OpenAI Connector

Foreign. Welcome my friends to another session of the HubSpot admins hug. Thank you for making this your weekly event to connect with, learn from, lean on and just be with other admins in the HubSpot ecosystem. I'm your host, Deanna Giloff. And of course, big thank you to our community members helping out behind and in front of the scenes. If this is your first hug, welcome, welcome, welcome. Please let us know in the chat how long you've been using HubSpot and if there's anything that you are just stumped on and need a resource for. And to our regulars, we welcome you. I'm glad you decided to come back again. I can't wait to see all of the amazing awesomeness that's going to be shared here by Matt and Kathy. Let's get this have housekeeping started. We've got upcoming events. Any of the events that are happening for the last part of the year on here. There's only six left my friends, so get those on your calendar. I do have the weeks that we aren't meeting for the hug on here so that you can get them on your calendar. You can take back that time for yourself and make sure that you are securing your Tuesdays all the way into 2026. If you do miss an event, that's okay. They are recorded. You can catch them on the YouTube channel. Make sure you're subscribing so you're getting notifications as soon as I publish them. I try to get them up there as soon as possible. Sometimes my calendar makes it so that I don't get them up there right away. But if you get subscription for the notifications then you'll know as soon as they're on there for the membership resources. I added this last week, but I just want to call it out again. You have the ability to add your membership profile to your LinkedIn, to anywhere else, just to show off that you are proud HubSpot administration member. Get that on there. If you have any questions on when you actually did join the admin hug, shoot me a message on LinkedIn. I'm glad to look it up and you can let everybody know the number of years that you've been a HubSpot Admin Hug member. We also still have that LinkedIn group. You definitely make sure you're going in there sharing resources, talking to your other admin members. Great conversations happening in there. We do have a space on the HubSpot community. Make sure you're checking those out. There's 937 of you already in that group, we are also doing the show and Tell. This is happening next week. If you haven't submitted your submission for an upcoming show and Tell, go ahead and get those on there. I have a bunch of submissions that I have to keep going through, so make sure you're doing that. And as soon as we get the 2026 calendar ready, I will be reaching out to you. We're going to try to do those quarterly, so get the submissions in. And I'm going to try to get a lot ahead so that people can plan for showing up and sharing. They're awesome. The boot camps, those are full for fall, but it looks like they are opening on December 19th for winter. So go ahead and make sure you're checking back here. Those boot camps fill very, very, very fast. You want to make sure you are signing up as soon as they are open. If you've got any issues happening, Bevy is our wonderful platform here that we use. Turn it off, turn it back on again. Check out those troubleshooting ideas and hopefully it will be working for you. If you have an idea for an upcoming hug, let me know. I'm trying to get ahead of 2026, so I'd love to hear your suggestions. I checked. I check these very often and I'm always looking for ways and resources not just for HubSpot team members, but also for community members. For amazing content that you as admin want to hear about, make sure you're connecting with myself, with Kyle, with the rest of your admin members. We do have space on for hiring for HubSpot is. So if you've got a role that you're looking for, if you're looking for your next role, make sure you're checking that out. All right, my friends, I want to have Kathy and Matt join us. Thank you guys for coming. It's. It's going to be a great session. We've got a unique combo today. We've got a HubSpot team member and a member of our community coming together and sharing what's going on with AI. Welcome, Matt and Kathy, go ahead and introduce yourself. Great.

I will go first and then I hand it over to Matt. It's so great to be here and to get together to be able to talk about HubSpot's AI connectors. We have a lot of information for you today and I can't wait to get into it. I'm with Matt from hubjoy, but before I hand it over to him, I want to introduce myself. My name is Kathy. My pronouns are she, Her. I'm a lead consultant at HubSpot. I help businesses with their HubSpot portals use the tool better. I help them implement better strategies and get into the weeds with all things marketing, sales and operations. I've been using HubSpot software for about 15 years and I really can't imagine a better tool for any business to scale with. So that's a little bit about me. Matt, I'm going to hand it over to you.

Thank you, Kathy. It's been super cool to get to know you over the last couple years. We've prepped this webinar. Yeah. My name is Matt Steffen. I'm the founder of Hubjoy. We're a modern HubSpot partner agency. We're growing at about 30% month over month right now, and we're kind of exclusively focused on helping our clients unlock the power of what we say is like a modern go to market stack. And of course that's centered around HubSpot. We're all about automation, intelligence and of course, layering in AI wherever we can. And I'm very excited to be here today because we have been deep in the weeds on the OpenAI connector since day one, you know, working with the HubSpot product team, trying to test it, break things, build frameworks, and really like, figure out how we can make it truly useful inside of real client environments. So the session today is really about trying to turn some of those learnings into something that you can use. Hopefully there are some very practical and tangible ways to connect AI to market motion smarter. So that's pretty much the goal for the day here for, you know, you to set the stage and we can walk through, you know why.

Yeah.

So as we talk a little bit.

About.

