ChatGPT Plugins

ChatGPT plugins let your team bring work from its existing tools into a conversation, then use supported actions to move it forward. For a Business or Enterprise admin, the useful question is specific: can an employee investigate a request, review the evidence and complete the permitted next step without copying everything between systems? Start with one workflow and check the plugin, its connected apps and its action permissions separately. OpenAI’s plugin overview

What ChatGPT Plugins Are Now

A plugin packages what ChatGPT needs for a repeatable job. A skill supplies reusable instructions. An app connects to an outside service and its supported tools. A plugin can contain either or both, sometimes alongside an app template that needs configuration. Think of it as a work kit: the checklist explains the job, while the connected app supplies the authorized access. Plugins in ChatGPT

This guide covers today’s plugin directory. OpenAI marks its March 2023 plugin launch post as outdated and points readers to the current catalog. Old beta tutorials are not the setup instructions for this system. Original plugin announcement

The connector naming change appears in OpenAI’s Business release notes. For existing connections, use the ChatGPT connectors admin guide.

Which Plans Can Use Plugins?

Select plugins are available on Free, Plus, Pro, Business, Enterprise, Edu and ChatGPT for Teachers; some require a paid plan. The directory being available does not make every listing available to every account. Check the plugin’s plan, region, workspace and device requirements. Plan availability, plugin requirements

For the two workplace plans, the defaults differ:

Treat defaults as a starting point and inspect your saved configuration. Feature-page defaults, detailed workspace defaults

How to Turn Plugins On for Your Team

Open Workspace Settings → Plugins, or Admin console → select your ChatGPT workspace → Plugins. Admin navigation

1. Open the plugin. Review its skills, required apps, optional apps and templates.
2. Choose its Installation policy. Available lets eligible members install it; Installed installs it automatically for eligible members or supported roles.
3. Enable the required apps. Review their access, permitted actions and provider restrictions.
4. Test with an eligible member. Complete individual provider authorization only when that app requires it. Official configuration steps

For Enterprise role settings, open Workspace settings → Permissions & roles, then the relevant role. Check Use plugins and the separate permissions for uploading, creating with MCPs, sharing and publishing. MCP is a standard connection method that lets ChatGPT call tools supplied by a service. Grant the permissions needed for the member’s job. Member permissions

An app template needs another setup step: an admin configures and publishes a workspace app before members use it. Installing its plugin does not complete that work. App template setup

For the employee, the path is Plugins in the sidebar → choose a plugin → Install plugin, where shown. If prompted, select Connect, choose the intended provider account, review access and finish authorization. Some apps use an admin-managed connection or need no individual sign-in. Then ask for the outcome, using @ and the plugin name when you want to select it explicitly. Account connection steps

What Data Can a Plugin See?

Check both directions: what ChatGPT retrieves, and what the connected service receives. Confirm whether the app uses an employee account or an admin-managed connection, and review its permissions. Connection setup

OpenAI describes these data flows:

• Into the app: relevant conversation context and, with Memory enabled, relevant remembered information may be shared to fulfill a request.
• Basic connection information: apps may receive IP address, device or browser type, language, region and approximate location.
• Back into ChatGPT: retrieved information can inform answers and web-search queries. Memory may retain relevant information where enabled and permitted by the source.
• Provider handling: information shared with an app is governed by that provider’s terms and privacy policy.

For Business, Enterprise and Edu, OpenAI does not train its models on information accessed from apps by default. That training policy does not mean data never reaches a connected provider. OpenAI’s connected-app data explanation

What the Admin Controls

Think in separate decisions:

These controls answer different questions; enabling one does not settle the others. OpenAI’s control map

Under Actions, admins can select supported read and write operations. Under New actions, choose how later additions are enabled; Disable new actions does not turn off existing ones. Domain and content restrictions are available for supported apps and workspaces. Some apps lack configurable action controls. Admin action controls

Allow read actions means reads can proceed without asking, while changes require approval. It does not disable writes. Always ask asks before reads and changes. Other permission options depend on the app and account; the standard workspace-wide selector does not offer Allow all actions. Approval settings

A workspace-enabled skills-only plugin can remain usable by eligible members whose role shows Disabled. Automatic distribution does not prove exclusive access. Skills-only role behavior

Five Workflows Worth Trying, Ranked

Start where staff repeatedly gather evidence before making a decision. These proposed pilots build on three examples from OpenAI: ServiceNow incident work, Salesforce deal reviews and Slack project updates. The payoffs below are reasons to test them, not measured customer results.

1. ServiceNow Incident Investigation

An IT service owner could use a plugin to prepare an incident assessment before deciding what to change. OpenAI describes finding similar incidents, examining change risk and updating a ServiceNow record. Customers connect their own instance and choose capabilities; ServiceNow’s Action Fabric applies existing rules and permissions. ServiceNow announcement

The live announcement is dated October 28, 2026. It says the MCP plugin is available, enhanced capabilities are expected by year-end, and interactive incident experiences are being previewed. Check the actions exposed in your installation before assigning a production workflow. Availability and preview status

A sensible first pilot would produce an evidence-backed assessment for a human reviewer. The potential payoff is less time rebuilding the incident’s history before useful work begins.

