custom GPT and its new plugin version to compare them.

Context:
- What the tool does: [DESCRIBE IT]
- Who uses it: [e.g., tutors, students, front desk staff]
- Rules it must follow: [e.g., never give final answers, only use our policy documents, refer students to the counseling center for personal concerns]
- Its instructions:
[PASTE THE INSTRUCTIONS]

Give me:
1. Eight test prompts: four typical requests, two edge cases, and two that test the rules above, such as a student asking for the answer outright.
2. For each, what a good response must do and what would count as a failure.
3. A simple scoring sheet with columns for prompt, old GPT result, new plugin result, and pass or fail.

Write the test prompts the way a real user would type them, including casual wording.

Sources

AI use statement

Sources come from my own reading, plus articles flagged by Elvis, an AI agent I built in Muse (Meta, 2026). I save the ones worth keeping with a note on why they matter for learning centers. Claude (Anthropic, 2026) helps me review those notes, choose topics, and outline and draft the article and its prompts. I revise every draft, check claims against the sources, and decide what gets published.

I tested the prompts in ChatGPT (OpenAI, 2026), Claude (Anthropic, 2026), Copilot (Microsoft, 2026), Gemini (Google, 2026) and Grok (xAI, 2026).

References

About the author

[[shortcode.authors.mark-woolwine]]

← Back to The Feed