Loading summary
Transcript107 lines
- [00:01]
A
Okay, the difference between your AI results and all of the experts out there is actually in the prompting. Well, we are going to turn you into an expert AI prompter in 30 minutes or less. But better yet, you won't even have to do the prompting because we're going to give you an AI assistant to do it for you so you can have expert level promptings without ever having to write something yourself. All of that on today's show. Marketing against the grain.
- [00:26]
B
We're back to the pod in just a minute, but first I want to tell you about something very exciting happening at HubSpot. It's no secret in business that the faster you can pivot, the more successful you'll be. And with how fast AI is changing everything we do, you need tools that actually deliver for you in record time. Enter HubSpot Spring Spotlight, where we just dropped hundreds of updates that are completely changing the game. We're talking breeze agents that use AI to do in minutes what used to take days. Workspaces that bring everything you need into one view and marketing Hub features that use AI to find your perfect audience. What used to take weeks now happens in seconds, and that changes everything. This isn't just about moving fast. It's about moving fast in the right direction. Visit HubSpot.comspotlight and transform how your business grows starting today.
- [01:32]
A
Okay, so we did an episode where we showed you all how to create one of the best landing pages in the planet using a range of different tools. We were using a range of different prompting techniques. That page is going live on HubSpot with a few tweaks. So this is how good some of these prompts work. And the entire comment section was filled with two asks. Integrate some of this stuff in the HubSpot, which I think is a good idea.
- [01:56]
B
It's all coming more on that soon.
- [01:58]
A
And then give us prompts. Templates.
- [02:00]
B
People want the prompts.
- [02:01]
A
Just give us the friggin prompts. So let's dive straight into like one prompt that I think maybe our listeners have seen. I think it's one of the better templates. Then we're going to give away some pretty interesting things for our subscribers to YouTube exclusively or for our subscribers to the RSS exclusively. And so this is a pretty well shared prompt that I'm showing for those not watching along on YouTube. It's called the autonomy of an O1 prompt. And this is actually from Greg Brockman, the co founder of OpenAI. And he was talking about the fact that he still sees a lot of people cannot prompt and that is why their results are much worse than other people for using these kind of LLM models. And so what he did is he provided this really easy to understand prompt and then broke it apart into its core parts. And I thought we could go through this. And then I want to show a quick hack that people can use to create their own prompt engineer from this template that not many people do. And so what this has is a couple of core components, right. If you want to construct a really good prompt. And this says for the O1 models, but I think this is just good prompt hygiene in general is that you want to have a clear goal. And so in this case, he has a clear goal around hikes in San Francisco. I will say anytime I see hikes and Americans say hikes in Europe, this means like going up mountains. I think in America this means going to the shops on a flat surface. It's not the same thing.
- [03:26]
B
I would say a hike in America is walking on. Not a sidewalk, not a sidewalk. Walking on a sidewalk. Like a trail of some type.
- [03:33]
A
Yeah, a trail, but it's not mountain.
- [03:35]
B
Climbing in some cases is. In some cases it isn't. Right. It's very wide definition of hiking in America. America.
- [03:40]
A
So he has a clear goal. You state your goal, you really want to understand what your outcome is. He has a return format. Now I'm going to get into a much more detailed version of this and then I'm going to get into a pretty awesome giveaway for a five page prompt that our listeners can just go take and start doing some cool stuff with. So this is a return format. What is the format? I want this to be returned in warnings. So what is the warnings that he provided here was like things like be careful to make sure the name of the trail is correct, that it actually exists. I think these are things you want to point it towards and that the time is correct. And so, you know, examples would be, if you are doing any kind of prompting to research something, you would say, hey, make sure the name of that company is real, make sure they have actual employees because you don't want it to hallucinate or whatever kind of warnings you want to create for the model.
- [04:31]
B
Kieran. I think this is a simple example of like AI might find something in one source that may not be that accurate. Right. And what it's basically telling the AI is like, hey, kind of check multiple times to make sure this thing's real.