Where we're at with the connectors, the AI deep research connector that we first launched at HubSpot was the OpenAI connector with HubSpot. And it really represents the next leap forward in HubSpot's AI forward roadmap. And it brings advanced natural language analytics directly to users workflows. It seamlessly integrates deep research capabilities and empowers every team to translate their CRM data into actionable insights at unprecedented speed. It makes AI innovation practical and it gives you an everyday asset for your growing business. We originally launched the AI connector with ChatGPT. Since then we've launched Claude and we're actually working on a Gemini Connector. So there's a lot that's sort of like influx right now and a lot going on.

Yeah. And it's been super, super cool. Like from a testing perspective, to be on the ground floor of that. Now a lot of this comes out of doing on the mcp, like model plane, model context protocol. You may have heard that term thrown around MCP servers. And one of the analogies that I like to use, it's almost like having a AI Somalia or like a data sommelier. You know, you give it a vague request like hey, you know, I want, I want something bold but not too heavy that goes with the steak. And then you know, an MCP is translating that into something precise. It's selecting the right wine and it's presenting it perfectly. So in HubSpot terms, it's taking your natural language prompt, it's pulling the right data, formatting that request, querying these AI, these large language models, and then handing back really clean answers without you ever really needing to touch the cr. You know, your request is like walking into a wine cellar and shouting your request with mcp. You know, you're getting these curated recommendations that are tailored to your data. And it's like you said, it's really bridging the gap between raw data and human language, which I think is just super, super excited. And obviously one of the things that we've been hearing, and I know you've been hearing it from your customers as well, is like, I'm super excited about this, but I don't really know where to start.

Absolutely, that's exactly what I hear a lot of.

So what we want to do today is actually just walk through the setup like the connector setup 101. So I'm going to just do it in a screen share here live and show you how to connect the Deep Research connector in the first place. Standpoint show a real life example can use these, some of the ops that we've developed and sort of craft specific use case and then have a bit of a conversation around like what are some of the outcomes and things that you can expect as you get into using the Deep Research connector. So I'm going to be using GPT to actually run this. Like we said, there's now Claude connector and Gemini coming online here as well. So to start things off we're actually going to hop into Deep Research mode. We're looking for adding HubSpot as a source here so I can come down into connect more. And here you can see all the connectors that OpenAI has available. There's HubSpot right there. As soon as I connect this gonna prompt me to connect my HubSpot portal and authorize. We're gonna Touch on privacy a little bit here but this is completely just read only access. So I'm gonna select, select the HubSpot portal that I want to connect. Once I've chosen that account, it's going to bring me back into chat. We are good to start prompting and doing search. One thing I want to call out real quick because this is, this is a question that I've been asked a couple times as well is can be used just in like a normal chat window as well. So once I get into a normal here you can see I've enabled HubSpot as a source. When you're using hubs of source Deep Research mode, it doesn't necessarily guarantee that GPT is going to be querying that source. It may or may not use data from your HubSpot. It's a faster mode where it's trying to pick and give you the best results it can with some more limited context. When we actually get into Deep Research mode, on the other hand, it's going to guarantee that it's actually pulling from your HubSpot source and then it's going to show you exactly when and where it's querying which really does make it ideal for these more like business critical tasks I would say. So that's it. Very simple, very straightforward in getting it connected. We'll send this slide deck out as a follow up as well and there's some resources there to the HubSpot knowledge base just to help you get connected directly to the Deep Research connector. So yeah, I think like now as we get into prompting, I think some of the things that.

Oh no, Matt, I think we're losing you.

Oh, is my, my network coming in and out? Yeah, I don't know if I'm back on. Sorry about that, might be screen share, too much going on. But yeah, I think, you know Kathy, maybe I'll let you kind of touch on this because we've really been collaborating on like what makes a great prompt.

Yeah, so this is a really good question. One of the things that makes a great prompt is it's a lot about like the goal. It's really about the structure of a prompt. Right. So it's about the goal, the scope, the output format and the, and the constraints. So it, what you put into a prompt is going to also dictate what you're getting out of a prompt. So one of the things that we think about when we're talking about data, when we're talking about prompt engineering is the structure of the prompt really matters. One of the things that we're going to do when we send out this slide deck. And then of course you're also going to get the swipe file that Matt has built. You're going to get access to over 190 prompts and they are all so well built that you're going to see kind of the structure of these prompts in action. And when we're talking about a great prompt, thinking of something that is really robust. And if you've played with AI really in any capacity, you'll kind of see the difference between asking AI.

Can you.

Give me a recipe for spaghetti? Vs can you give me a recipe for spaghetti that is rich with, you know, these types of ingredients? And I want meatballs instead of meat sauce and I'm looking for something that is slow roasted and like so you could see the difference between I want a recipe for spaghetti and I want something with all of these details. And then also when we talk about what constraints look like, I don't want anything that includes, you know, this because I'm allergic to it. So you get an idea of what that might look like when we're talking about prompting. If you give it garbage input, you're going to get garbage output. So I think that's really important to note. Matt, did we want to move on to looking at a prompt?

I think Matt might be frozen.

All right, well, all right.

I can hear. Yeah, okay, I can hear you guys. Can you, can you see my screen share still?

Yes, I can.

