ChatGPT Plugins vs GPTs vs Projects (2026): Before GPTs Retire

Key Takeaways

• OpenAI is retiring custom GPTs and moving creators to plugins. Standard retirement is December 11, 2026; FedRAMP workspaces and Enterprise workspaces with an approved deferral retire on February 11, 2027.
• New GPT creation already ended on October 26, 2026 for affected Enterprise workspaces (FedRAMP excluded), and personal accounts (Free, Go, Plus, Pro) cannot create or publish new GPTs. Existing GPTs stay usable until their date.
• ChatGPT uses installed plugins automatically when they are relevant, and one plugin can hold several skills and apps; a GPT is a configured chat you choose, limited to apps or actions, not both. That is why one plugin can replace several GPTs.
• A Project is not a GPT replacement. It is the container for one piece of work, and it pairs with plugins: connected apps work inside project chats.
• Migration (My GPTs > Migrate to plugin) turns instructions into a skill, connected apps into apps and knowledge files into reference files. Custom actions and the selected model do not carry over.
• Storyflow is a plugin, not a GPT: ChatGPT reads and builds on a board the whole team and the client open, so the plan outlives the conversation.

The GPT Retirement, Date by Date

You built a GPT for briefs, another for captions and a third for client updates, and the team uses all three daily. The question is no longer whether to move them but by when, and the dates differ by account type. Everything below comes from OpenAI's Creating and editing GPTs article, its plugins article and its migration FAQ, as read on 28 October 2026.

What stays usable until then

An existing GPT you have not migrated stays usable and editable until its retirement date. Once you migrate a GPT, the original becomes read-only, so treat migration as the moment the plugin becomes the version you maintain. On personal accounts, existing GPTs remain usable even though no new ones can be created or published.

What OpenAI says the replacement is

OpenAI's GPT article says it plans to retire custom GPTs and help creators move their workflows to plugins, which it describes as bringing "reusable instructions, reference files, and connected apps together." The plugins article adds that you may be able to migrate a GPT's published instructions and knowledge into a plugin instead of starting from scratch, and that migration depends on your account or workspace. The Custom GPT retirement and migration FAQ sets out what moves and what does not, covered in the migration section below.

Plugins, GPTs and Projects in One Sentence Each

The three names overlap because all three hold instructions. They differ in where they live and what they connect to.

A plugin is a toolkit you call into any chat

A plugin can include skills (reusable instructions and workflow guidance), connected apps (outside accounts, information and actions), app templates for administrators, and extensions such as in-chat views or forms. ChatGPT uses installed plugins automatically when they are relevant, and you can call one with an @ mention, by selecting + and entering its name, or from Plugins in the composer. OpenAI's developer docs describe a built plugin as skills plus MCP servers, the Model Context Protocol servers that give it tools and access to outside systems, with optional UI.

A custom GPT is a configured chat, now on a retirement clock

A GPT packages instructions, up to 20 knowledge files of 512 MB each, and capabilities: web search, image generation, Canvas, Code Interpreter and Data Analysis, and Apps. One constraint matters for migration: a GPT can use apps or actions, but not both.

A Project is a folder for one piece of work

OpenAI's Projects article describes Projects as keeping related chats, files and instructions together, and project instructions override your global custom instructions inside that project. Projects can be shared, connected apps work in project chats, and shared projects use project-only memory.

Side-by-Side: What Each One Holds

The rows are the ones that decide where a job goes; what each part of an existing GPT becomes is in the migration table further down.

The three "not described" cells are gaps in the articles cited, not verdicts, so check OpenAI's live articles before relying on GPT sharing or project-level admin controls.

A campaign laid out on a Storyflow board: the kind of artifact a GPT returned as text in a chat, and a plugin can build where the team works.

Why One Plugin Can Replace Several GPTs

The familiar setup is one GPT per job: a brief GPT, a caption GPT, an update GPT, each configured separately with its own copy of the instructions and files.

