Transcription
[Music] All right, welcome back to the Copywriters Podcast with your host, the world's greatest copywriting coach, David Garfinkle.
David, how you doing today?
Nathan: I'm good. How are you?
I'm good. Uh, we're doing brand new ways to use AI. And for the listeners that don't know, we record a couple of weeks in advance. And even if it was just a couple of days in advance, I don't know if we could fulfill on that promise, man.
That's right. I mean, this this is true as of February 20th, 2025. This may be old hat by the time you hear it, but it's the best we can do. And okay, let's—I think for a lot of people, some of these things are going to be new. So, oh, and you know what? I was talking with a client yesterday, today about this. He's deep into geeking out on AI, and most of the population, even most of the copywriters out there, are not as cutting-edge as some of the people in our crew are. So I think that it will still have a lot of relevance even if it's a month after we recorded it.
Yeah, and and if you're a super AI geek, you can look at this episode as an historical overview of AI in the last two weeks, weeks.
Okay, so let's go ahead and jump into it, man.
All right, so like we said today, we're going to talk about three new ways to use ChatGPT for copywriters and business owners, and another development from another company, which is a lot lower priced than one of the amazing things ChatGPT has come up with. Um, so we'll talk about that, and we'll talk about some other specific ways to use different AIs besides ChatGPT. But a lot of our focus today is on ChatGPT, and that's important because, to be clear, I wasn't all that hot on ChatGPT 18 months ago. But you know, as we were alluding to right in the beginning, in the AI world, 18 months is more like a couple of decades. Things have changed and definitely gotten better with ChatGPT, particularly in the last few weeks. The introduction of two new versions or apps, Operator and Deep Research. Now, at $200 a month, the combination of these two new services is 10 times as expensive as the old plain subscription ChatGPT. But if you need to get a lot more done quickly in certain areas of your copywriting business, or of your business that includes copywriting—if you're not a copywriter but a business owner that uses copywriting—these things could definitely be an asset for you. And we'll go into depth on ways to speed up key valuable tasks today, just after we go into depth on this: Copy is powerful; you're responsible for how you use what you hear on this podcast. And most of the time, common sense is all you need. But if you make extreme claims, and if you're writing copy for offers and highly regulated industries like health, finance, and business opportunity, you may want to get a legal review after you write and before you start using your copy. My larger clients do this all the time.
So Nathan, let's let's get started with the big ChatGPT breakthrough, Operator. And I'm getting mixed results with this one, and I was complaining to Nathan about it, who reminded me that Operator, quote, "is still an infant," close quote, which I suppose is technically true. Um, I'll tell you more about what it is and what it does in a minute, but the key thing to understand is that Operator is an Overlord AI; it can control other AIs and other apps, and you can get Operator to do tasks for you that in the past you had to do manually, supposedly much faster and more accurately than if you could do them yourself.
So here's how I tried it out: I had a bunch of transcripts, um, you know, digital transcripts I wanted summarized, and I wanted them summarized—each one into a title and five bullet points. Each original transcript was on a separate tab on a single Google Docs page. The process I wanted to do was only a few steps: open the doc, open one of the tabs, copy the full transcript, paste the transcript into another AI to boil it down to a title in five bullet points, then copy that output and paste that title into a new Google doc. After that, lather, rinse, repeat. So you'd end up with a series of summaries of these, you know, humongous one-hour, two-hour transcripts, and I wanted them in the same order the original transcripts were posted on the first Google doc with tabs. It's something that's fairly easy to do by hand, but to be honest, very tedious. Now, when I tried to get Operator to do it, the whole thing was extremely frustrating; it kept making mistakes and doing things differently than how I asked. And that was when Nathan told me I should consider Operator as of today an infant, which is good advice. Six months from now, which could be like six years or 10 years—years from now—Operator, you know, in in the AI world time, Operator will work much better and be doing things that we could scarcely imagine today.
But what about you, Nathan? Have you used—you you signed up for Operator too, right? What's been like for you?
