June 17, 2025 • 16 mins
On Episode 10 of Accelerated Velocity, Grace and Peter demo ChatGPT’s Deep Research Connector tool, which allows integration with apps including HubSpot, Google Drive, GitHub, Outlook, and more. After trying out a sample prompt using the ChatGPT <> HubSpot integration, they determined that the tool offers value in specific use cases, but might not be the go-to solution for all aspects of your marketing strategy–at least not yet.
The takeaway: leveraging AI for your marketing relies on choosing the right tools, checking for sources and accuracy, and maintaining a single source of truth. Listen for a rundown on how to protect your data on ChatGPT, how to leverage Deep Research, and what NOT to use it for (at least not yet).
Try HubSpot’s ChatGPT Deep Research Prompts
Chapters:
00:00 - Intro
01:23 - ChatGPT’s Deep Research Connectors
02:07 - Data Privacy Caution
02:40 - Testing ChatGPT <> HubSpot Connector
05:48 - Inaccuracies on ChatGPT
06:30 - Why ChatGPT Misses The Mark On Email Marketing Trends
11:45 - LLMs vs. AI Agents & Integrations
13:09 - Maintaining A Single Source of Truth
14:30 - Call for Input: What Do You Want to Hear Next Time?
15:20 - Outro
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:24):
GraceWelcome to Accelerated Velocity, the podcast that helps you move faster, smarter, and more strategically in the rapidly evolving world of AI. We'll break down the latest AI news and innovations and explore how they impact your marketing, sales, and business growth. We'll dive into the practical use cases, unpack new tools, and cut through the noise so you and your team can adopt with confidence.
(00:48):
GraceI'm Grace Mathews, director of content at inbound TV, a business development agency and HubSpot solutions partner dedicated to driving sustainable business growth. Each week, I'll be here with Peter Malik, our CEO and founder. Join us as we make sense of what's changing and what to do about it.
(01:10):
GraceHello. Welcome to episode ten of Accelerated Velocity. Peter, can you believe we're on episode ten already?
PeterI am beyond excited, grace. We actually officially in double digits and they said it couldn't be done.
GraceAll excited to see what's in store. And today we're taking a look at ChatGPT is new connector tool, which connects a number of different apps natively directly with ChatGPT. And that includes Google Drive, GitHub, HubSpot, Dropbox, a number of other apps. And basically what it is is that within ChatGPT, you can conduct a deep research query and pull on data from any of your connected apps to answer your deep research prompts.
(01:53):
GraceBut there are also opportunities for custom connectors as well to connect apps that aren't natively integrated with ChatGPT. Just a note on data privacy. For anybody who's interested in setting up any of these connectors, you first of all have to have a paid version of ChatGPT to use this feature, and ChatGPT will not use your connected data for training of their models.
(02:17):
GraceBy default. If you're on a team, enterprise or Edu account. However, if you have a plus or a Pro plan, you need to make sure to go into your settings and turn off improve this model for everybody. If you don't want the data from any of your connected apps and your deep research queries to be used by ChatGPT for training, so we'll jump into the meat of it.
(02:40):
GraceNow let's do it. Obviously we're a little bit biased because we're a HubSpot solutions partner. We're very interested in the HubSpot and ChatGPT integration. We set it up on our end, and we've run a couple of sample queries to see how it works. And talk about what's beneficial about it, what's not beneficial.
(03:01):
PeterHey, that's what I'm here for.
GraceSo the prompt that I asked ChatGPT initially was to create an analysis of our email marketing campaigns from the past three months that we've been sending through HubSpot, and I asked it to focus on identifying subject line trends that are highly effective, and then to also provide an analysis of our general email marketing statistics and focus on how we could increase click through rates and identify some opportunities that we may not have considered yet.
(03:33):
GraceSo first thing off the bat, once I asked that question, ChatGPT asked me a number of follow up questions to kind of narrow down my prompt. I think that's a big benefit of this deep research feature is that unlike a typical ChatGPT prompt where you might give it a weak prompt and you get back whatever ChatGPT is going to tell you with this deep research feature, ChatGPT is making sure to clarify important details in order to run the query.