Awesome. Okay, sorry about that. Yeah, so we'll be sharing this. This is the swipe file that Kathy was alluding to here. So it's broken down by job function. There's tons of examples in here. One of the ones that we were going to just look at today, which has been really, really helpful for our clients, is the idea of the win back watch. It's obviously a sales related prompt. You can see at this prompt it has all those elements that Kathy alluded to. It's telling HubSpot. Hey, here are the data analysis requirements or sorry, it's telling GPT. Here are the analysis requirements. You know, here's the output format we're expecting. Here are the additional deliverables that we require. So these prompts are really used as inspiration. You can simply copy them, take them and run them in GPT or you can leverage them as an example, feed them into something like a GPT or a CLAUDE and say, hey, based on this prompt, can you customize it for my specific use case? If you want things to be a little Bit different when we go into actually running this, we're just heading back over into deep research mode. I like to ensure that we're not searching the web. We just want to look at HubSpot as our source. I can run that prompt and now GPT is going to get into deep research mode. It's just going to confirm a couple details. Just want to ensure that we are looking at your HubSpot data. Yes, please proceed. It's going to now get into deep research mode depending on the complexity of the prompt and what you're asking. We've seen these research projects can take anywhere from let's say three to five minutes just for the interest of time. We've kind of actually already ran this and can show you what an example looks like once the output comes back to you. So you can see it's done its best to kind of provide a strategic overview. The prompt like what we're really asking for is hey, find us accounts in our CRM that are, you know, closed lost deals, have new activity, filter it by visitor behavior like don't include customers, let us know like what's the new activity, have they been on our website, was there a meeting and then give us some results with customized outreach emails that we can use for those accounts. Now I'm just using dummy data so there's not a lot here when it comes to the actual results. But you can see it prioritized a couple deals based on things like website engagement. You know, it's given us what was the original loss reason on that deal, what was engagement score like, why does it think that this is a good deal to re engage? And then it's actually presented us with some personalized win back emails. So now that we've got this sort of research.

All right, so I'm going to step in now that now that there's that research done, one of the things that you can do is you can take that information and you can move on to the next step. So one of the great things about you know, the OpenAI connector and deep Research is that what makes it different than a regular HubSpot report is that it helps you answer the question of like HubSpot gives you data answers but deep research allows you to do the next steps. So it's the data and so that's one of the things that you're looking for when you're thinking about, you know, you're looking at like for example right here we can see the additional deliverables. So it's looking, this is the Next step with your data. Matt, I see you're back, so I'll let you take over again.

Yeah, so sorry about that. No, I think you nail on the head, Kathy. It's like rather than a point in time report, we can take those, deliver like we can take this information and go many steps beyond it. Of course, this is just the start of the conversation with GPT as well. So one of the things that's great is it's, it's also sourcing all of its information. Right. We can see it, a very clear log of where it went in HubSpot to find this information. And then it's to our HubSpot so we can see the exact records that it's talking about and just start the conversation. We can ask questions, we can kind of continue that thread once we've started the deep research mode. I think just one thing to touch on that's been cool for us is rather than, you know, typically where like a sales leader or marketing leader will be the one that sort of controlling the dashboarding and reporting, the OpenAI connector really does open up the opportunity for maybe like a specific sales rep or if you have people running a specific region, they can actually be empowered by asking their own questions of the data. It's like, hey, how many, you know, I want to see the activity related to my region and my accounts, they can start asking GPT for their own responses related to the data in the HubSpot. So that's, that's been one really, I think, exciting use case that sort of come out of this. I think one thing to touch on with that, Kathy, and we've talked a lot around, how can people make the most of their HubSpot data to get the best results from the connector.

Oh, great question. So one of the things that we want to be thinking about is we want to be thinking about, you know, I was talking about the importance of a quality prompt and I was saying, you know, garbage in, garbage out. Right. The same thing kind of applies with your DEC data. When you're beginning to work with deep research, you also want to know what your data looks like. Anytime I'm working with a customer, one of the things that we want to talk about, first of all, I love reporting. Reporting is my jam. I love it all day, every day. And when I'm first working with a customer, we'll run reports together. And for example, we might run basic leads report and they'll, you know, sometimes we'll run a report and it's right on. Sometimes we run a report and they'll say, yeah, I did not get 100 leads last month. And what we'll find out is that there was some property, for example, that you know, there was a duplicate property and we put the wrong property into the report. That happens a lot. The older your portal is, the greater the likelihood is that there have been duplicate properties created and those properties are gathering information and your portal gets messy. That likelihood increases the more super admins there are or the higher turnover you've had in your HubSpot admin position. So what I tell people is when you're sort of embarking on sort of a deep research project, I think it's important that maybe you do a little bit of data cleanup before you get started. And that looks like kind of reviewing your properties. HubSpot makes a data model that you can go in and you can look at all of your properties. You can see if there are duplicates, you can see how filled out those properties are. And that gives you an idea of whether or not you can, whether or not you need to delete some. You can transfer data from one property to another. So there's a lot of things that you can do. So that is the kind of the first step. The next thing is to, to begin working through knowing what the internal values of your property names are. So the reason that's important is that the Deep Research connector is going to pull from the internal values of your properties. So if you, for example, there's external values and there's internal values. If the internal value of a property, say for example, you have. This is a great example. State properties. State properties are often, you know, sometimes they'll be spelled out and then somewhere along the line somebody decided that they wanted to be abbreviated for everybody externally so that when they're looking at a form, they're not looking at the fully spelled out name. So what, what makes that important is that when you're referencing the state name in your Deep Research, you're going to want to use the fully spelled out value. So I think that's also important is that you just want to be sure that you're using the internal values when you're using Deep Research and not the external values. And Matt, you shared a really great story about this with me that I thought would be really fun to share with this.