Um, it's been hit and miss. I've had very frustrating workflows with it; uh, I've also had some pretty good experiences with it. So just real quick, um, basically what it does is you can have it open up windows, take actions, have something happen in one app, copy and paste that over to another app. Uh, the biggest problem that I've had with it so far is it not quite understanding what to do once it's opened other apps and um wanting to log in to things I'm not—I'm not super interested in letting it know how to log into things for me. Certain things have APIs though. Google Docs has an API that you can integrate with ChatGPT, so it can skip past the logging-in thing. Um, I don't want it logging into my Bitcoin account; I don't want it logging in or into my Bitcoin wallet; I don't want it logging into my bank account or uh my QuickBooks or anything like that just yet. I don't trust it. Uh, but if there's an API, like there is with Google—Google—I've seen it work pretty seamlessly. And I had a real-world example of how I've been using it, and then I have something to have you kind of roll around in your mind and get your takes on. So the one thing that I have used it for is I've got a bunch of different um AI copywriting bots for writing emails, and getting up each morning and figuring out, do I want to do a paino email? Do I want to do an aha moment email? Email for this client, for this product. Um, using Operator, I can have it go in, and I can say, "Hey, use this bot to write this type of email; use this bot to write this type of email; use this bot to write this type of email." And once you've written all three types of emails with two variations of each, go into Google Docs, use this document, copy and paste over what's there, and then wait for me to give you the okay on which email—email to use. Now, next step, once it integrates with an API for GetResponse or Constant Contact, the next step of this will be, "Okay, take this email, once I've picked which one, log into my email platform, pick the list, send it to this list." I haven't got it doing that part yet correctly, consistently again because of the login issue. Um, but it's a great way to wake up and have 80% of my work done for me, and I just have to pick, make a couple alterations and tell it, "Hey, this is the one I want to go with."
Yeah, it it kind of gives you time to wake up slower, doesn't it?
Without worrying about it.
Exactly. The thing that I was thinking was you have built some pretty cool copy editing and copy reviewing bots, and I was thinking, having like a daisy chain: "Okay, this is how I'm going to write my my uh hook and headline; this is how I'm going to write my pain points and my bullet points and my call to action; here's how I'm going to put them together," and have different bots that can do all of that, and then after that process, "Hey, here's the checklist to make sure that the copy is persuasive; to make sure that it flows nice from one section to the other; to make sure that I'm getting all of these key things." So we're not trying to cram all of that into one prompt; we've got prompts that do a specific thing, then a prompt that puts it all together, and then a prompt that runs through and has—as as close as you can come to having David Garfinkle actually check your copy for you—and have a daisy chain. And since all of that would be inside of ChatGPT, you'd never actually have to leave ChatGPT. Operator works great inside of its own software, and so that's kind of where I'm wanting to go next with this stuff.
Yeah, I I think, think that's a really good idea. The the one caveat, after you get the the critique or the review, um, don't ask ChatGPT or Claude or any other AI to fix it; that you need to do that, because there are all kinds of things that you can't count on it to know or to do. Um, however, I I think what you're talking about is extremely cool: rewrite, ready first draft, I mean totally ready, complete, within instructions and suggestions. Uh, I'd love that idea. Are you are you part of the way there now?
I am not. I was actually going to talk with you about, because you've built—I think two different—you've got like a grumpy old man one and then a grumpy old David one, and I was going to talk to you about that, because that's kind of like the final step in the chain, and um, but I I will say one thing that I have noticed though, especially with the email thing: I am going through a lot of tokens, creating a lot of stuff that never actually gets used because I say, "Hey, I want to wake up to six different variations; I'm only going to use one of them, possibly." And between using all the different bots for all the different variations and then using Operator to go through and take the variations and post them into a Google doc, it does burn through a lot of tokens. So you you've uh you've got to be conscious of the expense of using, besides just the $200 a month.
Yeah, but when you go through the tokens, what does that actually translate to in dollars and cents? I mean, you get a million points on Poe, for example, which is uh some translation of tokens, and it was only the first month when I couldn't stop using it every spare minute that I went through all million.
Yeah, that's true. The first month is always the most costly, and I think a lot of it is 'cause we're experimenting; we're trying new things, and as we get better and better with it, we know what works; we we know what not to try again, or if we're going to try it again, how to do it more effectively. But my question is, are all those tokens costing you real money?
No, but you hit your limits.
Yeah, but with um like Poe, you can go up to $50 a month. I mean, you know, what if if you were to hire a proofreader or a copy—or a junior copywriter, you're not going to get him for $50 a month.
Yeah, yeah. The the amount saved is absolutely worth it.
Yeah, I mean, I I don't think the tokens—it it's more of a competitive issue; it's like, yeah, I want to stay within these limits, but it's not—it's not a real economic problem, right?