- [04:43]
A
Right, Right, exactly. One of the things I do internally, when I'm prompting, I use this prompt all the Time. But again, I'm going to give you an easy way to use this so you never really have to remember the format. And so even though I'm giving you the format here, I'm going to show you a way where you never have to actually remember it. But one of the things I do, and I use this internally for internal projects and I give it a ton of context, which we're going to go into now. I'll always say, like, site the documentation and the exact place where you got the information to make the statement. So when I get the reply from the LLM, it always cites all of its sources and reasons why it made the assumptions or kind of give the output that it did. So I can see exactly it's getting it from documents that actually exist. And then the most important thing that people still don't get, I don't think.
- [05:28]
B
No, they definitely don't.
- [05:29]
A
People think context is like a sentence. So even here, the context that Greg gave in the example, it's like a couple of paragraphs, right? We what we did on the landing pages, if you recall in our landing page episode, and if you haven't watched that, you should go watch that. The context we provided to create the world's best landing page was actually a 30 page style guide, right? And so this is the beauty of these LLMs is that they have huge amounts of ability to, for you to add context. You can add a ton of things at a context window and really tailor those results. And so kicking off this episode, if you are stuck with prompting in general, this is a great first place to start, right? Do I have a goal? Have I stated my return format? Have I given some sort of warnings to say like these are the things I really need you to do and have I given it good amounts of context? And I think that's a good starting point for this episode. That will make people much, much better at prompting if you haven't been doing this to date. So we don't remember all of that because we're lazy. So one of the things I do is basically I'm constantly taking prompts and then turning them into templates and then creating custom GPTs to replicate that template for me. So what do I mean by this? So I go to my GPTs, I go in here and we're going to get into a couple of more advanced prompts in this episode. I really love this one that I've created, which is taking a deep research prompt and creating a YouTube video.
- [06:50]
B
You're going to get asked for all of These, just so we know the viewers are going to be hitting you up.
- [06:54]
A
Because I love our viewers, I am going to give away the deep research to YouTube video, which I think is actually incredible. I'm going to show you something even better. Taking a marketing tactic and being able to templatize it into something where you can have an AI replicate that marketing tactic for you. I am not yet going to give away this prompt, but I'm going to show you the results. What we will do in the future, if you're a listener, I think we're going to figure out how to give our listeners some exclusive stuff at some point. Yeah.
- [07:22]
B
So you want to hit subscribe?
- [07:23]
A
Because you want to hit subscribe.
- [07:24]
B
We're going to try to do some subscriber exclusives, but this YouTube one is dope.
- [07:28]
A
I was like, maybe I should keep this one for myself because it's got a style guide for a YouTube video that is actually insane. All right, so what I've done here is I'll go into this prompt engineer 01. So this is my prompt engineer for creating 01 prompts. And so basically what it does is it provides a little bit of context for what I want, and then it tells it to follow the exact template. And so anytime I need a prompt, what I find really easy is I can come in here and I can say, okay, I'm looking to create an incredible Twitter thread on the latest Google. You don't have to worry about spelling boy announcements, which is great for me because I don't have grammar in here. Let's go, announcements. Create me a prompt to research those announcements that marketers would love and create. Create an actionable thread on how they can use the AI to be way more productive. Whatever. Obviously you would have an actual real conversation here. Go back and forth, and then you say, cool. That's a cool conversation. Create me a prompt for this. Let's create a prompt to create a strategic memo on how AI is going to disrupt the. To disrupt. What is it going to disrupt? Give me something that's going to disrupt. They go into software. Yeah. It's going to disrupt the industry. Okay. I'm going to provide both internal and external data to analyze as part of the memo.
- [09:01]
B
All right, so that's two sentences, very basic, very straightforward. And it's taking the template that you just showed everybody, you know, with the goal, the format, the warnings, and it is giving you a full prompt. It's making sure you follow that template every single time.
- [09:18]
A
The outcome is a strategic memo that pitches how we need to reformat our sales team. The return format should be a two page memo detailing the core changes. You should only make it applicable to software sales. Okay. Obviously you'd have provided a ton of context. All right, so it's gonna do it now. And so it is a prompt engineer. It will make sure that you give it all of the things it needs, which I think is a good actual checkpoint for people. I use this all the time. I use this to create all of my prompts for O models. But I will provide it with a lot of information upfront and then it will craft a pretty good prompt. And you know what's actually pretty interesting? Can you make this prompt a lot more detailed? I want it to be very advanced so you can actually play around and it will improve upon that. Now the other thing is, one of the things I would get into the habit, to be honest with you in doing is I believe this about all AI is I would create a first version yourself to get used to prompting. I don't think you want to rely solely on AIs, but if you are like a novice at prompting, doing what I just showed you would take you up to like above good pretty rapidly.
