Should You Connect AI to HubSpot? The Answer Just Got a Lot More Interesting.

Updated May 2026 — originally published June 2025

 When HubSpot announced its Deep Research Connector for ChatGPT back in June 2025, I wrote about it with equal parts excitement and caution. As someone certified in AI for Business and AI Data Governance, I couldn’t just geek out over the possibilities without also flagging the risks. And there were real ones.

But a lot has changed since then. HubSpot has been moving fast — and honestly, in the right direction. The concerns I raised about data exposure and lack of admin visibility? HubSpot heard that feedback (from a lot of people), and they built real controls around it. So let’s revisit this, because the conversation has evolved.

What the Integration Does

With this connector, you can ask ChatGPT questions about your CRM data in plain English. No code, no complicated filters—just natural language and immediate insights.

Here are a few examples of what it can do:

• “Show me the companies with open deals over $50K that haven’t been contacted in 10 days.”

• “Summarize our pipeline for Q3 by sales rep.”

• “What are our top support ticket categories this month?”

It’s powerful, accessible, and flexible—especially for teams that don’t have time (or patience) for dashboards and reports.

What’s Actually Available Now

The connector ecosystem has expanded significantly. HubSpot now supports direct integrations with:

• ChatGPT (the original Deep Research Connector)
• Claude (Anthropic)
• Gemini (Google)

Each of these connects to your HubSpot CRM and allows users to ask questions, pull data, and take action — in plain English, without building a report or knowing a filter exists. It’s genuinely powerful. But the key update is who’s in charge of it now.

The Benefits

1. Time-Saving: No need to pull reports or manually filter data—just ask and get what you need.

2. Better Insights, Faster: Team members can spot gaps or trends instantly, without relying on a data analyst.

3. Cross-Team Use Cases: Marketing can segment by persona or region, Sales can analyze pipeline risk, and Support can forecast ticket volume.

4. Built-In Permissions: ChatGPT respects your HubSpot user permissions—so users can only query data they already have access to.

The Risks (And They’re Real)

Despite the benefits, this isn’t plug-and-play AI magic. There are very real concerns—especially if your business handles sensitive customer data or operates in regulated industries.

Here’s where things get tricky:

1. Data Exposure

Prompts can easily include personally identifiable information (PII), confidential notes, or internal strategy. If a user types “Summarize all contacts with churn risk and include notes,” you’re potentially sending sensitive info into an external AI system.

2. Audit Trails

When I originally published this, HubSpot had no audit trail for AI activity — that’s now been addressed (more on that below), but it’s worth knowing it wasn’t always the case.

3. Compliance Concerns

If you’re subject to GDPR, HIPAA, or other regulatory frameworks, this could raise red flags—especially if the integration isn’t backed by clear internal policy or legal review.  Having worked in the healthcare industry – this is HUGE and cannot be understated.

4. User Behavior is the Wild Card

The tool isn’t dangerous—but untrained users can be. The AI can only be as safe as the person crafting the prompt. If you have curious wanderers on your team, the allure of using ChatGPT in HubSpot could be a problem.

The Controls Super Admins Have Been Waiting For

Here’s what’s changed, and it’s significant.

HubSpot has introduced centralized App Management controls that give Super Admins real authority over the entire AI connector ecosystem. You can now:

-Approve which apps can be installed at all — users can’t just go rogue and connect ChatGPT on their own
-Control which users can install each app — role-based, not blanket access
-Set data permissions per connector — read-only vs. full edit access, decided by you
-Manage AI access by data type through Settings > AI > Data Governance, including object-level scopes (you can allow Breeze to read Contact data for summaries while blocking it from Deal or Revenue data entirely)
-Review an AI Audit Log that shows which users are prompting AI tools and what data is being surfaced

That last one is huge. One of my original concerns was that HubSpot had no visibility into what your team was actually asking. That gap is now addressed.

Additionally, Super Admins can control AI model training — whether HubSpot may use your account’s customer data to train its own AI models. Toggle changes are logged. You’re in control, and you can prove it.

If your account has Sensitive Data turned on, you’re automatically opted out of AI model training and certain connectors like Claude won’t access engagement data at all — adding an extra layer of protection for regulated industries.

