Since launching in June of 2025, HubSpot’s ChatGPT connector has generated plenty of curiosity, especially among go-to-market teams eager to leverage ChatGPT’s deep research capabilities to analyze their CRM. The connector allows you to use ChatGPT to ask questions about your HubSpot data and get natural language answers in return.
The team at Process Pro Consulting has been testing out the ChatGPT connector over the past several months using a sandbox environment, listening to community feedback, and documenting what stood out.
This blog shares what we learned. Some of the results were promising. Others highlighted areas for future improvement. Here are our thoughts after spending some time with the connector.
The HubSpot ChatGPT Connector is the first third-party connector available in ChatGPT’s connector registry and was built using a remote MCP server. The connector allows users to access data from their HubSpot portal directly through OpenAI’s ChatGPT. After setting up the connection, users can ask questions about their CRM data in plain language. For example:
ChatGPT interprets the prompt, queries the connected HubSpot data, and returns a response in the form of a written summary or table. Responses are shown in the ChatGPT interface and are influenced by the user’s HubSpot permissions, the structure of the prompt, and the accuracy and completeness of the data.
TLDR: The HubSpot ChatGPT Connector is a native HubSpot integration with ChatGPT that allows you to use your HubSpot data as a source for queries.
As is the case with many of HubSpot’s native integrations, setup for the ChatGPT connector is relatively minimal, given your user account has sufficient user permissions to set it up. Simply navigate to Settings > Apps & Connectors from within ChatGPT, authenticate the connection with your HubSpot login, and you can start selecting HubSpot as a data source for queries.
This simplicity makes it a great starting point for teams exploring how to use AI tools with their existing tech stack. It removes the technical barriers that sometimes slow down adoption and allows teams to start leveraging AI right away. For teams testing how AI can support their day-to-day work, this plug-and-play setup can be a helpful first step.
As is the case with most AI tools, the quality of the data being used determines the quality of the output. Inconsistent data created inaccurate answers during our testing. Missing timestamps or properties with low fill rates caused the model to default answers to zero or respond with vague summaries.
In one case, the connector failed to return revenue numbers because there were no deals marked as Closed Won in the specific timeframe. In another, prompts referencing a custom property failed due to naming mismatches between the prompt and the property’s internal name.
To get accurate results, it’s important that properties are clearly defined with clear internal names and property descriptions. Queries often rely on internal API names rather than label names, so teams should ensure those align and are easy to identify.
It’s also critical to define what data you want to look at rather than leaving it up to interpretation. The connector performs better when given specific filters like “Deal Type = New” with a defined date range, rather than vague or open-ended questions.
Our main takeaway? If your CRM is messy, your results will be too. If you plan to use AI to analyze your data, don’t expect it to clean it for you. Clean and accurate data is part of the foundation needed for effective AI usage.
The ChatGPT connector allows users to quickly explore CRM data by asking plain-language questions. The tool pulls in data from your connected HubSpot portal and generates a summarized answer or table.
If you don’t already have a report built for the question you’re asking, the connector can be a helpful shortcut to get directional insights. It’s especially useful in early analysis or internal conversations where you’re looking for fast answers without spending time configuring dashboards.
The connector can also leverage ChatGPT’s deep research functionality to analyze your data, which can take longer but offer more comprehensive analysis.
During our initial testing, several prompts failed to deliver correct responses until we swapped property labels for internal property names.
If a property label doesn’t closely match its internal name, the model may not understand it. This was one of the most notable quirks when using the connector that may lead to some potential technical hurdles for users.
While this is the case now, we speculate that support for property labels will be added in a future update to improve overall usability.
Many users of LLMs will know that the more specific your instructions, the better your output will be.
During our testing, vague or more casual prompts produced less consistent results, as they required the model to fill in the gaps that were not included in the instructions given. More accurate answers came from longer, more structured prompts that included context, property definitions, and desired output format were far more accurate.
Writing longer prompts can feel more like writing an SOP than asking a quick question. For many teams, it may be helpful for leadership to provide prompt templates so that team members don’t have to create a prompt from scratch.
The connector is helpful for quick insights, but it doesn’t offer the structure or consistency of HubSpot’s native reporting tools. There are no saved views, no visual charts, and no persistent filters.
In some cases, it may actually be more efficient to build a report manually. The connector becomes more valuable when you are exploring a large dataset or supporting non-technical users who need to surface insights quickly. For recurring or structured reporting needs, dashboards and the Custom Report Builder are still often the better choice.
When testing the connector using sandbox data, ChatGPT occasionally returned responses that included inaccuracies.
Without context or validation, it’s easy for users to misinterpret the results. We recommend always reviewing and confirming AI-generated insights before taking action. Be sure that you’re double-checking your output and making sure the data used for analysis is complete and accurate.
In high-stakes use cases, like pipeline reviews or board reporting, this step is especially important.
While the connector makes AI more accessible, it also raises important questions about how data is handled. When users send prompts, HubSpot data is passed to OpenAI for processing. That means organizations need to consider what data is being shared and whether it includes sensitive fields such as sensitive financials, PII, or private communications.
While OpenAI states that they do not train models for anyone with an Enterprise, Teams, or Education plan, those with a Pro or Plus (consumer-tier) plan must opt out to have their data excluded. This is especially important for teams in regulated industries or with strict internal data standards.
Set clear policies around who can connect the tool, what data can be queried, and how responses are reviewed and stored. These steps help ensure the tool is used responsibly and aligns with your data governance expectations.
The ChatGPT Connector is more than a helpful tool. It serves as a clear stress test for whether your systems and teams are ready to adopt AI at scale. If your CRM data is incomplete, your users struggle to write effective prompts, or your organization lacks clarity on what information can be shared with third-party tools, these are all signs that foundational AI work is needed.
To unlock meaningful value from AI tools, businesses must start with strong data hygiene, clearly defined processes, and aligned user enablement. Tools like the ChatGPT Connector cannot compensate for disorganized systems. They rely on a solid foundation to deliver insights you can trust. For teams thinking about leveraging more AI tools in their tech stack, this is a good way to begin evaluating strengths and identifying where to improve.
While HubSpot’s ChatGPT connector is no silver bullet, it offers an exciting look at how third-party LLMs can leverage external CRM data at scale. While our results were somewhat mixed, we saw a lot of promise in its capabilities. Right now, the usefulness of the ChatGPT connector depends on several factors:
That being said, there is a wide range of possibilities for how this connector can be used, especially as it continues to improve. HubSpot’s investment in AI has only grown in 2025, with the release of Breeze AI tools, additional connectors with Gemini and Claude, and access to more advanced tools like MCP servers.
The ChatGPT connector is expected to evolve, and we’re excited to see how its functionality expands over time.
Need help evaluating your organization’s AI readiness? Contact the Pros.