Um, we're privileged, David; we we get paid a lot of money for what we do. For the average person out there, $200 a month is probably not—for sure—I—for sure. But the average person out there isn't going to be writing six emails every morning either.
There you go. Exactly. All right, we can we can bicker about this, but let's let's talk about some other things you can do with Operator, and I don't know how to do this yet; um, you would probably be better at this than I would, Nathan, but um, just some ideas: You could use it to segment your list better, figure out what people are buying, update it, and and send them emails around those topics of interest or with similar offers. And you could um probably do this much faster and more accurately than if you did it by hand. Um, you could coordinate posting on social media, scheduling, because who has time for that when there's actual real work to be done. And um, you could collect and analyze customer feedback—back—not just the stuff that comes into your customer service department or the random email or the random review; you could have it go out and search and and uh and and look for stuff, and and that kind of takes us to the second new feature for the $200 a month, and that is Deep Research.
So Deep Research, you know, it's kind of like what what's that button you have on on some really souped-up cars? A turbo button or something? It it—it is—it's bottom of the $200 a month version of ChatGPT, and it is—I'm all in on Deep Research. It is a quantum leap ahead of what research with ChatGPT used to be like, which I thought was pretty lame and often pretty inaccurate. Um, so I've used Deep Research on a number of projects. To give you an extreme idea, I have a mentoring client who's in a very high-stakes market, and he needed to come up with a new lead. Now, Deep Research didn't come up with a new lead; I came up with the idea, and that took several hours, but the idea was, "Let's look in in adjacent markets about raw material prices and how raw material prices increased rapidly when the sale of the product that raw materials were in increased." And took Deep Research eight minutes to give me a detailed 16-page report that was very well organized and very well sourced; plenty of links. And then I used a simpler version of ChatGPT to look at that 16-page report and find the spikes, and I found one raw material that increased 11 times—over 10,000%—in a one—time span, which would lead to—if you had a stock in a company that you know sold that raw material—the stock would go up. And so this made a great case for the lead in, and I know I'm being kind of sketchy—or not sketchy—but I'm leaving lots of holes in in what I'm saying; that's on purpose because it's private; it's—you—I don't want to share it; it's my client's proprietary info. But that's—that's one way to use Deep Research that really worked for me.
Now, here's the thing, and and this this gets back to what we were talking about in the beginning, how you know um a a week is like a year or more in in AI terms. A few days after ChatGPT or OpenAI introduced Deep Research, Perplexity, which was the go-to research AI before, introduced a competitor to Deep Research, and oddly enough, they also called theirs Deep Research. Um, the difference is, Perplexity Deep Research is free instead of $200 a month, and it's free on Perplexity for up to five requests a day, and you get unlimited requests for Perplexity Pro subscribers, which is $20 a month instead of 200. So I did a little test; I gave both Deep Research apps the same task: create a bio of one of our previous podcast guests based on a simple one-sentence prompt. Then I had Claude analyze and compare the two reports, and here are some things I found: The one from ChatGPT OpenAI was longer—2100 words compared to Perplexity's 1500 words. OpenAI was better organized and referenced; it was more businesslike and concise with references for every fact. Perplexity's was more conversational and did not include links at the point of the interesting facts, but included the list of links at the end. OpenAI's writing style of reports was more straightforward and factual, which is surprising to me because in the past I found OpenAI to be anything but straightforward and factual. But times change, and with AI, they're changing very fast. Perplexity report was more analytical and interpretive; it's conversational, not as informative and factual. So depending on how factual you want your research to be and how much backup you'll need—like to put in footnotes or to show a lawyer for an okay—OpenAI's Deep Research might be your best bet, but Perplexity Deep Research is very impressive for less demanding projects.
So I was a big fan of Perplexity until OpenAI came out with their Deep Research, which was, I think, intentionally positioned as a competitor for Perplexity, and then they kind of fired back. Um, I still like Perplexity; I also like AI—OpenAI Deep Research. I've had issues though with OpenAI Deep Research where it crapped out on me last week for like three days in a row; I couldn't access it; it just kept saying, "We'll let you know when Deep Research is available again," and that was pretty frustrating. Uh, but I will say this: I use it specifically for market research. I say, "Hey, go out and find all of the information that you can find about people who are dealing with this frustration and using this type of product, and bring me back the top 25 pain points that they're dealing with, and categorize them on ways that I could position my product as something that solves those problems, or uh put it into their words, summarize their words, how they feel about these." And Perplexity—one thing that I do like about it over uh OpenAI—you gave some good—some good uh reasons to prefer Deep Research from OpenAI—Perplexity has the ability to go into Reddit, which I try to stay away from Reddit myself, but people go mask off in Reddit; people—since there's kind of an anonymous aspect to it—people will tell you what they really think, and Perplexity has uh kind of like an exclusive permission to just go in there and scrape Reddit for all of the best uh feedback. And people talk about all kinds of products, all kinds of services, all kinds of frustrations. And so going into Reddit specifically with Perplexity, I've gotten some great insights on what people are thinking about inside of my market for different products and different services. And in that one instance, Perplexity has actually done a better job than Deep Research from OpenAI.
