Connecting ChatGPT and Claude to HubSpot, Safely

Which HubSpot AI integration should you use?

Use the ChatGPT or Claude connector when people want to ask questions of the CRM and make small updates from a chat. Use HubSpot's remote MCP server when you're connecting another AI tool or building your own. Use AI inside HubSpot workflows when every record should be processed the same way, without anyone opening a chat.

What do you need to connect ChatGPT or Claude?

Any HubSpot plan, and a Super Admin or a user with App Marketplace access to connect without approval. Everyone else needs a Super Admin to approve the connector first and choose its data permissions. Claude needs a paid plan, and on Claude Team or Enterprise an owner enables connectors before members can authenticate individually.

• One account at a time: a ChatGPT or Claude account connects to one HubSpot account. People who work across several portals switch between them.
• Feature tiers: the connectors work on every HubSpot plan, but some data needs a specific hub, such as campaigns on Marketing Hub Professional or quotes on Revenue Hub.
• Claude Enterprise: admins can stop users connecting HubSpot to Claude accounts outside their organization.
• ChatGPT setup: HubSpot's article says Settings, Apps; OpenAI now calls this Plugins. Look for HubSpot there.

What can the connectors read and change in HubSpot?

They read CRM records, activities and, depending on your hubs, marketing and content data. They can create and update records such as contacts and deals and log calls, emails, meetings, notes and tasks. They can't delete anything, they change at most 10 records at a time, and they draft workflows rather than build them.

• Validation rules: HubSpot's ChatGPT article says custom validation rules, including pipeline stage rules, aren't applied to its changes. Its Claude article says those rules are applied, except association label validation. Check before letting either update deal stages.
• Sensitive Data: if your account uses HubSpot's Sensitive Data setting, the connectors don't get engagement data.
• Content: only published pages and posts are available in full; drafts return metadata.
• Audit log: HubSpot records connector writes against both the user and the connector.

Answers are only as good as the CRM behind them. If deal stages, lifecycle stages and owners aren't kept clean, an assistant will summarize the mess confidently. Our HubSpot data quality guide covers the basics.

How is your HubSpot data handled?

It depends on the AI plan. OpenAI says data from apps may be used for training on ChatGPT Free, Plus, Go and Pro when the Improve the model for everyone setting is on, but not on business plans. Anthropic says it doesn't train on commercial plans by default. HubSpot routes requests from EU-hosted accounts through its EU data center.

• ChatGPT: on Free, Plus, Go or Pro, turn off "Improve the model for everyone" for anyone connected to HubSpot, or use a business plan.
• Claude: on consumer plans, Anthropic says training data excludes raw content from connectors, though content copied into a conversation may be included.
• Once data leaves HubSpot, the AI provider's terms apply. HubSpot also says it may collect anonymized summaries of tool calls on its MCP server.

How do you use ChatGPT or Claude inside HubSpot workflows?

HubSpot's Use a custom LLM workflow action calls OpenAI, Anthropic and other providers with your own API key, on an Enterprise subscription. HubSpot's own Data Agent custom prompt action works on Professional and Enterprise and uses HubSpot credits. Data Hub Professional and Enterprise can also call any AI API from custom code or a webhook.

• Data Agent prompts: limited to 3,000 characters, and the model isn't connected to the internet.
• Custom code actions: must finish within 20 seconds and up to 128 MB of memory, so keep AI calls short or hand long jobs to another tool such as n8n.
• Write results to a property, such as a summary or a category, and review a sample before using it in routing or reports.

How should you roll it out, in order?

Decide what people should use AI for, then have a Super Admin approve the connector with the narrowest data permissions that support those uses, and choose who can install it. Confirm the AI plan's training settings, test updates on sample records, and review the audit log after the first weeks before widening access.

1. List the use cases: meeting prep, account summaries, logging notes, quick property updates. Note who needs each.
2. Approve the connector as a Super Admin, granting only the data permissions those uses need, and choose who can install it.
3. Check the AI side: plan, training settings, and whether a ChatGPT workspace admin or Claude owner needs to enable it.
4. Test updates on sample records, especially deal stages, given the validation differences above.
5. Pilot with a small group and a written rule for what AI may change.
6. Review the audit log, then widen access.

HubSpot AI connector checklist