Yeah, it's a funny one. And the insight in the chat, I'll keep my video off. In the meantime, if it's really not working for us, please let me know and I'll try and switch networks but the funny one, yeah, because we obviously whenever we kick off with the client, we start with an audit and it's always about that. Taking a look at the data model, looking at the internal property values, also seeing how many duplicates there are. But this one was really funny because it started, the company was started in Latin America and then migrated to North America. So all of the internal values were in Spanish and all the external values are in English. Which of course when you start to leverage something like the Deep Research connector, where it's using the API values, you're just getting really muddied responses. I think one thing that we've kind of come across with our testing as well is actually this theme fits into all of the other AI products that HubSpot is starting to ship as well. So if you're starting to build your own AI assistant in HubSpot within Breeze Studio, leveraging the internal values for even something like the pipeline, like which pipeline do you want to point your assistant to, and giving it that numeric value for the pipeline that you want that assistant to run on is going to just give you that much cleaner results. I think as a concept, this is something that's really foundational to understanding as you start to get into playing with these AI tools more and more.

Absolutely.

Now, one of the things that we wanted to just be really explicit about is there are some known challenges. I have been having a lot of conversations with people on LinkedIn about, hey, I've been trying the connector and I'm just not getting the results. I think some of the known challenges are that when it comes to prompting, if they're really long and lengthy prompts, that can cause a bit of a timeout. If you're asking it just basically to do too much, breaking your prompts down into more specific action oriented prompts is going to be very beneficial. The other thing is cross object handling. If you're asking the ChatGPT to do research on contacts and how that associates with companies and how that associates with tickets, that type of cross object joining has really been challenging from everything we found in our testing. And then finally, this is not a limitation, it's just the reality where it's at today is there's no access to engagements, so you're not going to be able to ask ChatGPT things about non core objects. So if it's calls, meetings, notes, tasks, it doesn't have access to those objects, so it just simply can't do reporting on them. And then the reality is it's AI. There are occasional Hallucinations, which is, I think, just bolsters the points that you made, Kathy, about you still need to know your data and be able to take a look at the results and say, okay, through a critical lens. Is this. Can I trust this 100%? Go into the sources and just confirm what it is that you're seeing. That all being said, we'll chat three months from now, Kathy, and I'm sure you've got a great insight on just how fast the development is on these things.

Yeah, it's happening so quickly. The speed by which technology is developing is faster than we can develop behind it. So one of the things that I'm telling people is that if there's a prompt that isn't working today, I don't think that you just abandon it. I think that you take that if there's, I mean, if there's a prompt that you really want to work, I think that you sort of put it on the back burner and sort of table it, put it on a separate tab of like, try this again in three months. Try this again in six months. Because AI and technology is developing so quickly that it may not work today, but it may work again in six months. And I think a perfect example of that is Breeze Assistant. The things that Breeze Assistant is doing today. Listen, I work at HubSpot, and I was not a fan of chat spot. I felt like the things that I wanted it to do, it couldn't do. And some of that was the technology, and some of it was that AI just wasn't, Wasn't fast enough. And now I look at what it's doing and it's so far beyond even what I'm asking it to do. I'm, I'm way behind its, its capabilities. So I just think all of that to say, if there's something that you really think that it should be able to do, don't abandon it just because it can't do it today. Take it, put it on the back burner, and then try it again in three to six months and see what it can do.

I couldn't agree more. And I think that's part of the. We're all being challenged to be even more curious, right? The more curious that you can be, the more benefits and the more results you're going to get from. From AI. So now there's a lot of questions flying around in the chat, and one of the ones that we get all the time is like, yeah, privacy and security. So I think just being super upfront and transparent about this is here's what we know today the connectors are 100% read only. So what does that mean? When you connect HubSpot to OpenAI.

It.

Doesn'T have two way access. It can't write any information into your HubSpot, it can only read information from your HubSpot account. The other thing there is that it's session based, so there is no storage or training. The asterisk there is. We know for sure. When you're on a paid business or enterprise plan from OpenAI, you're opted out of sharing training data by default. Now if you have a paid individual plan, something like a plus subscription, it's upon you to actually opt out of sharing data for training purposes. So once you've opted out of sharing data for training purposes, that just double guarantees that any of your HubSpot information is not being shared with OpenAI. And the fact that it's session based again means that that's, there's no long term storage of that information. We've included some links to both of those privacy like privacy agreements if you want to read into them a bit more. But I think that's sort of what I've, I've come to learn from the OpenAI side of things. Kathy, if you want to share from HubSpot's perspective, like how are they approaching privacy and security around these connectors?