- [10:37]
B
Definitely above average.
- [10:38]
A
And then if you actually play around with it and actually ask it to continue to edit it. Okay, what we've done so far is we've taken just a single template, we've created a custom GPT to be your prompt engineer to replicate that template whenever you want. But the interesting thing is if you go back and forth between models, you did a whole video on 2.5 Gemini. And so I've realized that actually you can go between the models and say, here is a detailed prompt for an OON model.
- [11:06]
B
This is what I do. This is smart.
- [11:08]
A
How can I improve this? So the output and even if you.
- [11:11]
B
Don'T say a one, you can just say advanced reasoning model if you want to be like kind of inclusive for any of the reasoning models, which is a good tip, I think.
- [11:19]
A
And then you can put in your prompt and then it will give you a much more detailed version of that. Now what I would do is I would go through the prompt and I would say, cool, I like these parts, I don't like these parts. I would want these kind of improvements. But. But if you flip flop back and forth in the models, they can edit on each other's work. So that's our first kind of just standard 01 prompt.
- [11:39]
B
Wow. Way more detail. It's giving you 2.5 Pro is really good.
- [11:43]
A
Yeah, it's really good. All of the other prompts that I'm going to show now for deep research are actually done this way. It's a combination of two models. So let's just quickly show the same sort of format that, that kind of template that Greg created. But we'll do it for deep research. Now what people are using deep research for mainly is to do a research report, right? That's basically what it's being used for, is analysis. Okay, so there's a couple of things. Again, if you're following along on YouTube, this is going to be really easy. But if you want to really maximize your results on deep research, one of the things is context really, really matters in deep research. And so an example would be whenever I am doing something for HubSpot, I give it full details on my role, the company size, the industry that we are in. I give it information about who our customers are. I give it information about our tech stack, if that's applicable to what I'm doing. I give it a lot of context about what are the key goals and challenges. And so what I found is if you set the context about who you are, you know, the company you're in, all of the things that matter to the research you're about to do, the results are way, way better. And so the blurb here is this is obviously very simplistic. It would be more detailed than this. But I'm a content manager in a 200 person SaaS company. We use HubSpot and Jasper all day. We're evaluating new tools to scale our content and grow our audience faster. So you set context, you give a clear assignment, you want to clearly state what research you want done and what kind of answer would be most useful. You want to go through what your objectives and key metrics are. And I'll go through this because this one we can just give away. It's. It's out there already. I give it away already. Scope and priorities. You want to set your scope and priorities. So after we set the objectives and key metrics, we're then like, okay, well this is the scope of the research, this is what I'm trying to focus on. And then deliverable format. And what I want to spend the rest of this episode riffing on is this part matters because the deliverable format is like pretty interesting. So you can create an internal report, you can create a template. What I'm going to show you is go from like the research into a YouTube episode. And so the five core components of a deep research prompt is like Set the context, make sure you have clarity on the assignment, and it's really clear to the AI what you're trying to do. Set the objectives and key metrics, set the scope and priorities, and then set the deliverable format.
- [14:12]
B
Kieran, while you're transitioning here, what I want to really help people understand is that that last step, the deliverable format's so important. Cause I think what most people I talk to do is like, they do deep research and just like, hey, I want a bunch of information on this thing and they get it and then they go and then they do something with it and they often just put it in an LLM with a different prompt. And you're way better off to just tell the deep research just like, hey, I need this formatted to go with this prompt that I'm planning to use in this LLM. And it will package everything up so that you'll actually get a far better result from your prompt, which is, I think, what you're going to show. But that's why that's so important.
- [14:49]
A
Yeah, yeah. And so what we have here is another GPT trained to reproduce that prompt, right?
- [14:57]
B
Yep.