(04:04):
GraceAnd so I answer those questions, and I got our deep research analysis back, which includes a performance summary, some subject line trend analysis recommendations, and some benchmarking comparisons about our email performance versus others in our industry that we got back from.
(04:26):
PeterSo I think first of all, the the numbers that you just scroll past, here we go that, you know, I'm really happy about our open rate. It's not like I wasn't aware of it, but it's a good healthy open right. It's around 30%. It's saying industry benchmark 42%. Average across all industries. I have a hard time believing that.
(04:50):
PeterHonestly, as I've seen so many, so many accounts that we've dealt with that have 4% open rate or 8% open rate and 12% open rate, and and it's our job to get them up. But there are a lot of emails going out there that are not, optimized to be opened. So I take issue with that. But nevertheless, I'm happy for the report about us.
(05:16):
PeterClick through rate says 3%. And, you know, the range is, 2% overall average. So we're running ahead of that, which is fantastic bounce rate is up approximately 1%, which is actually it's higher than I would like it. But ChatGPT is saying normal is less than 2%. So I guess I'll accept that. But I'd love to have, definitely under 0.5% of, bounce rate and a subscribe rate.
(05:46):
PeterThat's that's totally acceptable. I like.
GraceThat. One other thing that I wanted to mention, I, I went and did a separate little search while you were reviewing these results, Peter. And the performance summary that ChatGPT is giving us, it's not entirely accurate to what I'm seeing on our email marketing dashboards in HubSpot for the exact same time period for example, our current open rate for the past three months for newsletters specifically is just under 37%.
(06:21):
GraceWhereas the performance summary report here is saying that the open rate for the past three months is 30%, that 7% is a pretty large disparity in my mind.
PeterSo that's a huge, huge event. We move on to subject line trend analysis, and it gets dicey because if you look at personalization in use of recipient's name, so it says emails that included the recipient's first name or tailored content saw significantly higher opens. And showing personalization subject lines can boost open rates by 50% or more. And so first of all, my understanding is that that was true.
(07:05):
PeterI think it's fair to say maybe 4 to 6 months ago that was true might have been a little bit longer, but it's not true right now. And I realized that since it's something that's changing literally every week, that it's at least right now, it's impossible for ChatGPT to give you actionable and accurate data back on this particular point, which I think is interesting.
(07:29):
PeterAnd it really sort of goes to the overall subject of like, you want to pick your battles if you're using AI, relying on it for factual information, you want to figure out what you can rely on and and what you probably can't rely on. And, and I would say, flip through this overview and also, you know, sort of just knowing how all UM's work, if something is changing on a daily, weekly basis, you're probably not going to get accurate information on ChatGPT or any other low ball things.
(08:03):
PeterSo subject to change. But but that's my thought on that. And it's not it's not to fault ChatGPT in any way, but it is just to say that there are things that all the all I'm so much better at, things that they're not so good at.
GraceAnd I think some of these challenges are potentially, in part due to the difference between accessing AI tools within a platform like your CRM or your marketing automation platform, and accessing your CRM or your marketing data within. And, which is what we're doing now, though HubSpot has a number of native AI features which can, you know, provide you insights on your reporting, analytics and whatnot, which is is very useful.
(08:49):
GraceAnd there's still development to be done there as well. But I think, Peter, to your point, if you want specific, tailored insights to what you've been doing versus looking at maybe more so, general industry best practices, this might not be the best approach right now.
PeterAnd obviously all of that subject line data is not taken from our HubSpot account. It's it's just chat going out and being ChatGPT.
(09:18):
GraceRight. And you can see all of the cited sources are from blog dot, hubspot.com, not from our CRM data, even though we do have the connector set up. That is not what ChatGPT is using our CRM data for.
PeterAnd by its by its very nature blog that HubSpot Com is going to have a history, you know, of a lot of articles about this, and if you probably went in and analyzed them against each other, you'd see a lot of shifts. But by its very nature, the blog is not going to provide up to date, accurate information.