I mean, I think that's exactly right. You know, they are. When they rolled this out, they were very careful to tell us that there was no information that was being mingled with other customers data, that there was no information that was being used to train OpenAI. So that was really important for us to know and for us to share with our customers that your data is protected, it is safe, you know, so as long as the only thing that we are careful to do is to ensure that when I'm working with a customer and I'm asking them if they're using AI tools, if they're using Chat GPT, if they're using any, any AI tools outside of, you know, in either within HubSpot or outside of HubSpot, that they have an AI policy that allows them to do it. Because really it's up to your legal team and your AI policies as to whether or not it allows you to do that. So some of my customers right now are really eager to see the Gemini AI tool get launched so they can start using that because that's what their company allows them to use. Once we sort of get all of the tools kind of rolling out, then I Think there's going to be a lot more coverage out in the marketplace between HubSpot and our customer base. But from our perspective, we take data privacy very seriously and we don't want your data being commingled with other customer data. We don't want your data being mingled with Internet data. So we don't want like, I shouldn't be able to just be chatting about spaghetti and asking about, you know, customer X and you know, pulling up your data through ChatGPT. So that's really important to us and we do take it very, very seriously.

Yeah, no, I think that's, I think just. Yeah, hitting home on that point too. There's also some like GDPR compliance as well where depending on the type of data that you're actually querying, the connector will actually route that accordingly to the right HubSpot data centers, which is I think really, really cool. It just commits the privacy and security. I think one other thing just to. To touch on as well. I'm kind of watching the chat and I think there's a lot of great questions around. Like people are mentioning Claude, people are mentioning, you know, Gemini and OpenAI and I think from like a sort of a power user of these systems. One thing that I would definitely encourage is when it comes to prompting, especially like play around with the different AI solutions that are available. Like if you've only ever used ChatGPT, I would strongly encourage you to go check out Claude because I think what you'll start to, you, what you'll start to understand is that they kind of each have different like things that they're really good at. Like I find that for actually writing these prompts, Claude is really, really great at writing prompts. It gives very thorough outputs and is very detailed in the way that it does outputs. I find for. For myself, like Gemini is really strong when it comes to, you know, helping with workflows or, or like actually coding like things that are a little bit more technical because the output context is a little bit longer. So that's just a bit of an aside from all the conversations going on in the chat thread is if you're like a loyalist to one of these platforms, I would definitely strongly encourage try out some of the other ones, test them out and you know, when the new Gemini connector comes out, Gemini and try it out if you've never been in there. So I think we are at. Yeah, I mean we wanted to leave lots of time for Q and A. I sorry that my. Work kind of cut off there. So apologies. But yeah, we Want to open this up to Q and A. So, Deanna, if you want to lead us through the questions that have come through the chat and allow people to hop on, that'd be, that'd be awesome.

Perfect. My admin friends, if you could go over to your Q and A tab and upvote any of the ones that are over there right now, there are quite a few. I will sort those and go from the most voted to the least and let's see how many we can get through. We've left quite a bit of time, so hopefully that helps. So, Kathy and Matt, if there's anything that you want to add any color to or if you've already answered it, to add any additional sparkle to it, that would be great. Hopefully I've given you guys enough time to go over there and add your votes. Let's see what we've got. Wow. With 19 votes from our very own here on the host, how do we answer the security compliance questions internally? Very big question. 19 of votes.

Yeah, I think, I mean explicitly, I guess, to get a little more color on that. Jeanette, is that from like within your organization? How do you have that conversation internally about, you know, can I use this and hear the security concerns?

Yeah, and I'll, I'll just add some, some color to it. So we tend to be very, very, very conservative with, with any integrations. And although a lot of people are playing with AI on their own, we don't, we've got a policy that's like, don't connect any agents, don't connect anything that has anything to do with any of our customers. And so it's, so it's kind of hard from a. Because, because our compliance officer basically says that, that a lot of the, a lot of the verbiage in the privacy policies basically say it's up to you to use it, how you, you know, to feed it what's important to you and kind of take, take the onus off of the vendors. So that's the, that's the struggle I'm having internally.

Yeah, I mean, I think from my perspective, a compliance review, depending on how stringent it gets, there's always going to be intricacies there. I think some of the things we highlighted today about it being session based, it's not actually accessing your data until you're putting in that prompt and explicitly pointing it to say, hey, can you go and find this information for me? Which really isn't much different than when you're leveraging really any of the other HubSpot integrations. So I would sort of challenge you to say, like, what are the other integrations maybe that you're leveraging? Do you have a, you know, a QuickBooks integration or, you know, whatever popular integrations you have in your platform, they're leveraging the API to get data from your HubSpot portal, which is essentially the same mechanism of GPT is its answers within that session window that bolstered with, if you're on a business plan for GPT, the sort of promises that have been made there within their policies around the not sharing training data. I think those kind of two things combine to say that's probably the best response that I would sort of carry forward to a compliance officer and to say, like, it's really no different than another integration that we have in the system. I don't know, Kathy, if you want to share any other color or add any light to that as to how you've navigated it with other, some of your clients.

