ChatGPT for Sales
In current ChatGPT terminology, a plugin is an installable workflow package that can include skills, apps and app templates. A skill is a reusable method with instructions and supporting resources. An app connects ChatGPT to an external service so it can retrieve information or perform supported actions within the user’s and workspace’s permissions.
For a seller: a skill defines how to do the work; an app provides where the data or action lives; a plugin packages one or both into a discoverable workflow.
Product details verified 30 August 2026. OpenAI migrated the app directory to the Plugin Directory on 9 July 2026. Installation and use vary by plan, workspace, role, surface, region, app capability and administrator settings. See Apps in ChatGPT, Plugins in ChatGPT and Codex and Skills in ChatGPT.
| Layer | Owns | Sales example | Does not automatically provide |
|---|---|---|---|
| Plugin | Discoverable, installable workflow package | Account-brief workflow with a research skill and Drive app | Permission to every included service |
| Skill | Reusable instructions, method, resources or code | MEDDPICC evidence-extraction procedure | Live CRM or email access by itself |
| App | Authenticated connection to an external service | Retrieve approved Google Drive files or Slack context | A good sales method |
| App template | Guided setup pattern for a provider or internal service | Admin-published company app configuration | Authorization outside existing user/workspace access |
| MCP | Underlying protocol/infrastructure for tools and data | Custom enterprise connection | A user-facing content taxonomy or sales workflow by itself |
The old 2023 “ChatGPT plugins” ecosystem is historical. Do not use third-party lists from that period as evidence for what can be installed now.
An eligible email app can help locate a thread, summarize the latest customer message or prepare a reply draft. A scheduled task can respond to a supported new-message event where documented.
Review boundary: verify recipients, quoted commitments, dates and attachments. Do not send automatically unless the connection, action and approval are explicit.
Google Drive is an official example of a current app. It can make approved files available to eligible ChatGPT surfaces, subject to provider permissions and workspace controls. Use it to retrieve the correct product brief, account plan or customer proof rather than uploading uncontrolled duplicates.
Review boundary: access follows the connected account’s permissions. Connecting the app does not make every file appropriate for use.
OpenAI currently names Google Calendar among tools that eligible managed-workspace agents may connect to. Use it for meeting context only where the workspace, role, app and surface support it.
Review boundary: do not claim universal calendar support. Limit preparation to relevant meetings and approved account context.
Slack is an official current app example and a supported event source in scheduled-task documentation. Use it to retrieve approved team context, post or prepare summaries where the action is supported, or operate an eligible workspace agent in connected channels.
Review boundary: channel access, message action and data sensitivity still follow organizational permissions and policies.
Do not claim a native Salesforce, HubSpot or other CRM connection merely because a query implies one. Verify the current Plugin Directory listing, supported action, plan, region and administrator state at the time of implementation.
Where a CRM app or approved custom connection exists, use the minimum required scope. Prefer read-only retrieval and proposed updates before enabling writes.
Some apps support search, syncing or Deep Research. Support is capability-specific: an app may work in Chat but not Deep Research, or support retrieval without write actions.
Review boundary: “connected” does not mean “available in every mode.”
A skill should encode a stable procedure, not a vague persona. A strong sales skill contains:
Strong candidates include account-research methodology, discovery-evidence extraction, call-quality review, stakeholder mapping, proposal compliance and pipeline-risk inspection.
Before connecting or enabling an app, answer:
If the team cannot answer these questions, do not connect the app yet.
Use this four-part test:
Test with a non-sensitive sandbox case. Confirm what the app read, what it proposed, what it changed and what approval appeared.
| Need | Better fit |
|---|---|
| Repeat the same procedure with different inputs | Skill |
| Reuse a conversational specialist | Custom GPT |
| Keep one deal’s evolving context | Project |
| Combine specialist behavior, tools, skills and schedules | Workspace Agent |
| Package a method and app connection for installation | Plugin |
Start with one read-only, high-value use case: retrieve the correct account file, summarize an approved email thread or gather a defined internal source for research. Add actions only after the retrieval, evidence and approval behavior is proven.
Browse Promptifi workflows or see ChatGPT setup and data controls.
Current terminology is the reverse of the old assumption: OpenAI now uses the Plugin Directory as the primary discovery layer. A plugin can package skills, apps and app templates; apps remain external-service connections.
A skill is a reusable set of instructions and supporting resources that defines how ChatGPT should perform a task. It does not provide external-service access by itself.
No. App access remains subject to the provider account, user authorization, workspace settings, role and supported ChatGPT surface.
Some eligible apps can provide sources to Deep Research. OpenAI currently says research uses available read actions and does not use app write actions.
Each capability owns one seller job. Start with the smallest one that produces the outcome.
