ChatGPT Connector for Salesforce: Options and Permissions to Check

A ChatGPT Salesforce connection lets ChatGPT Work read CRM data — accounts, contacts, opportunities, cases — and help with summaries, pipeline analysis, and drafting follow-up content. Getting it right means choosing the right connection option for your plan, reviewing which objects the AI can access, and deciding clearly which actions need a person to approve before they reach Salesforce. Tools like fenbs keep cross-team follow-up decisions and handoff notes in one place so work continues cleanly between sessions.

Check your ChatGPT Work connector settings first. Available connections depend on your plan and on what Salesforce editions support. If a native connector appears, the steps below apply. If not, this guide covers alternatives.

What ChatGPT can do with Salesforce access

With read access, ChatGPT can help you summarize an account's recent activity, identify open opportunities by stage, draft an email to a contact, or answer questions about a case. It can surface information that would otherwise take several clicks to assemble.

Write access is more sensitive. Creating or updating a Salesforce record has downstream effects on pipelines, reports, and the work of other people. Treat any AI-proposed change as a draft to review rather than a completed action, and establish a review step before changes are committed.

Connection options for Salesforce

ChatGPT Work may offer a native Salesforce connector in its connector settings. If no native connector is available for your plan, alternatives include exporting Salesforce reports as CSV files for upload to a ChatGPT session, using the Salesforce API with a custom connector configuration, or routing through a third-party automation platform.

Salesforce also has its own AI features built in. For work that stays entirely within Salesforce, those features may be a more natural fit. A ChatGPT connector is most useful when you need to bring Salesforce data into a broader workflow that also includes other tools.

If your Salesforce instance uses a managed package — a third-party add-on installed from the AppExchange — check whether that package adds its own permission restrictions on top of Salesforce's native ones. Some managed packages block API access to their objects by default. Your Salesforce administrator can check whether any connected app permission set covers the objects and fields your workflow needs.

Permissions to check before connecting

Salesforce permissions are granular: read and write access can be set per object type (accounts, contacts, leads, opportunities, cases, and so on). When setting up a connection, request only the object types and permission levels the workflow actually needs.

• Confirm whether the connection uses read-only or read/write permissions.
• Review which Salesforce objects are accessible — not all object types may be needed.
• Check whether the connected profile has access to sensitive fields such as contract values or personal contact data.
• Confirm the connection uses a dedicated service account or OAuth user, not a shared admin credential.

For the current Salesforce connected app permission model, see Salesforce's developer documentation↗ (opens in a new tab). Your Salesforce administrator will need to approve the connected app before it can be used.

Reading records versus changing them

Reading a contact record to prepare for a call is low risk. Updating an opportunity stage or creating a case affects other people's pipelines and reports. Before giving write access, define which actions the AI may propose and which must go through a person for approval.

If your workflow involves cross-team handoffs — a sales person drafts a follow-up, a support person creates a case from it — record those handoff steps as tasks. The next person, or the next AI session, picks up from the recorded state rather than reconstructing it from memory.

Salesforce data has a natural lifecycle: a lead becomes an opportunity, an opportunity becomes a closed deal, a deal generates a case when something goes wrong. AI sessions that read Salesforce data at one stage of that lifecycle need to understand where a record sits to give useful output. Recording the lifecycle stages and what each means for your team helps the AI avoid treating a closed-won deal as an open opportunity. That context belongs in the project record, not re-explained every session.

How to do this with fenbs

1. Create a free account and add a project for your sales or CRM work.
2. Record approval rules as Decisions and rules: which Salesforce actions need a person to confirm before they are applied.
3. Add follow-up tasks as items on the board, so they are visible to the next person or AI session that picks up the work.
4. Connect your AI via Integrations and ask it to read the project context, including rules, before working with Salesforce data.
5. Before ending a session, ask your AI to save progress: what it completed, what it checked, and what is waiting for a person.

For connector problems, ChatGPT Connectors Not Working covers common failure points. For a broader overview of what connectors support, see ChatGPT Connectors and Apps. Other guides are in the connect-ai topic.