Yeah, you know, it's interesting because a lot of my customers, you know, when they do have AI policies, sometimes they're almost like intentionally vague and it's like, you know, don't do this. But if you do, be careful. You know, it's like, oh, okay, thanks so much. I did put a link in the chat that has, that we also share, I think in this deck that has a lot of really great information about how exactly HubSpot uses AI and that it's read only that it's not training on your data. I think it goes pretty deep. I think in terms of, you know, how you, how the data is used, how it's not used. And I think this is pretty good at like in terms of the high level talking points with your, with your legal team or with your IT team. Some IT teams are just like, it's a hard no. I think also in the chat I've been seeing that it's important that you go into any AI model and turn off the option to share your data. So that I think is a beginning point for anyone using AI is that you should be going in and you should be ensuring that your settings limit how you're sharing your data. So that I think is the starting point and then reviewing that link, we've got it in the deck as well. But I think it's just important to understand that HubSpot doesn't mingle your data. We don't use your data to train the AI tool. And so it sort of like builds like, I don't want to use the Word firewall, because I believe that it has bigger, you know, instances, but like sort of builds like a, like a holding tank for your data. It reads it and then it doesn't go anywhere else aside from that. So I think that it's just important for you to know that that's how the data is protected.

I will make sure all of those links are shared as well. So they're on the recording and everybody has access to everything. Let's see. Karen's question upvoted with 28. So this is a big one. How does using the connectors with an external LLM differ from using HubSpot's Breeze Assistant?

Yeah, that's a great question. I'm happy to take this one. Karen. The reality of the Breeze assistance is when we talk about mcp, that sommelier, if you imagine the sommelier can go into depending on which Somalia you're talking to, like you've got OpenAI, you've got Claude, you've got GPT. They all are going to have a bit of their own different style and flair. As I was kind of talking about, they each are kind of have different strengths. What we know about Breeze so far is that the Breeze assistants are actually powered by kind of a combination of large language models, but at the core it is still OpenAI's GPT4O mini and GPT4O and then stable diffusion as well. Sort of one of the other models that's used. So it's taking a very similar approach, approach of using HubSpot data as a source and your user input, running it through those models in a secure way and then updating data sources and providing things like summaries. So I think one of the benefits and the cool things of using these other large language models as well is again, they're going to have their own sort of style and flair and be able to kind of provide responses in potentially different ways. And then of course, with GPT you can flip between models. While, you know, we have the legacy models now we're on GPT5. So when you're using the connector, you can sort of experiment with those different models and see what type of responses and results that you're getting versus with Breeze. Like I said, it's been sort of hard coded into GPT 4.0 for right now, and I'm sure that will progress over time, but that's sort of the way that they're powered at this point in time.

Awesome. Let's see. Robin has a little bit more of a technical question. So hopefully this is Clear. If we connect our GPT as super admins, can anyone in our association connect too? So Robin, I'm going to infer. You mean to your instance of your connection?

Yeah, that happened to touch on that as well. The way the connector works right now is it's more like connecting like a Google Calendar to HubSpot like it is an individual connection. So whoever wants to query the HubSpot will set up the connector within GPT directly. So what we're doing is empowering all of the users with the HubSpot or the GPT Pro subscription and then just giving them the directions on like hey, if you want to ask questions of the HubSpot, here's how you connect it via the connector.

We've got another one with 19 of votes. What prompts can we use in GPT or ChatGPT to extract data that HubSpot filtering won't be able to extract? I see that the most useful prompts are focused on deals companies, but sales teams in my company work outside HubSpot so I wonder what it provide for marketing automation and email creation team. Also, as it can't read custom objects and email conversation parameters, are there any tips and tricks that you can share covering other areas?

Yeah, it's a big question. Kathy, I don't know if you have any thoughts and ideas about the specifically the HubSpot filtering. I'm happy to touch on some of the tips and tricks. It's like maybe how to get around that but if you have any thoughts.

Yeah, my initial thought went to workarounds is well since right now it's not able to pull from those custom objects and from, from some of those other modules, I would say my, my initial thought was probably to go around it as well. So I'm thinking to export the data and then import it as a spreadsheet and then report on it that way which is probably what I would do using, you know, deep research and, and then asking it to report on the, the data that you're inputting via spreadsheets which you know, defeats the purpose of. I fully understand that is what I'm saying. On a webinar we're talking about AI connectors. But this goes back to what I was saying is that it can't do that right now. But I know that they are feverishly working on what else they can do with these and trying to get, get more data out of them. So you know, over time I think more things will be uncovered. I mean for you know, the sake of transparency, we, we just started Being able to report on like forms, you know, in HubSpot. So there are, there are a lot of things that are continually evolving within the HubSpot platform. So I expect that in not very long we will be able to do more in some of these connectors. So I will would say stay tuned but in the meantime I would export the data and then import it into Your, into your ChatGPT tool and then do your research that way and still use similar prompts. I mean I would still like build very robust prompts using the swipe file as a, as a template and then run your prompts that way. I don't know Matt, if you have something different that you would suggest.

