8 Ways Marketers Are Using Generative AI Right Now

The person who wrote this is human. But how can you be sure?

We live in the age of generative AI. That means computers have entered into the realms that we previously believed to be unique to the human experience.

What does it mean for the writer, now that ChatGPT can write long blocks of beautiful prose in mere seconds? And what of the artist, now that Midjourney can create impressionistic paintings in the style of Van Gogh in less time than it takes to microwave a bag of popcorn?

These are big philosophical questions, all of which are begging to be answered by the great thinkers of our time. But since generative AI has already entered our workplace, we can’t wait for philosophers to tell us what it means to be a person. We can’t wait for managers, who don’t have time to research this stuff, to tell us what to do.

We have to figure it out on our own.

So today, I’d like to share with you some ways my own marketing agency is currently using generative AI in day-to-day life.

But first: what exactly is Generative AI anyway?

If you’ve been hiding under a rock for the last year, and you’ve made the poor decision to emerge into the real world, then you should probably know what generative AI is and why everyone’s talking about it.

Generative AI is a type of artificial intelligence that focuses on creating new and original content, rather than just recognizing and categorizing existing content. This can include generating text, images, music, and even video. It involves training machine learning models on large datasets and then using these models to generate new content based on the patterns and relationships they have learned.

The results can be highly creative, sometimes even surpassing human-generated content in terms of originality and quality. With the rise of generative AI, we are seeing the blurring of lines between human and machine-generated content, and the possibilities for what can be created are truly endless. So get ready to sit back and enjoy the show as AI takes the reins of creativity!

ChatGPT, a generative AI describing generative AI

The definition given to us by ChatGPT is a good one, even though it appears to be threatening my job. It’s true – computers are dipping their toe into the waters of creativity and doing a pretty good job at it.

We are actively deploying generative AI tools in our client work at this time. Even more interestingly, our clients are aware of what we’re doing and fully support our endeavors. Generative AI has cut about 15-20% of the time sucks out of our work, allowing us to create higher quality deliverables for our clients.

I know that a “15-20% reduction in wasted time” doesn’t sound like the breathless press releases of Big Tech, but don’t scoff at that figure. When you’re working 40+ hour workweeks, this is a life-changing efficiency gain.

A few common generative AI tools

If you’re reading about generative AI tools, especially in the marketing agency world, you’ll probably come across some of the following names.

• ChatGPT: A language model capable of creating long blocks of human-like text based on prompts provided by the user. It’s a stunningly versatile tool.
• Midjourney, Stable Diffusion, and DALL-E: AI platforms that generate art using machine learning algorithms. You can read about it in-depth in this article.
• Jasper: Similar to ChatGPT, this is a language model that can create human-sounding text. However, whereas ChatGPT is stylized as a chatbot, Jasper AI is explicitly for copywriting and long-form content generation, even allowing users to create blog posts from outlines.
• Descript: An AI-powered platform that allows users to edit audio and video files as if they were a Google Doc. It can be used for tasks such as transcribing speech, dubbing audio, and editing video content.

Indeed, we’ve used all of these, with the exception of Stable Diffusion, in our own work. All of these AI tools have different use cases, strengths, and weaknesses. But what they all have in common is that they are absolutely fascinating to use.

This is best demonstrated when talking about specific use cases in the wild, so let’s talk about some of those.

1. Summarizing complicated concepts

“Joe Client just sent me a 15-page email.”

Not to worry. One use of AI that has personally saved me an enormous amount of time is copying and pasting long stretches of text into ChatGPT and asking for a summary. Long emails, web pages, news articles – you name it.

But ChatGPT’s ability to summarize complicated concepts extends beyond that. I have asked it to explain why Rollercoaster Tycoon was coded in assembly language, complete with examples of programming language code. It has explained to me the similarities and differences between Stoicism and Taoism. I have even used it in the kitchen to help me know which spices pair with beets and carrots.

This sounds like a bit of a parlor trick, but think about the business applications in marketing. It’s not uncommon for copywriters, artists, content writers, and marketing strategists to land clients in industries they know nothing about. Most marketing skills are transferrable, but you still have to have a basic understanding of how the industry works before you can do good work.