Installed plugins come to the request

OpenAI says ChatGPT automatically uses installed plugins when they are relevant to your request, and you can also call one by @ mention or from +. So the brief format, the caption rules and the board connection can all work in the same conversation: you draft the brief, ask for the captions and update the board without leaving. It is not a better GPT. It is a different shape: a toolkit ChatGPT reaches for on its own, rather than a configured chat you have to pick.

Apps or actions, not both, was the old ceiling

A GPT that needed to read a file store and call your project tracker had to pick one route. A plugin can include several apps, plus skills, so a brief skill and the apps it needs can sit in one package, as in OpenAI's own example of a plugin that gathers progress from Slack and Linear. If an optional app inside a plugin is disabled, OpenAI says the plugin's other capabilities may keep working; a required app that is unavailable stops the related capability.

Plugins come under workspace governance

In Business and Enterprise workspaces, plugins come under their own admin controls: installation policies, role access, allowed actions and approval settings, set out in OpenAI's admin controls article. Our admin guide to ChatGPT plugins walks through each one. For a team, that is the strongest argument for migrating early rather than on the last day.

How to Migrate a Custom GPT to a Plugin

There are two routes, and the right one depends on what the GPT did.

Migrate from My GPTs

Migration runs from My GPTs > Migrate to plugin, where your account or workspace offers it. OpenAI's migration FAQ sets out what moves:

• It uses the GPT's latest published version. Drafts do not transfer, so unpublished edits stay behind.
• Instructions become a skill, connected apps become apps, and knowledge files become reference files.
• Custom actions do not transfer, and the GPT's selected model does not carry over.
• The migrated plugin starts private. The original GPT becomes read-only but stays usable until retirement, and existing conversations stay accessible.

That route suits a GPT that was mostly instructions, files and apps. A GPT whose value sat in a custom action needs that integration rebuilt as an app, and a GPT tuned on one model needs its output checked again, because the model choice stays behind.

Or rebuild with Plugin Creator

Rebuild when the GPT's shape was wrong, for example when its knowledge files mixed several clients. OpenAI's plugins article gives this path:

1. Install Plugin Creator from the plugin directory.
2. Start a chat with @plugin-creator and describe the plugin you want, including any template the skill should follow.
3. Work with ChatGPT to create it. Local plugins install automatically and may not show an install card; install others if prompted, and complete any app connections.
4. Try it on a real task, check the result against your template and, if the plugin can make changes, test those actions before sharing.
5. Use it by mentioning it with @.

You can also start from the directory: go to Plugins and select Add (+) to open a conversation with Plugin Creator. To change a plugin later, mention @plugin-creator with the plugin's name and describe the change.

What each part of a GPT becomes

Test before you share

OpenAI's own example is a weekly-update plugin that gathers progress from Slack and Linear: try it on a real week, check that it reflects the activity and follows the template, then share. Apply the same rule to every migrated GPT. Because a migrated plugin starts private, you can run it on three recent real tasks the GPT handled, compare the output side by side and fix the skill before anyone else sees it.

When a Project Is the Right Container

A Project answers a different question from a plugin. A plugin says how ChatGPT should work; a Project says what it is working on.

What a Project holds

Chats, files and instructions for one piece of work: a client, a campaign, a launch. Project instructions override global custom instructions inside the project, which makes them the right home for rules that apply to one client only, such as "this brand never uses exclamation marks". Shared projects use project-only memory, so context from one client's project does not spill into another's.

Project limits by plan

Invitees get chat or edit access. If a client project needs more files than your plan allows, move the reference material to a connected app instead: OpenAI lists Google Drive files and folders and Slack channels as supported project sources.

Plugins inside a Project

Connected apps work in project chats: select the tools menu (+) or refer to the app by name. OpenAI's connected apps article draws the line between the two words: an app connects ChatGPT to one service, and a plugin packages one or more apps and skills into a workflow. That pairing is what replaces a client-facing GPT. The Project carries the client's files and rules, the plugin carries your house format and the connection to the tools where the work lives, and neither has to duplicate the other. For a full Project setup for creative work, see ChatGPT Projects and apps for creative work.