Yeah. Um, that's good. Have have you used Deep Research? Well, I guess you said you're used it for market research; use it for anything else?
Nope. I learned it specifically for that 'cause uh I'll I I'll take that back actually. Um, booking guests for one of my other clients; we're trying to—we're in a strategy right now to get that overlap of, "Hey, YouTube—our audience or our potential audience likes this influencer; we want to get our podcast in front of them; so help us find that type of influencer to get on the show." Um, Open—OpenAI actually did a better job at that of finding—find people who have posted YouTube videos—at least four in the last month—that have gotten at least this many views in the first 48 hours, that have at least this many comments, that have more positive comments than negative comments—giving it that type of task, uh the OpenAI Deep Research did better than the Perplexity one.
Okay, that's interesting. So when we were talking about this show, you point out you shouldn't put all your eggs in one basket, because you find that for specific tasks one AI is going to work notably better than another. Do you want to talk about that?
Yeah, so this has just been my experience: uh, ChatGPT for brainstorming, for um bouncing ideas off of something else, is wonderful; I love it. Um, for actually writing copy, it's not as good as Claude Sonnet. Uh, Claude Sonnet's copywriting—exact same prompts—I'll use the exact same prompts in both, and what I get out of Claude Sonnet is 98% of the way there, where ChatGPT is 95% of the way there. Um, so I like to have—I have a subscription to ChatGPT; I have a subscription to Poe; um, I have a subscription to uh Deep Research—the the more expanded pro version of ChatGPT; I've also got Perplexity; um, I've also got—for image generation—I use Midjourney for certain types of images, and then I use Runway AI Flux for different other images. And so um I think—and again, we're privileged; we have the—it's not a big deal for us to put 20, 40, even 200, $100 per subscription and have a bunch of different subscriptions. Um, maybe not everybody is in that position, but if you're using this for business, if you're using this for marketing, if you're using it for copywriting, it makes a lot of sense to have access to different tools for different um aspects of your creation and not have all your eggs in in one basket, like you said.
Yeah, I mean, one of the ways I like to look at it, because yes, uh clearly I I can afford these things, and I'm I'm trying to look at it, you know, as a frugal entrepreneur: You know, if I get this, will it save me time or make me money, or save me time where I can make more money elsewhere? And if it's—if it's not going to do that, um, doesn't matter—where I can afford it—I mean, I really can't afford it; it it'd be a waste of money, you know.
Well, and that's why I'm a big fan of Poe, is that's one subscription, and you get Claude, you get ChatGPT, you get Grok, you get all the different models—maybe not as good as they are if you buy a subscription just to that one model—um, but you got—you get access to image generation; you get access to all kinds of stuff. So I would say, if you had to just pick one to go with and you're just kind of dipping your toes in into all of this AI stuff, I would say Poe would be the one that I recommend.
Yeah, I mean, you you have to learn an extra step with Poe; there are like three or four—you don't just type in one prompt; there's—you can save a prompt; you can type in a prompt; you can type in a greeting message; you can type in a bio; there there's a lot to learn with Poe, but you can use it just the way you'd use a uh um ChatGPT or or Claude or or Grok on on X.