(09:50):
GraceSo I guess the the conclusion here is this specific connector might not be ready for these sort of analytical queries just yet. Now, HubSpot does have a number of sample prompts that you can pull from their website, and I'll link the URL in the show notes. That might be a little bit more suited for this. This feature, and I think it's great to play around with.
(10:17):
GraceIf you have a HubSpot account and you're interested.
PeterYeah. You know, I'm going to give into my obsessive leanings and say, in this podcast right now that what we just explained there is huge. You want to start using an LLM to cut your hours to retry and stuff that would take you a long time to do and do it in a well-organized way. And that's certainly possible.
(10:41):
PeterBut at the same time, for where we're at right now, you have to consider the hours of questions you are asking. The Lim questions at the lab can give detailed and accurate and up to date answers for. And so the answer to that question is sometimes yes, sometimes no. But if you're a company that's looking to use Lims in your day to day work, to, to, you know, to make your work more efficient, etc., all that good stuff, you really have to do a sort of a 40,000ft analysis of, okay, what data can I count on when I'm asking the lab?
(11:25):
PeterAnd that's kind of that's kind of what we do right now when we're working with a client to build a custom AI solution, is we want to make sure that the data that's coming in is pristine. Is it that it's not going to be 100%, but it's it's going to be at least pretty much as accurate as you can reasonably expect.
(11:45):
GraceWhen is the right time to rely on native AI tools, like, for example, native HubSpot AI, other tools native to your marketing automation platform, or your CRM of choice, versus to seek custom integrations that bring specific AI capabilities into the platforms that you rely on as a business every day?
(12:09):
PeterWell, I think that, you know, there's a couple more than two categories here. One of them is AI intelligence based on certain products. So HubSpot being one, they they've really gone all in on AI. They acquired the company. Clear a bit and incorporate that into HubSpot for for different data enhancement functions and all AI functions. We use as an example, Clickup as project management tool.
(12:33):
PeterAnd Clickup now has its own AI, which is specific to Clickup. And and I would say in general, those integrations are pretty solid because they're just dealing with a very finite data set. And so and so you can rely on it. I think the general chat is it's helpful for many things and it's helpful for research. It's helpful for generating ideas, it's helpful for a general overview.
(12:59):
PeterAnd it's helpful for detailed overview. But you still have to really check your work. And one way to do that is to do it on different LMS.
GraceYeah. And you know, Peter, this gets me thinking about something we talk about all the time as HubSpot partners, which is the value of a single source of truth. In other words, having a centralized platform with all of your data, all of your marketing, sales, service activities, all talking to one another so that there aren't discrepancies. And I think that's shifting or being disrupted a little bit.
(13:36):
GraceBut so many new enticing AI tools coming out from so many different companies, how do we maintain AI, at least to the best of our abilities, the single source of truth right now, while also kind of experimenting with all this stuff?
PeterYeah, I think that I think a single source of truth in regards to ILM and in regards to getting, you know, single source of truth into HubSpot, probably the best strategy right now is to build an agent that that again, researches in multiple platforms and then has some way of reconciling the conflicting data that's coming in. And maybe, maybe that even looks like going out to additional platforms to really see what the consensus is.
(14:22):
PeterAnd sometimes in today's world, especially, the consensus isn't necessarily that accurate. But nevertheless, I think that's the best shot for getting a single source of truth as far as I type data into HubSpot. For anybody who's watching this podcast, I'd love to get your feedback about your opinions of how you want to consume information, especially around AI. You know, do you want the unvarnished truth or, you know, is it is it more fulfilling to find the bright, shiny object that may or may not be the real bright shiny object?
(14:58):
PeterSo I love I love your input on that. And, because it's it's a complex world out there.
GraceAnd also, I think there's a different value of knowing what the, the latest buzz is on a day to day basis versus letting something settle and then coming back to it. Thanks for listening, everybody. Make sure to give this podcast a positive review if you can.
(15:23):
PeterAnd subscribe, subscribe. Subscribing is a good thing.
GraceAbsolutely. And any other link or reference you might need or want will be down in the show notes as usual. Thanks for listening.
PeterSee you next week.
