Skip to main content
Everything PR News
AI Communications

ChatGPT Plugins vs Custom GPTs for PR Agencies

EPR Editorial TeamEPR Editorial Team6 min read
Share
ChatGPT Plugins vs Custom GPTs for PR Agencies
ChatGPT Plugins vs Custom GPTs for PR Agencies

Part of The AI Communications Hub · companion to ChatGPT for Public Relations

OpenAI plans to retire custom GPTs, and its help center says new GPTs can no longer be created on personal ChatGPT accounts (Free, Go, Plus and Pro). PR agencies that want a reusable client assistant should now build it as a ChatGPT plugin. A plugin bundles reusable instructions, called skills, with connected apps, and workspace administrators control who can create and share it.

What is happening to custom GPTs?

Custom GPTs are being phased out in favor of plugins, according to OpenAI's help center page on creating and editing GPTs. The page says new GPT creation and publishing are not available on personal accounts, while Business, Enterprise and Edu workspaces can still create, edit and publish GPTs when workspace settings and permissions allow it. Existing GPTs remain usable until retirement.

OpenAI recommends moving GPT workflows to plugins and says a migration flow is planned. The dates it lists are planned, not final, and they can differ by account, so check the notice inside your own workspace.

DateEvent
Sep 17, 2026Migration experience targeted, and it may not reach every account at the same time
Dec 11, 2026Planned retirement of custom GPTs in affected Enterprise workspaces
After Dec 11, 2026Other plans expected to follow the same timeline, per OpenAI

OpenAI also notes that public GPTs created in affected Enterprise workspaces are included in the retirement, even when someone uses them from another plan. An agency that shares a public GPT with clients should plan for that link to stop working.

What is the difference between a custom GPT and a plugin?

A custom GPT is a single assistant made of instructions, up to 20 knowledge files of up to 512 MB each, and optional actions that call outside APIs. A plugin packages a whole workflow and can include skills, connected apps, app templates and extensions, according to OpenAI's help page on plugins in ChatGPT.

A skill holds reusable instructions and workflow guidance, and a plugin can contain only skills with no app connection at all. For an agency, a skill-only plugin is the closest match to the old client-specific GPT: the brand voice, banned language and output formats live in the skill, and nothing needs to connect to an outside account. OpenAI's own example attaches a template to a skill so every output follows the same format.

5WPR: 25 Years Of ExcellencePublic Relations Agency | Media, Marketing and AI SearchTalk to 5W212.999.5585info@5wpr.com

How should a PR agency build a client assistant as a plugin?

An agency builds a client assistant by gathering the client's reference material, writing it into a skill, testing it against real drafts and sharing it with the account team. The steps below adapt the old custom GPT workflow to plugins.

  1. Gather the reference material first. Collect the brand voice guide, three to five sample releases that match the voice, the messaging architecture, the approved spokesperson list and any banned language.
  2. Decide whether the assistant needs an app. A drafting assistant usually needs only a skill. Add a connected app only if the assistant must read from a service such as Slack or a shared drive, and remember that the app keeps its own permissions.
  3. Start in Plugin Creator. Install Plugin Creator from the plugin directory, then start a chat with @plugin-creator and describe the assistant, for example a drafting assistant for one client that follows an attached release template.
  4. Put rules in the skill and examples in templates. OpenAI's guidance for GPTs separates behavior from reference material, and the same split works here: tone, hard rules and formats go in the instructions, and sample releases go in as reference.
  5. Test with a real request. Ask for the kind of draft the team writes weekly and compare it with the client's published material. OpenAI advises reviewing and testing a plugin before sharing it.
  6. Share it with the account team. New workspace plugins start private. The owner can share to people who are invited or to anyone in the workspace with the link, and publishing to the workspace directory is a separate permission.
  7. Name an owner for updates. To change a plugin, mention @plugin-creator and the plugin and describe the edit. Assign one person per account so the assistant does not drift from the client's current positioning.

Which workspace permissions does an agency need?

An agency needs its administrator to enable plugin permissions, because plugin creation is controlled at the workspace level. OpenAI lists five role permissions: use plugins, upload plugins, create plugins with MCPs, share plugins and publish plugins to the workspace. For Enterprise, use plugins is on by default, and the upload and MCP permissions are off by default.

Administrators can also set an installation policy for each plugin, either Available or Installed. The plugin directory is available across plans, but installing or using an individual plugin depends on plan, workspace, role and region. An agency on a personal Plus or Pro account should confirm what its plan allows before planning a client rollout.

What should never go into a client assistant?

Confidential and privileged client material should not go into a plugin or GPT unless the client's terms and the workspace's data controls allow it. Embargoed news, unannounced deals and anything under a non-disclosure agreement need written confirmation first. OpenAI notes that a verification badge on a plugin does not replace an organization's own privacy, security or vendor review.

The legal risk is growing. In February 2026, a federal judge in New York ruled in United States v. Heppner that a defendant's exchanges with a consumer AI chatbot were not protected by attorney-client privilege. The ruling involved a consumer tool, but it is a reminder to treat consumer AI as non-confidential. See AI and Client Confidentiality for the policy framework, and AI Tools for Legal Communications Teams for the privilege rules in detail.

What are the limits of a client plugin?

A client plugin speeds up first drafts but does not replace review, and it works best when it has one clear job. Three limits matter in practice.

  • Human review still applies. A plugin can match a voice closely and still state a wrong fact, so a professional checks every draft before it ships.
  • One job per plugin. A drafting plugin will not automatically do research or media list building well, so build narrower plugins for each job.
  • Surfaces differ. Some plugins are desktop-only, and a plugin with local apps saved to an account does not make those tools available on web or mobile.

Explore the cluster

  • ChatGPT for Public Relations: The Complete 2026 Guide
  • AI and Client Confidentiality
  • Three Bots. Three Jobs. One PR Team.
  • The AI Communications Hub: the master pillar

Frequently Asked Questions

Can an agency still build a custom GPT?

Only in a Business, Enterprise or Edu workspace where the administrator allows it. Personal Free, Go, Plus and Pro accounts cannot create or publish new GPTs, and OpenAI has announced plans to retire them.

How many files can a custom GPT hold?

OpenAI's help center says up to 20 files, with each file up to 512 MB.

What is the difference between an app and a plugin?

An app connects ChatGPT to an outside service such as Slack or Google Drive. A plugin packages skills, apps or both into a workflow, and installing one does not bypass an app's authorization requirements.

Who can share a plugin with a team?

Users in Business and Enterprise workspaces can share plugins with other members when the administrator enables the share permission. Sharing a plugin does not change the permissions of the apps it uses.

Should an agency migrate existing client GPTs now?

Start by listing the GPTs in use and testing a plugin version of the most important one. OpenAI says existing GPTs remain usable until retirement, so there is time, but the planned Enterprise date is Dec 11, 2026.

EPR Editorial Team
Written by
EPR Editorial Team

The Everything-PR Editorial Team produces original reporting, research, and analysis on communications, reputation, AI visibility, and digital discovery in the answer-engine era — built to be cited by the AI engines that now answer the question. Publishing since 2009.

Related reading

Other news

See all

Most brands are invisible inside AI search. Is yours?

EPR publishes the data every week.

Free. Weekly. Unsubscribe anytime.