I I want to um go over something really basic, which I thought was awesome. I found this graphic um on Twitter X, and um I'm not sure who wrote it originally, or I'm not sure they even said who wrote it originally, but apparently like the the president—it was not Sam Bankman-Fried, but it was like a senior—the the chief of OpenAI had forwarded it on his blog or on some social media accounts, so I thought I'd read it. It's the anatomy of a prompt, particularly an 001 prompt for for ChatGPT, and 001 is the super reasoning um uh version. But this—this is—this is a great basic thing, which I really hadn't seen before, and I'm sure if I'd taken all the classes you've taken and watched all the YouTube videos you watched, Nathan, I would have seen it, but I I thought it was good. It said there's—you—you know, your your prompt should be four parts; it should—the top should be a goal; the second should be a return format; the third should be warnings; and the fourth should be—I love this—language context dump, because most people will basically write a one-sentence prompt, and I've been able to get very good results sometimes doing that, but if you don't—if if you want to really frame up what what you're asking the AI to do, it's good to at least know this. So at the top, goal—and I didn't write this, but I I I guess I could have if I were a hiker—"I want a list of the best medium-length hikes within two hours of San Francisco; each hike should provide a cool and unique adventure and be lesser known." So that's the goal; in other words, you're implying um that you want the AI to give you—well, you're saying it—"I want this list." And then the second part, right underneath it, the return format, is just one paragraph; it says, "For each hike, return the name of the hike as I'd find it on AllTrails," which I guess is a website for hikes, "then provide—"
The starting address of the hike, the ending address of the hike, distance, drive time, hike duration, and what makes it a cool hike—cool, cool, and unique. Adventure. Return the top three.
Okay, so that's that's the return format. This is this is what I want. This is how I want you to achieve the goal. The third one, um, is warnings. And this warning is: be careful to make sure that the name of the trail is correct, that it actually exists, and that the time is correct. Okay, so, um, you would think, of course, you know, any any responsible professional researcher would do that, but AI needs to be reminded.
And then here's context dump. This is this is the fourth part, and this is so important because—and by the way, in in the prompt it doesn't say goal or return format or warnings—this this is just like labeling of these different sections. So the fourth one: for context, my girlfriend and I hike a ton. We've done pretty much all the local SF hikes, whether it's Presidio or Golden Gate Park. We definitely want to get out of town. We did Mount Tam pretty recently, the whole thing from beginning to the end to Stinson. It was really long, and we are definitely in the mood for something different this weekend. Ocean views would be nice. We love delicious food. One thing I loved about the Mount Tam hike is that it ends with the celebration of arriving in town for breakfast. The old missile silos and stuff near Discovery Point is cool, but I've just done that hike probably 20 times at this point. We won't be seeing each other for a few weeks; she has to stay in LA for work, so the uniqueness here really counts.
Okay, maybe I never would send a prompt quite like that, but I I think it it gives, um, a good example. U and I, I want you to realize this requires a lot of thinking on your part that people aren't used to doing. You need, you need, when it talks about the return format, you need to actually figure out what you want it to look like and what you wanted to include. Most people think, well, it's artificial intelligence, that means it's intelligent, that means it'll figure it out for me. No, it won't. It isn't artificial clairvoyant. So we should do a full episode on this, this what you're talking about. It's called prompt adherence, and we, I'm just going to give a real quick 50,000-foot view, the way that me and you usually do this is we give it the 50,000-foot view of what we want: Hey, I need some good hooks written for this offer. Then we give it the uh the format that we want it back in. Then we give it: Make sure you follow these rules for writing and avoid these words. And then we give it some examples, the way that it should be outputted should look something like this. So a slight, very variation on what he was talking about, specifically for copywriting, but yeah, having this, um, four-step process to writing your prompts makes all the difference, and you get much more consistent output. And so I would just say, if you're listening to this and you want to learn a little bit more about it, go look up the word prompt adherence, and you'll dive down a rabbit hole that'll make your copy, your copy AI outputs way better.
Yeah, um, let's let's do something about that uh in a future episode. And I I want to end with a thought for the day. This is from someone I've never heard of before, but I saw him yesterday, BAGI, on um Twitter X, and he says: "Programming isn't going away; prompting is programming. And the better you are at articulating what you want in clear written English, the better your results." So that kind of sums up what we were just talking about in a couple sentences.
Absolutely, David. Fantastic episode, man. I'm uh I remember when I first started getting into AI copywriting and you were kind of like, "I'm not sure I want to do episodes on this," but now that you've dived into it, you are just delivering gold, man. So I'm I'm I'm glad that you decided to jump into the deep end when it comes to AI copywriting.
Well, me too. But to be fair to me, that was the Stone Age when when we actually talked about um AI the first time, and now we're in the 21st century. Who 18 months ago was in the Stone Age? That's crazy. All right, if if you enjoyed this episode, you want more, or you don't want to miss the upcoming episode where we dive a little bit deeper into this, make sure you're subscribed to the podcast. Catch all the episodes over at copywriterspodcast.com, and until next time, we will catch you later. Catch you later.