GraceWelcome to Accelerated Velocity, the podcast that helps you move faster, smarter, and more strategically in the rapidly evolving world of AI. We'll break down the latest AI news and innovations and explore how they impact your marketing, sales, and business growth. We'll dive into the practical use cases, unpack new tools, and cut through the noise so you and your team can adopt with confidence.
(00:48):
GraceI'm Grace Mathews, director of content at inbound TV, a business development agency and HubSpot solutions partner dedicated to driving sustainable business growth. Each week, I'll be here with Peter Malik, our CEO and founder. Join us as we make sense of what's changing and what to do about it.
(01:10):
GraceHello. Welcome to episode ten of Accelerated Velocity. Peter, can you believe we're on episode ten already?
PeterI am beyond excited, grace. We actually officially in double digits and they said it couldn't be done.
GraceAll excited to see what's in store. And today we're taking a look at ChatGPT is new connector tool, which connects a number of different apps natively directly with ChatGPT. And that includes Google Drive, GitHub, HubSpot, Dropbox, a number of other apps. And basically what it is is that within ChatGPT, you can conduct a deep research query and pull on data from any of your connected apps to answer your deep research prompts.
(01:53):
GraceBut there are also opportunities for custom connectors as well to connect apps that aren't natively integrated with ChatGPT. Just a note on data privacy. For anybody who's interested in setting up any of these connectors, you first of all have to have a paid version of ChatGPT to use this feature, and ChatGPT will not use your connected data for training of their models.
(02:17):
GraceBy default. If you're on a team, enterprise or Edu account. However, if you have a plus or a Pro plan, you need to make sure to go into your settings and turn off improve this model for everybody. If you don't want the data from any of your connected apps and your deep research queries to be used by ChatGPT for training, so we'll jump into the meat of it.
(02:40):
GraceNow let's do it. Obviously we're a little bit biased because we're a HubSpot solutions partner. We're very interested in the HubSpot and ChatGPT integration. We set it up on our end, and we've run a couple of sample queries to see how it works. And talk about what's beneficial about it, what's not beneficial.
(03:01):
PeterHey, that's what I'm here for.
GraceSo the prompt that I asked ChatGPT initially was to create an analysis of our email marketing campaigns from the past three months that we've been sending through HubSpot, and I asked it to focus on identifying subject line trends that are highly effective, and then to also provide an analysis of our general email marketing statistics and focus on how we could increase click through rates and identify some opportunities that we may not have considered yet.
(03:33):
GraceSo first thing off the bat, once I asked that question, ChatGPT asked me a number of follow up questions to kind of narrow down my prompt. I think that's a big benefit of this deep research feature is that unlike a typical ChatGPT prompt where you might give it a weak prompt and you get back whatever ChatGPT is going to tell you with this deep research feature, ChatGPT is making sure to clarify important details in order to run the query.
(04:04):
GraceAnd so I answer those questions, and I got our deep research analysis back, which includes a performance summary, some subject line trend analysis recommendations, and some benchmarking comparisons about our email performance versus others in our industry that we got back from.
(04:26):
PeterSo I think first of all, the the numbers that you just scroll past, here we go that, you know, I'm really happy about our open rate. It's not like I wasn't aware of it, but it's a good healthy open right. It's around 30%. It's saying industry benchmark 42%. Average across all industries. I have a hard time believing that.
(04:50):
PeterHonestly, as I've seen so many, so many accounts that we've dealt with that have 4% open rate or 8% open rate and 12% open rate, and and it's our job to get them up. But there are a lot of emails going out there that are not, optimized to be opened. So I take issue with that. But nevertheless, I'm happy for the report about us.
(05:16):
PeterClick through rate says 3%. And, you know, the range is, 2% overall average. So we're running ahead of that, which is fantastic bounce rate is up approximately 1%, which is actually it's higher than I would like it. But ChatGPT is saying normal is less than 2%. So I guess I'll accept that. But I'd love to have, definitely under 0.5% of, bounce rate and a subscribe rate.
(05:46):
PeterThat's that's totally acceptable. I like.