Decision Rules: Plugin, Project or Both

Use the table to place each job.

Build a plugin when the job repeats across clients

If three account managers run the same brief format for twelve clients, that format is a skill. Putting it in twelve Projects means twelve copies to keep in sync.

Start a Project when the job is one client

If the rules, files and history belong to one client, they belong in that client's Project. The test: would another client's chat be worse for seeing this? If yes, it is project material.

Use both when the deliverable lives outside ChatGPT

A campaign plan the team edits, a deck the client reviews, a shot list the crew opens: these should not live in a chat at all. Work in the client's Project, call the house skill for the format, and have a plugin like Storyflow put the result on a board.

Where Storyflow Fits: A Plugin That Holds the Board

GPTs were good at producing text that the team then copied somewhere. Storyflow's plugin removes the copying: ChatGPT reads and builds on the board where the project already lives.

What ChatGPT can do through it

Through the Storyflow plugin, ChatGPT can list your projects, boards and documents, read a board's elements and layout, summarize it, report what changed on a board in the last day or week and who changed it, and search your workspace. When you allow changes, it builds notes, to-dos, tables, documents, walls and whole sections, mind maps, flowcharts with labelled connections, charts and dashboards, slide decks, carousel pages, shot cards, strip boards grouped into shoot days, colour palettes and ready-made board templates, and it can place Storyflow Tactics, expert frameworks such as AIDA or the Hero's Journey, on the board. It does not generate AI images, read board comments or run the in-app Storyflow AI's menu workflows, and a photo attached in the chat does not reach the board; the plugin gives you an upload link instead.

How to install it

Open the Storyflow listing, select Install plugin and sign in to Storyflow. Choose which projects ChatGPT may use, all of them or only the ones you pick; read access is always on, and you decide whether ChatGPT may also make changes. You can also allow documents outside projects. In a chat, type @ and pick Storyflow, or choose it from the + menu. Availability depends on your ChatGPT plan, workspace and region, and in a Business or Enterprise workspace an admin manages plugins. On Free and Go accounts, OpenAI's troubleshooting article notes that plugin extensions, such as in-chat views, are not available. Storyflow is in the directory, so it does not need developer mode. Full steps are in how to install plugins in ChatGPT, and the in-chat side in how to use plugins in ChatGPT.

What it costs

The plugin runs on ChatGPT's model and your ChatGPT plan, so its actions use no Storyflow AI credits. Storyflow is paid-only during early access, with Plus at $7.99 a month billed annually, Pro at $14 and Max at $39 (pricing), and its Free plan launches before the end of 2026. Plus, Pro and Max have no weekly limit on plugin actions. Anyone a paid member invites to a board joins free, and an invited free account gets 25 plugin actions a week, resetting on Monday (UTC). Connections end after 30 days, after which ChatGPT asks you to sign in again.

Worked Example: Ashgrove & Hart Retires Three GPTs

Ashgrove & Hart is a fictional nine-person agency on a ChatGPT Business workspace, making brand and social campaigns for about a dozen clients. It runs three GPTs and wants all three replaced well before December 11, 2026.

The audit

The audit took the shape it did because of one question asked of every file and instruction: does every client need this, or one client? Everything shared went into the plugin; everything client-specific went into a Project. That is also why Brief Builder is rebuilt rather than migrated: migration would carry all six knowledge files, three clients' guidelines included, into the plugin's reference files.

Building the Ashgrove House plugin

The creative lead installs Plugin Creator, starts a chat with @plugin-creator and asks for a plugin with two skills: a Brief skill that follows the agency's attached brief template (objective, audience, insight, message, deliverables, mandatories, timeline) and an Update skill that drafts a Friday email in five short sections. She tests both on three real past briefs and one real Friday before the account team sees them, then shares the plugin with the workspace.

One client, one Project, one board

For Dunmore Coffee, a fictional roaster launching a spring blend, the account manager creates a Dunmore Project, uploads the brand guidelines and last year's campaign recap, and writes project instructions: "Dunmore's voice is warm and plain, never exclamation marks, always name the origin of the beans." She installs Storyflow, grants only the Dunmore project and allows changes.