No, yeah, I think, I think you hit the nail on the head and even just what I'm excited about to make that type of like cross object spreadsheet work easier is the. We just got enrolled in the new Data Studio beta, which is really, really cool. Right. Allowing you to take all these different HubSpot objects, external data sources, join them in what looks like a really hyper powered CSV type of environment where it's columns and rows. Once you join all that data intelligently, you can use it in workflows, you can write it to the CRM, you can export it. You can see where HubSpot's going with this, which is we're going to help you organize your data really, really well. And then as these connectors evolve, it's like plugging them into something like a data studio so you can start to ask questions on it. I think that's where we're going. But like you said, wait three months and come back and I'm sure that answer is going to be a little bit different.

Awesome. All right, this one's a two part question. I'm going to break it into two because there's the related but different. I assume the connector cannot read highly sensitive data properties. Correct.

That's an interesting one. In terms of like I. From my understanding, I don't think there's any explicit way to turn off read access to like specific properties that are tagged in some way. I don't know Kathy, from your perspective. Yeah. If there's any. Because there's no real flag on a property in HubSpot that marks it as highly sensitive where GPT would not have access to it.

I don't think so. I'm going to see if I can find out live while we're on this call, but I don't. I don't think so. I'll see if I can answer that in the chat as well.

It's a good, It's a good question. Yeah, I don't know specifically the answer to that either. And then. Yeah, Dan, I'm looking at the two parter there. Does it use credits? It doesn't use. Yeah, it doesn't use Breeze. Breeze credits at all. Honestly, depending on what type of GPT subscription you have, you are limited in the amount of deep research you can do. So there's always credit somewhere, right?

Somebody's always charging for something.

Yeah, exactly.

Let's see. Chris wants to know what kind of guardrails exist that would prevent a user from exporting data from the GPT connector that they would be prevented from exporting from HubSpot directly.

Yeah, it's another good question. Yeah. And to be honest, Chris, this is kind of in line with the last question of like, what are the. What are the bounds that we can put on the connector? And I think from our testing thus far, I don't. I don't also have an explicitly good answer to that either. About which what are the permissions that you can put on the connector and how do you limit the data that HubSpot or that GPT has access to from HubSpot? I don't. I don't. I. Yeah. In all our testing, I have been more so conducting it from like a super admin perspective. So I don't have a great answer for that. It's a great question, Kathy, Maybe in your research. I don't know if anything's come up from, from your perspective.

No, but I think that, that, that is also a good question. I'll have to do some research and maybe follow up on that.

Yeah, Some of the admins in the chat confirmed there are specific API permissions for sensitive data properties. So if you're experiencing. You have sensitive data properties turned on, which I'm pretty sure you have the ability to turn that on and off as super admin that explore what that looks like in terms of the API permissions. I don't know. I can't speak to what that means with the connector itself. All right, we have another one. Let's see. Brianna wants to know any timeline on releasing the HubSpot AO audit. Could you give a quick rundown of what's in it and how it helps with GTM performance? Brianna might be a different group we're looking for, but I can definitely share this over. Okay, let's see. Nero wants to know. My organization uses Langdoc since we're in the EU and it has our HubSpot account integrated? How best can I utilize the AI connector?

That's a. Another great question. I mean specific to the eu. The only real insight that I have on that is I know if you're prompting with OpenAI and you're based in the EU, the data, all the data that gets returned is routed through the HubSpot EU data centers. So I know that there's like, there's a level of GDPR compliance that comes with the connector for EU customers. Beyond that, that's about as much as I know. I hope that answers your question but it's kind of a specific.

Yeah, if it's, if it's not, definitely reach out and you know, use your HubSpot admins as a resource. If you have questions, use that LinkedIn group use, you know, you have the chat here connect with people. These are people that are going into their platforms all the time trying things out. So you know, if the group here doesn't know, bring it to the Internet. Liz wants to know the chatbot linkage is not available at the moment in Europe. When will this be sorted?

Linkage? I don't know. Kathy, do you, have you heard about that, the chat bot linkage?

I have not.

Is that like the chat bot. I'm just. Sorry, I'm just curious. There's a lot of chat bots floating around. Is that like the, the Breeze chatbot agent or like specifically the Deep Research connector Customer agent? Yeah.

Liz, if you want to clarify in the chat can definitely give a little more information to that, that'd be great. While you do that, let's answer Paul's question. Does having a rich and consistently filled in property description help the connector select the right properties when HubSpot is used as a source?

That's a great question. I think from an API perspective, I don't believe descriptions are being querying. That's really where from our testing being more explicit about the internal property names has been producing the best results but it's actually, it's a great question Paul. Like I haven't explicitly tested in an environment that has rich descriptions versus not rich descriptions. I can't speak outside of my own perspective but yeah, the internal values has been the most important piece so far.

Then there's a follow up question probably for Kathy. If so, will HubSpot add in generate Breeze action for property descriptions like it has for workflow descriptions?