GraceThat. One other thing that I wanted to mention, I, I went and did a separate little search while you were reviewing these results, Peter. And the performance summary that ChatGPT is giving us, it's not entirely accurate to what I'm seeing on our email marketing dashboards in HubSpot for the exact same time period for example, our current open rate for the past three months for newsletters specifically is just under 37%.
(06:21):
GraceWhereas the performance summary report here is saying that the open rate for the past three months is 30%, that 7% is a pretty large disparity in my mind.
PeterSo that's a huge, huge event. We move on to subject line trend analysis, and it gets dicey because if you look at personalization in use of recipient's name, so it says emails that included the recipient's first name or tailored content saw significantly higher opens. And showing personalization subject lines can boost open rates by 50% or more. And so first of all, my understanding is that that was true.
(07:05):
PeterI think it's fair to say maybe 4 to 6 months ago that was true might have been a little bit longer, but it's not true right now. And I realized that since it's something that's changing literally every week, that it's at least right now, it's impossible for ChatGPT to give you actionable and accurate data back on this particular point, which I think is interesting.
(07:29):
PeterAnd it really sort of goes to the overall subject of like, you want to pick your battles if you're using AI, relying on it for factual information, you want to figure out what you can rely on and and what you probably can't rely on. And, and I would say, flip through this overview and also, you know, sort of just knowing how all UM's work, if something is changing on a daily, weekly basis, you're probably not going to get accurate information on ChatGPT or any other low ball things.
(08:03):
PeterSo subject to change. But but that's my thought on that. And it's not it's not to fault ChatGPT in any way, but it is just to say that there are things that all the all I'm so much better at, things that they're not so good at.
GraceAnd I think some of these challenges are potentially, in part due to the difference between accessing AI tools within a platform like your CRM or your marketing automation platform, and accessing your CRM or your marketing data within. And, which is what we're doing now, though HubSpot has a number of native AI features which can, you know, provide you insights on your reporting, analytics and whatnot, which is is very useful.
(08:49):
GraceAnd there's still development to be done there as well. But I think, Peter, to your point, if you want specific, tailored insights to what you've been doing versus looking at maybe more so, general industry best practices, this might not be the best approach right now.
PeterAnd obviously all of that subject line data is not taken from our HubSpot account. It's it's just chat going out and being ChatGPT.
(09:18):
GraceRight. And you can see all of the cited sources are from blog dot, hubspot.com, not from our CRM data, even though we do have the connector set up. That is not what ChatGPT is using our CRM data for.
PeterAnd by its by its very nature blog that HubSpot Com is going to have a history, you know, of a lot of articles about this, and if you probably went in and analyzed them against each other, you'd see a lot of shifts. But by its very nature, the blog is not going to provide up to date, accurate information.
(09:50):
GraceSo I guess the the conclusion here is this specific connector might not be ready for these sort of analytical queries just yet. Now, HubSpot does have a number of sample prompts that you can pull from their website, and I'll link the URL in the show notes. That might be a little bit more suited for this. This feature, and I think it's great to play around with.
(10:17):
GraceIf you have a HubSpot account and you're interested.
PeterYeah. You know, I'm going to give into my obsessive leanings and say, in this podcast right now that what we just explained there is huge. You want to start using an LLM to cut your hours to retry and stuff that would take you a long time to do and do it in a well-organized way. And that's certainly possible.
(10:41):
PeterBut at the same time, for where we're at right now, you have to consider the hours of questions you are asking. The Lim questions at the lab can give detailed and accurate and up to date answers for. And so the answer to that question is sometimes yes, sometimes no. But if you're a company that's looking to use Lims in your day to day work, to, to, you know, to make your work more efficient, etc., all that good stuff, you really have to do a sort of a 40,000ft analysis of, okay, what data can I count on when I'm asking the lab?
(11:25):
PeterAnd that's kind of that's kind of what we do right now when we're working with a client to build a custom AI solution, is we want to make sure that the data that's coming in is pristine. Is it that it's not going to be 100%, but it's it's going to be at least pretty much as accurate as you can reasonably expect.
(11:45):
GraceWhen is the right time to rely on native AI tools, like, for example, native HubSpot AI, other tools native to your marketing automation platform, or your CRM of choice, versus to seek custom integrations that bring specific AI capabilities into the platforms that you rely on as a business every day?