The prompt chain

1. "@Storyflow read the Dunmore spring launch board and list what is decided and what is still open."
2. "@Ashgrove House draft the spring launch brief in our format, using the guidelines in this project and the open questions you just found."
3. "@Storyflow put the brief on the Dunmore board as one wall per brief section, with a to-do for every open question."
4. "@Storyflow add a mind map of three campaign routes next to the brief, one branch per route with its hook, channels and the risk."
5. Every Friday: "@Storyflow what changed on the Dunmore board this week and who changed it? Then @Ashgrove House draft the client update from that."

The realistic output shape: step 1 returns a short list in chat; steps 3 and 4 put walls, to-dos and a mind map on the board, each change shown on an approval card if the workspace asks before changes; step 5 returns a list of what changed and who changed it, then a draft email the account manager edits before sending. The account manager still sets assignees and due dates on the to-dos herself, and the client reviews the board through a view-only link that needs no account.

A campaign brief laid out as walls on a Storyflow board, the shape the Brief skill and the Storyflow plugin produce together for Dunmore.

What the old GPTs did not do

The Brief Builder GPT produced the same brief text, but in its own chat, with three clients' guidelines loaded into every conversation, and the result still had to be pasted into a document. The Friday Update GPT summarized whatever the account manager typed into it. The replacement reads the board itself, so the update reflects what the team actually changed rather than what someone remembered.

Common Migration Mistakes

Each of these is cheap to avoid before December 11 and expensive after.

Rebuilding every GPT one to one

Ten GPTs do not need ten plugins. Group them by what they share: one plugin for the agency's house formats, Projects for clients, and connected apps for the tools. One to one rebuilds carry the old shape forward.

Putting client files in the plugin

A GPT's knowledge files were shared with everyone who used it, and migration moves every one of them into the plugin's reference files, so the problem comes across intact. Client-specific files go into that client's Project; a shared Project uses project-only memory, which keeps one client's context apart.

Treating Project instructions as a skill

Project instructions apply inside one project only. A format you want in every project belongs in a skill; a rule for one client belongs in its project.

Leaving the team on the old GPT

After migration the original GPT becomes read-only but stays usable, and the new plugin starts private. A teammate who keeps opening the old GPT gets frozen instructions while you improve the plugin nobody else can see. Share the plugin and announce the cut-over the same day.

Expecting actions and the model to come across

Custom actions do not transfer, the selected model does not carry over, and drafts stay behind because migration uses the latest published version. List each GPT's actions before you migrate it, and plan the app that replaces each one.

Waiting for the date

An unmigrated GPT stays usable and editable until its retirement date, which makes waiting tempting. But the plugin needs testing on real work, and the workspace admin needs time to set policies. Start with the GPT the team uses most.

What OpenAI Has Not Said Yet

As of the articles read on 28 October 2026, several questions have no answer in OpenAI's help centre. Do not fill them with guesses in your team's plan.

• The GPT article gives one standard retirement date and one later date for FedRAMP workspaces and approved Enterprise deferrals; it does not break the standard date down by plan.
• Whether a public GPT Store listing carries over to the plugin directory is not covered; a migrated plugin starts private.
• Whether capabilities such as web search, image generation, Canvas or Code Interpreter carry over is not among the migration facts cited here.
• Plugin sharing is described for Business and Enterprise workspaces; sharing a plugin with people outside a workspace is not described.

The Bottom Line

The GPT era ends on a schedule: December 11, 2026 for the standard date, February 11, 2027 for FedRAMP workspaces and approved Enterprise deferrals. The replacement is not one thing but a split: skills in plugins for what repeats, Projects for what belongs to one client, and connected apps for where the work lives. Audit your GPTs by asking of every file and instruction whether every client needs it or one does, migrate the most used GPT first, and test the plugin on real work before the team switches. If your GPTs mostly produced plans that someone then copied into a deck or a doc, make the board the output: install Storyflow and let ChatGPT build the campaign plan where the team and client already look.