You know, I feel like they've been doing a real. I don't know if that's on the roadmap, but given that they have built that in for a lot of other things like reporting and workflows and things like that. I would guess that, that that property descriptions may. It may be next or may. Well, I wouldn't say next, but it may be on there. But I'm speaking completely out of turn. I'm literally just a business consultant. So like, I don't want to say I'm super low on the list, but they don't, they don't talk to me or ask me or you know, like I, I am nobody here. So I would say that given the direction of like adding more information to HubSpot's portals and descriptions, I would say that it would be a reasonable, reasonable guess that eventually that could be where they're going. So that's a cautious enough statement so that nobody comes back and says, well, Kathy said on an admin hug one day that it's definitely on the roadmap. It's. I'm not saying it's definitely on the roadmap, but it would be a logical guesstimate that perhaps maybe it is on there. I would also say too that just as somebody who does a lot of employees implementation and works with a lot of customers, that more information is always better than less. I know a lot of people who have taken over for people who are no longer there or, you know, you're doing a lot of data hygiene or data inventory and you're like, what was I thinking when I was doing this, you know, six months ago, five years ago, whatever. And you know, you have no idea what you state of mind you were in. So those descriptions are really worth their weight in gold.

Gold.

So I think investing in that is really worth its, worth its weight whether or not ChatGPT uses it.

And just for everyone's information, I will share the chat and the Q A with Kathy so she can share that with the product team on her side and saying that out loud, I'm going to roll that into the next question, which is probably something for that team as well. But I'm going to say it out loud because it was Ovid upvoted 10 times. Why doesn't HubSpot just build an agent within the application to do deep research that keeps everything secure? Why create a connection that introduces potential risk factors to your data?

So I have my guesses, Matt. Do you want to take it or do you want me to take it first?

You go ahead. You go ahead.

Because I think so. It would be like saying, why doesn't McDonald's make, you know, why doesn't McDonald's make like burritos? Right? Because somebody's already cornered the market on something that's way better and more robust and it would take an extraordinary amount of time, energy, resources to build something that somebody else has already done very, very well. So I'm not, I'm not saying that, that Taco Bell is like the best in the game when it comes to being burritos. I'm just saying that they've already like, they've already cornered the market in streamline burrito making. So all of that to say that, you know, could we eventually get to a place where there is some sort of internal tool that is doing deep research as good or better than open AI?

Maybe.

But for right now to be able to bring that to market and have your data doing deep research in a way that is very secure by the way, I think getting that in your hands so that you can make some of these connections and start getting really deep insights on your data, I think is, I think that connection now exists without having the time, money, resources, you know, like sunk into something that is probably not going to be as robust as something that OpenAI already is. That's my estimation.

Yeah, I think it's a little bit about context and like meeting users where they are as well. One of the things that we, we didn't really touch on but when you're connecting HubSpot, like there's a whole list of other connectors there in the chat GPT options, right? So you can toggle as many of those on as sources as you want. So obviously HubSpot is, is fighting to be the source of truth for your go to market data. But as a user, if you want to cross query different sources, look in Google Drive and then go pull things from, you know, my Gmail calendar and then go look at Notion and look at HubSpot data. You know, GPT already has all those. So I think it's just that from a user perspective it's like why not get into that, you know, participate in that versus trying to just be like a walled garden for some of these, these things.

All right, we probably got one more question available for us. Let's see, let's get one that has a bunch of upvotes, sort. Can you guys keep adding them in here? Let's see if we're a partner, do we connect ChatGPT to each portal or if we connect once, is it connected to all portals or do we have to do it on a per portal basis?

Yeah, so it's definitely Definitely per portal in the demo there you decide which portal you're authorizing and right now it's on a per user basis as well, which we kind of touched on. So whichever user is actually going to be querying GPT is going to have to connect HubSpot as a source in their GPT account.

I love that.

I appreciate all of the questions. I you guys are forward thinking and you're at the very beginning of the adoption curve so I appreciate everyone asking all of those questions. I know it's hard for Matt and Kathy to be able to answer everything but if you are having those questions, make sure you're using your resources that you have available to you as an admin in the this ecosystem. We we've got all the different avenues to ask those questions. Get it out there. No question is stupid. So I appreciate you asking all of them here at the at the admin hug. I'm going to share the rest of the ones that didn't get answered with Kathy and Matt and if they can answer any more and share those that information with you as a community, that will be great. Kathy, Matt, thank you so much for coming to the admin hug. I will see the rest of you next week when we are highlighting our fellow admins at the show and tell. Have a great week everybody.

Welcome to the admin hook no flashes, facts and a whole lot of love get it done accidental or all in by choice administrative fields and how packing.

Time.

Making magic with no headline.

Tiana.

Brings fire in a crew that's here to lift you higher we swap the whims and share the pain no solo in this game it's the action hug where legends walk in from chaos to clarity that's how we win. Study that Challenge accepted Portal perfection fully connected. Quick fix Queens workaround kings custom coded everything tabs open coffee strong you've been solving all along We've got the know how we've got the map even HubSpot product bulls up to chat you're not just her you help shape the tools admins rewriting all all the rules thanks for building with the hug. Thanks for powering the hug. The hinds of the campaign Bush nudge there's a whole crew hold inside it and together powered by insight.