- [14:58]
A
And so this is again, a quite simplistic way. So I'm going to basically go, I want to create a report on what AI SDR tools can improve my sales outreach, open rates. Okay. So we've given it a little bit information. Obviously you would give it a lot more. Again, you're going to get the guide here. So all you need to do is create your own custom GPT, give it the prompt template that we're going to give you at any time. You want a deep research prompt to create a detailed report on. Anything really. You could do that. So we've given it some contacts, we've said what kind of what industry we're in, who we're targeting. These are like a pretty basic here. So we've given it some of this. And so it should actually create a pretty detailed deep research prompt here to create your entire AI SDR tool.
- [15:45]
B
You know, one of the things I love, I love all the emojis.
- [15:47]
A
Yeah, I know. It's like really bigger emojis.
- [15:48]
B
The emojis make me happy. I don't know why. It's just like, I know it doesn't do anything, but, man, I just, I like the emojis.
- [15:55]
A
It brings a bit of color to.
- [15:56]
B
It, makes it feel like less just like staring at black and white.
- [16:00]
A
So it's gonna craft you the whole, wow.
- [16:03]
B
The. This is a very, very deep, deep.
- [16:05]
A
Research Prompt again, like the size of the prompt, right? And so again you can go from beginner level deep research prompter to like pretty expert level pretty quickly by creating these custom GPTs that are trained on a prompt that really works. And so you can just basically copy and paste this and kick off your report and it will bring you back a pretty detailed report on AI tools that can really help you increase your cold email open rates by 10%. We've given it some scope and priorities. Has to be a HubSpot integration. There's a pretty cool thing here, right? It includes case studies or performance benchmarks where available. It will do a lot of cool stuff. Honestly, the level of report you can get in most companies, you can probably just like use this and send it internally and no one would know is not from you.
- [16:50]
B
Totally.
- [16:57]
A
The concept here is going from research to output, right? You're doing a detailed research on something and it's creating something for you.
- [17:05]
B
So what he's doing is he's taking that advanced prompt that was very dope as we just saw. And he's just putting it in to regular 4.0 deep research. Basic model, Basic deep research turned on a deep research. Now, just so everybody knows, it's always going to ask clarifying questions. It's going to ask clarifying questions. I think it's doing that because they don't want you to keep running deep research when you didn't get a couple of the things you needed the first time. So I think they're trying to reduce the number of runs that you're doing, but it's always going to ask you a handful of quick clarifying questions. He's answered those and now bam, it's going to run the deep test.
- [17:40]
A
Off it goes. All right, let's get into a pretty cool prompt here. Anytime you want to just create a report on anything, right, it's like having a McKinsey based consultant. You can just say, okay, I want to report it, whatever it may be. You can go into this GPT assistant, you can ask for the report, it will ask you some clarifying questions and boom, away it goes. The next one I want to show is this YouTube one.
- [18:01]
B
I haven't seen this one yet by the way.
- [18:03]
A
I'm actually just going to go through the prompt in detail. And so again, you would definitely be better to be following along in YouTube.
- [18:10]
B
It's definitely a YouTube first episode, folks.
- [18:13]
A
I'll make it work for all of the people in rss. All right, so that was her report, but I suspect from Marketers like what are marketing type things we can do? And so this one here, it does the same sort of thing, right? It says context. And so again it's saying you're a research analysis and narrative strategist, an audience specialist and a creative video screenwriter. So you're expert at a lot of things.
- [18:33]
B
I was going to say you're like five different people in this case, right?
- [18:37]
A
Your mission is to perform a deep multifaceted research on the topic provided and then transform that research into a world class YouTube video outline. It's designed for viewer engagement, retention, shareability and impact. I wouldn't actually put retention in there, if I'm being honest.
- [18:51]
B
Kieran doesn't care about retention.
- [18:52]
A
No, I care, but that's in viewer engagement.
- [18:54]
B
I'm joking, I'm joking.
- [18:55]
A
And then it says, okay, well like you'll provide a topic, you'll provide an audience, the viewer will provide a video length. And then once it's done that, it will go and research that topic in detail. It will do the research scope, it will look at the core narrative, it will define key takeaways, it will pinpoint you really engaging elements. So this is like really cool, right? It goes and looks to see, okay, well, what are surprising statistics, what are counter intuitive findings, what are relatable analogies, what are ethical dilemmas? Look for quotes. It's got some really cool things. It's trying to find all of the things in this topic that would actually make it a good YouTube video. And then it says, okay, I have those things. Now I'm going to convert it into this outline. And so this outline is a very detailed outline.