(12:09):
PeterWell, I think that, you know, there's a couple more than two categories here. One of them is AI intelligence based on certain products. So HubSpot being one, they they've really gone all in on AI. They acquired the company. Clear a bit and incorporate that into HubSpot for for different data enhancement functions and all AI functions. We use as an example, Clickup as project management tool.
(12:33):
PeterAnd Clickup now has its own AI, which is specific to Clickup. And and I would say in general, those integrations are pretty solid because they're just dealing with a very finite data set. And so and so you can rely on it. I think the general chat is it's helpful for many things and it's helpful for research. It's helpful for generating ideas, it's helpful for a general overview.
(12:59):
PeterAnd it's helpful for detailed overview. But you still have to really check your work. And one way to do that is to do it on different LMS.
GraceYeah. And you know, Peter, this gets me thinking about something we talk about all the time as HubSpot partners, which is the value of a single source of truth. In other words, having a centralized platform with all of your data, all of your marketing, sales, service activities, all talking to one another so that there aren't discrepancies. And I think that's shifting or being disrupted a little bit.
(13:36):
GraceBut so many new enticing AI tools coming out from so many different companies, how do we maintain AI, at least to the best of our abilities, the single source of truth right now, while also kind of experimenting with all this stuff?
PeterYeah, I think that I think a single source of truth in regards to ILM and in regards to getting, you know, single source of truth into HubSpot, probably the best strategy right now is to build an agent that that again, researches in multiple platforms and then has some way of reconciling the conflicting data that's coming in. And maybe, maybe that even looks like going out to additional platforms to really see what the consensus is.
(14:22):
PeterAnd sometimes in today's world, especially, the consensus isn't necessarily that accurate. But nevertheless, I think that's the best shot for getting a single source of truth as far as I type data into HubSpot. For anybody who's watching this podcast, I'd love to get your feedback about your opinions of how you want to consume information, especially around AI. You know, do you want the unvarnished truth or, you know, is it is it more fulfilling to find the bright, shiny object that may or may not be the real bright shiny object?
(14:58):
PeterSo I love I love your input on that. And, because it's it's a complex world out there.
GraceAnd also, I think there's a different value of knowing what the, the latest buzz is on a day to day basis versus letting something settle and then coming back to it. Thanks for listening, everybody. Make sure to give this podcast a positive review if you can.
(15:23):
PeterAnd subscribe, subscribe. Subscribing is a good thing.
GraceAbsolutely. And any other link or reference you might need or want will be down in the show notes as usual. Thanks for listening.
PeterSee you next week.
Popular Podcasts
New Heights with Jason & Travis Kelce
Football’s funniest family duo — Jason Kelce of the Philadelphia Eagles and Travis Kelce of the Kansas City Chiefs — team up to provide next-level access to life in the league as it unfolds. The two brothers and Super Bowl champions drop weekly insights about the weekly slate of games and share their INSIDE perspectives on trending NFL news and sports headlines. They also endlessly rag on each other as brothers do, chat the latest in pop culture and welcome some very popular and well-known friends to chat with them. Check out new episodes every Wednesday. Follow New Heights on the Wondery App, YouTube or wherever you get your podcasts. You can listen to new episodes early and ad-free, and get exclusive content on Wondery+. Join Wondery+ in the Wondery App, Apple Podcasts or Spotify. And join our new membership for a unique fan experience by going to the New Heights YouTube channel now!
The Breakfast Club
The World's Most Dangerous Morning Show, The Breakfast Club, With DJ Envy, Jess Hilarious, And Charlamagne Tha God!
Fudd Around And Find Out
UConn basketball star Azzi Fudd brings her championship swag to iHeart Women’s Sports with Fudd Around and Find Out, a weekly podcast that takes fans along for the ride as Azzi spends her final year of college trying to reclaim the National Championship and prepare to be a first round WNBA draft pick. Ever wonder what it’s like to be a world-class athlete in the public spotlight while still managing schoolwork, friendships and family time? It’s time to Fudd Around and Find Out!