2. Salesforce Deal Reviews

A sales-operations lead could turn a quarter-end deal review into a ranked list of risks and next steps. OpenAI’s example uses Salesforce to review closing deals and prepare an action plan. OpenAI’s business examples

Ask the pilot to distinguish recorded facts, missing information and proposed follow-up. That separation could reduce meeting preparation and make incomplete CRM records easier to challenge. A plausible summary of a weak record is still a weak basis for a forecast.

3. Slack Project Updates

A delivery lead could ask for progress, blockers, owners and next steps from the team’s conversations, an example OpenAI presents for Slack. Slack workflow example

Use a reporting skill to keep the same format each week. Require links to supporting messages and leave disputed ownership unresolved for the lead to decide. The useful saving would be in collecting and formatting the update, while the lead remains responsible for its meaning.

4. Incident Handover Preparation

Building on the ServiceNow workflow, an on-call lead could request a handover draft containing the current issue, relevant history, unresolved questions and proposed next check, using only information the installed connection supports.

Test this after the basic assessment works. Ask the outgoing engineer to mark omissions before anyone relies on the draft. A shorter handover is valuable only when the incoming engineer still knows what is uncertain.

5. A Weekly Decision and Blocker Brief

Building on the Slack example, a department operator could narrow the output to decisions awaiting an owner and blockers that need escalation. This is a more focused task than summarizing everything that happened.

A team could compare the brief with its existing meeting agenda and count useful additions, false alarms and missed blockers. That tells you whether the plugin improves coordination or merely produces another document to read.

The business consequence is fewer manual handoffs. A conventional reporting loop might involve finding records, copying excerpts, reconciling them, writing a summary and returning to the source tool. A plugin pilot could bring retrieval and drafting into one conversation, with supported follow-up actions subject to policy. Measure total staff hours across the full loop, including review and corrections, before claiming a saving.

Two Things Worth Building Around This

Strongest: A Governed IT Triage Workflow

Build a narrow incident-assessment package for a service-desk manager. Combine the team’s assessment instructions, approved connection and review criteria. Sell the configured workflow and its validation to organizations already struggling with incident preparation.

DataForSEO estimates 480 monthly US searches for “ai help desk”, checked October 30, 2026. That is broad interest in the job, not a count of buyers for a ChatGPT plugin. It supports testing the problem with service-desk teams; it does not establish demand for this particular implementation.

The smallest useful version would handle one incident category, produce a reviewable assessment and document what it cannot access. Start without changes to production records; add a specific write operation only after the installed integration supports it and the organization approves the workflow.

The catch is integration and maintenance. ServiceNow configuration varies, some announced experiences remain previews, and the provider may absorb a generic triage feature. The defensible work is knowing the customer’s escalation rules, evaluating failures and keeping the configuration useful. That makes this the stronger opportunity than another general-purpose chat wrapper.

Smaller: A Department’s Weekly Reporting Plugin

Build a reusable update workflow for a delivery or operations team. The buyer pays for a consistent report based on its own definition of progress, risk and ownership.

DataForSEO estimates only 10 monthly US searches for “automated status reports” in the same check. That narrow query is a weak standalone demand signal. Treat this as an implementation add-on for an existing customer, then validate interest through actual usage.

The smallest version could combine a reporting skill, an approved Slack connection and a set of example reports. OpenAI supports building plugins for internal workflows. Building for your organization

The catch is that a template is easy to copy. The customer-specific work is defining a useful report, handling missing evidence and deciding who maintains it. Stop expanding the package if the team does not use the first report.

What This Does Not Solve

A plugin does not repair inconsistent source records or decide who should own a process. My recommendation is to avoid a broad rollout until one workflow produces dependable evidence with tolerable review effort. Keep the existing process available during the pilot.

Do not assume a custom GPT becomes an equivalent plugin automatically. OpenAI schedules standard GPT retirement for December 11, 2026, with February 11, 2027 for FedRAMP and qualifying approved Enterprise deferrals. New GPT creation ended October 26 for affected Enterprise workspaces, except FedRAMP; follow your workspace notice. Migration timeline

Migration moves published instructions into a skill and copies knowledge files into reference files. Custom actions do not transfer, the selected model does not carry over, and the replacement begins private. Test the integration and restore the intended sharing before moving colleagues across. Migration behavior

Use the custom GPTs retirement guide to plan that separate piece of work.

How do I use ChatGPT plugins?

Open Plugins, choose an available plugin and finish any required app setup. Start a new chat and describe the task; use an @ mention to select the plugin explicitly. Check device restrictions, including Desktop only listings. Using plugins

Are ChatGPT plugins free?

Select plugins work on all plans, but some require a paid plan. Check the individual listing and its account requirements. Normal ChatGPT plan limits apply, and external plugins may impose their own caps. Availability and limits

Do I need a Chrome extension to use ChatGPT plugins?

The official starting point is the plugin directory inside ChatGPT. A third-party Chrome extension is not a substitute for installing an approved directory plugin. Follow the selected plugin’s setup instructions. Official plugin directory guidance

Monday move: pick the person who owns incident triage or weekly reporting. Choose one plugin, one recurring task and one reviewer. Have the admin confirm the connection and actions, then compare a real output with the existing process. Expand only when the team can explain what improved and what still needs checking.

If you want that workflow configured for your team, we build AI automations for business operations.