- [19:41]
B
How'd you get this outline, Kieran?
- [19:43]
A
This was done using the thing I usually do which is taking videos that have worked really well, taken the transcripts, adding them into context, and then working with AI to distill it into a style. So this is like a YouTube kind of style outline.
- [20:00]
B
Did you do that in ChatGPT or did you do that in NetBookLM?
- [20:02]
A
I did it in Gemini and ChatGPT. I've been switching between both models. Just iterating.
- [20:07]
B
I have been too.
- [20:08]
A
It's super cool. Actually.
- [20:09]
B
It works way better between two or three models. This is why you kind of need multiple seats. If you're just a one Frontier model shop, you kind of need more than one.
- [20:17]
A
Yeah. I haven't been using Claude as much, which I feel bad about, but I just have.
- [20:22]
B
You were the number one Claude, super fan of Cloud.
- [20:25]
A
I know, I was the Claude person. I don't know what's happened to me.
- [20:28]
B
The pace of innovation there is a little slow right now.
- [20:30]
A
Yeah. I think Google have come out with some pretty incredible things. ChatGPT continues to come out.
- [20:35]
B
Google has come up with so much stuff recently that nobody knows about. We're just going to do a series of shows coming up soon with some of the cool stuff like that. The freaking AI in Google sheets is incredible.
- [20:46]
A
Now they have so many releases. We'd have to do like one show per some of the cool releases. The agentic workspace.
- [20:52]
B
Yeah, we're going to do an overview and then we're going to a couple of the big ones. We'll do just a quick run through and kind of marketing use case for.
- [20:58]
A
So it's broken into. I'll just go through the core headlines. It breaks into a hook. It breaks into the intro and it says you have to set up the premise and the promise. Clearly state the video. Central topic in the question goes through like how you create engagement right up front. It sets the stage, context is key, unpacks the research, the so what impact and implications, units and critical perspectives. So it goes through a whole litany of things, right?
- [21:22]
B
Yep.
- [21:22]
A
And then it has all this kind of phase through discoverability and polish elements. It basically goes through compelling title suggestions, the video description, keywords and tags even goes into the sound so good. And production and source. And then it goes into sources. Product producer Darren's checking this out so that one's running. Okay, so we can do this, right?
- [21:43]
B
Let's do it, man.
- [21:43]
A
We actually want to create a video for YouTube around Google.
- [21:47]
B
Let's do it.
- [21:48]
A
So I want a prompt to create a YouTube video for Google's latest AI launches from next 25 and that will ask me those three questions. Okay, so I'm in my custom GPT. That custom GPT again is trained to reproduce prompts in the style that I showed you. We go in and we say this is a real video that Kip and I want to do. So let's go. Okay, cool. I've got the topic, which is Google's latest AI launches. Who's the target audience? So we're going to say marketers, startup founders, ideal video land. We're going to say 30 minutes. All right, we're going to put that in. Okay. So now it's going to give me the detail prompt and we're actually going to run this one in Gemini Deep Research. If you see the time I mentioned 30 minutes. So you can see it's given the exact time, gives the hook. It gives the Exact time, length of the intro. It gives the exact time to set the context. This is actually really good. Start using this. It gives the exact time to do all of the core insights. It does a really cool thing where you have five minutes at the end where you talk about the impact and implications. Like, why does this matter? You provide the nuance and critical perspective in that minute 25 to 27. And then you have a call to action, an engagement loop. And then you end screen. So it's given like an exact breakdown mapped to the 30 minutes, which again is pretty incredible. And so let's run this prompt in Google and Kip and I can probably just use this and do a show and see if it's good, right? Like we can actually use this for our video shows.
- [23:30]
B
Would love that.
- [23:31]
A
All right, so this is what it's produced. It started with a pretty good little hook here. 50,000 images a month. Eight weeks of work done in eight hours. Google's Next 25 conference unleashed a tsunami of AI innovations that promise to supercharge marketing. So again, it's for marketers. Did Google just hand marketers a creative superpower or are we looking at the beginning of the end for traditional marketing as we know it?