But Here’s What Breeze Can’t Do — And This Is Where It Gets Interesting

HubSpot’s native Breeze AI is excellent for what it’s designed for: summarizing records, drafting emails, scoring leads with its built-in model, generating content. It lives inside HubSpot and works within HubSpot’s logic.

What it can’t do is think sideways.

That’s where connecting an external AI like Claude becomes genuinely valuable — not as a replacement for Breeze, but as a layer of deeper reasoning on top of your data. Here’s how I’ve been using it with clients:

Lead Scoring That Actually Reflects Your Best Customers

HubSpot’s native lead scoring is rules-based. You define the criteria, assign points, and hope you guessed right. The problem? Most teams pick criteria based on intuition, not pattern recognition.

What I’ve done instead: pull a structured export of your closed-won customers, bring it into Claude, and ask it to identify the common threads — industry, company size, deal cycle length, first-touch source, number of contacts engaged before close, whatever data you have. Claude will find the patterns you didn’t know to look for. Then you build your HubSpot scoring model around what your actual customers look like. The result is a lead score grounded in evidence, not guesswork.

Advanced Reporting That HubSpot Can’t Build on Its Own

HubSpot’s reporting is powerful, but it has a ceiling — especially when you’re trying to connect behaviors across objects or analyze qualitative data alongside quantitative data.

Here’s a real example: a client wanted to know whether leads who asked for a follow-up during a sales call actually got one. On the surface, that sounds like a simple pipeline report. But the follow-up request lived in a call transcript (qualitative), and the confirmation lived in an activity log (quantitative). HubSpot can’t natively cross-reference those two things to tell you whether your team closed the loop.

The approach that works: export the relevant call transcripts and activity logs, bring them into Claude, and ask it to analyze whether transcripts that include follow-up requests have a corresponding follow-up activity logged within a set timeframe. Claude can work through that in ways HubSpot’s report builder simply isn’t designed to handle.

This isn’t a workaround — it’s using the right tool for the right job. HubSpot stores and manages your data beautifully. Claude helps you understand it at a level of nuance that structured reporting can’t reach.

My Recommendation

1. Set your permissions before you open access. Go to Settings > AI > Data Governance first. Decide which data types AI can touch, and which it can’t. Don’t connect and configure later — configure first.

2. Decide who gets access based on role, not trust. This isn’t about whether you trust your team. It’s about whether a junior coordinator needs to query deal revenue data through ChatGPT. Probably not. Scope it accordingly.

3. Use the Audit Log. Make it a habit to review it. Not because you’re policing your team, but because you’ll catch unintended data patterns before they become a problem.

4. Train your team on prompt hygiene. AI is only as discreet as the question you ask. A prompt that includes a customer’s full name, email, and account notes is sending all of that into an external system. Teach your team to keep prompts general and anonymized where possible.

5. If you’re in a regulated industry — GDPR, HIPAA, CCPA — loop in legal before you connect anything. The controls are solid, but compliance is a legal question, not just a technical one.

---

Final Thoughts

A year ago, my answer to “should you connect AI to HubSpot?” was a cautious “maybe, with significant guardrails you have to build yourself.”

Today, my answer is different: yes — with intention, with the right permissions in place, and with a clear picture of what you’re actually trying to accomplish.

HubSpot has built the guardrails. Breeze handles the routine. External AI like Claude handles the complex reasoning, the pattern recognition, and the analysis that structured reporting can’t reach.

That’s not a gap in the system. That’s a feature.

If you’re not sure where to start — whether that’s setting up your AI governance settings, building a smarter lead scoring model, or figuring out what reports you’re missing — that’s exactly what I help clients work through.

UPDATE

This blog was posted after midnight on June 6.  When answering client questions at 6am, I noticed the knowledge base had been updated to include the language 

• HubSpot customer data is not used for AI training in ChatGPT. To ensure that your data is not used for any AI training in ChatGPT, it’s recommended to turn off the Improve the model for everyone setting.

There is further information stating that users are only going to be able to access information that coincide with their access within HubSpot – but we know that ChatGPT will be able to see everything in HubSpot so how will it not veer into other areas to answer a question?

Further, ChatGPT isn’t supposed to have access to sensitive information.  But again, once it’s connected everything is there so how does it not use all of the information as a resource for answering questions.