- [23:50]
B
Ooh, dramatic. I love the sim.
- [23:53]
A
That's a pretty great opener, right?
- [23:55]
B
It is.
- [23:55]
A
Okay, so that's the hook, that's the opener. Then it goes into premise and promise.
- [24:00]
B
Yep.
- [24:00]
A
A tsunami of AI hits marketing and then it gives an overview. I mean, this is pretty incredible, right?
- [24:06]
B
It's really good.
- [24:06]
A
So like we're looking at a pretty detailed overview. Hooked yet? Buckle up marketers. Let's dive into the game change in highlights and what they mean for you. Pull out some quotes. AI will drive a transformation of advertising in 2025 even more significant than the mobile revolution. It's pulled it out from Twitter, actually.
- [24:22]
B
Mm, interesting.
- [24:23]
A
Did you know it could pull stuff from Twitter?
- [24:25]
B
I didn't know. That's actually pretty wild.
- [24:26]
A
It's pulling quotes from all over the place.
- [24:28]
B
News, search, Twitter.
- [24:31]
A
Yeah, look, this is a show, right?
- [24:32]
B
It's really good. We probably will make a show out of it.
- [24:35]
A
It's even got, see here. It's included things like sound, keyboard, typing as user to taste a task. It's thought of like all of the ways to integrate sound into this episode. It's got tons of citations, tons of citations for like quotes we can use. I'm really curious about the Twitter quote. The only thing I would want to play around with is I think at what I would do. Remember when we Give it warnings to say explicitly what we don't want.
- [25:01]
B
Yeah.
- [25:01]
A
The thing I would definitely ask it to ensure it does is map to this outline.
- [25:06]
B
Map to the outline and map a little bit more to, like, the marketing use cases. Right.
- [25:11]
A
Again, don't think that this is. Oh, cool. I've got a prompt now that I can create any type of outline for a pretty incredible YouTube video. But don't just stop there. You can actually use that deep research and just swap out the YouTube outline for any outline, a blog post outline, and use that outline, a Twitter thread outline. The thing to learn here is going from a deep research analysis to deep research, actually combining that with an output template. Any output you want.
- [25:43]
B
I was going to say this has been more than one show's worth of content even. We might even have to do a part two here.
- [25:48]
A
All right, I won't show. The last one is just to get people's thoughts racing here. This one here takes any marketing tactic and creates a detailed template execution guide for an AI assistant to replicate that. And what it basically does is distill out exactly how to recreate this and then provide a bunch of prompts here for the AI to use. And the AI can then replicate any marketing tactic you give it. And I think this is also a pretty great use of deep research because people are just thinking of deep research as a thing to create a report analysis. I think of it as a way to actually get very specific on things you can execute on, like, take action on. And I think that's one of the takeaways from this episode. So if we run through what we give you really quickly, we give you this O1 prompt, we put that.01 prompt into a prompt engineer, so you can recreate that anytime you want. We give you a much more detailed deep research prompt. Then we put that deep research prompt into your own custom GPT so you can replicate that whenever you want. Then we give you a version that is actually something you can build upon. I just happened to do the YouTube video. So this kind of deep research prompt that actually creates a very detailed YouTube video, and we showcase that by running it for this Google AI's next 25 launch, that we're going to actually use the results from this deep research prompt and we will create a show on.
- [27:11]
B
Yeah. So, Kieran, when I think about today's show, in like, literally like 30 minutes, we taught somebody how to be an expert prompt engineer. We gave them all the tips, all the tricks to become an expert prompt engineer. And we gave this bonus of you should never do dumb AI anymore. You can always have deep research to provide really powerful context to any problem you're trying to solve. With AI, that is one of the power hacks that you and I use. It is very transformative. Nobody does it. It's something you're going to continue to see on the show and something we've talked about in the past. But the use of custom GPTs, the actual anatomy of amazing prompt, and how to use those prompts to get really powerful Deep Research is a game changer for how everybody's going to be using AI.
- [27:57]
A
Yes, go forth and prompt go forth in prom.
- [28:00]
B
I love that. We'll see everybody real soon on marketing.
- [28:03]
A
It's the greatest.