You could go to Google, scroll past the ads, then start reading everything you can find. Sure, buy all the books and watch all the YouTube videos. Get lost on Whatever Tok, figuring things out. Going down a rabbit hole is a surefire way to learn.

But if you have specific, targeted questions, the search engine can’t handle those. ChatGPT can. (Just don’t forget to fact-check it!)

2. Proofreading our writing and rephrasing complicated text

This is a pretty straightforward use case. Just about any solid large language model, such as ChatGPT or Jasper, is qualified to proofread writing. In general, it has a better track record than spell check or even tools like Grammarly. While it’s by no means perfect, it still dramatically cuts down on the time-consuming task of proofreading.

Some language models are also quite good at rephrasing complicated text. For example, I’m working with a client that is making a really complex board game series called Vrahode. It’s a brand-new fantasy world, and describing it poses a challenge for even the best copywriter.

Often, my client and I do our best to get the basic idea across. Then I take it to tools like ChatGPT and ask for a rephrase. You have to scrutinize the output, but it’s still great for saving the all-too-limited brainpower it takes to come up with stuff like this.

3. Generating ideas and providing inspiration

Traditionally, when you think of brainstorming, you imagine people gathered in a conference room at a whiteboard, frantically writing down the first things that come to mind.

Here’s the trouble with that. People are afraid to speak up. Sometimes the wellspring of ideas runs dry in everybody’s heads. And even if you have a group of minds fertile with ideas, you are still limited to the speed of typing or speaking to document them.

However, if you ask ChatGPT for a list of pros and cons, you’ll get a response back in seconds. Again – not every response rings true, but it beats the hell out of trying to come up with this stuff on your own.

It gets weirder, though. I’ve heard that some folks are starting to use Midjourney, DALL-E, and Stable Diffusion to do the same thing with imagery. They can rapidly prototype sketches of art to get a sense for the kind of style they want on their graphic design. Movie studios can use AI art to create storyboards.

Think of the applications! Instead of having a laborious back-and-forth, artists and their clients can turn to the AI oracle for some direction, and instead of the client giving vague direction like “make it pop,” they can say “I like this image for X, Y, and Z reasons, but it falls short on qualities A and B. Can you take this and make it better?”

4. Copywriting

I keep talking about ChatGPT, but that’s because it’s the current king of generative AI. One of the reasons I believe this is because it is a pretty decent copywriter.

I’ve already discussed its ability to rephrase and brainstorm ideas. When you’re trying to copywrite for a brand, this is already a massive time-saver. But if you’re truly stuck, you can ask for a bunch of options for email subject lines and ad copy, headlines and body text. You almost always have to adjust them to fit the particulars of your project, but it sure makes it easier to get started.

5. Long-form content writing, such as blogging

Let me first state an important caveat: you have to be really careful with long-form content writing and AI. Because of some problems that I’ll discuss later in the post, you have to do a lot of editing to take the quality from mediocre to great. You also have to be super careful about plagiarism, meaning you’ll be doing a lot of rephrasing and rework.

But still, imagine this: you can ask ChatGPT to outline a blog post for you. You make some tweaks to that outline, flesh out some of the details and gather some important statistics online. Then you can take that outline and plug it into Jasper AI. Jasper will then crank out the bulk of the blog post for you.

The text that results won’t exactly knock your socks off, but it’s still a damn good first draft. Treat it like the work of a talented intern, and don’t be shy on the edits. Do this, and you can end up writing longer, more detailed posts in less time.

I write almost all Weird Marketing Tales content by hand because I have a very particular voice in mind here. However, for more corporate and dry blogs, I’ve found that using this method can result in creating a 2,500-word draft in two hours instead of a 2,000-word draft post in three hours. That’s a 33% reduction in time spent writing and a 25% increase in output.

For best results for my clients, I usually spend that extra hour creating better graphics, sourcing better photos, adding better sources, and generally putting more polish into the work than I would be able to manage without AI assistance.

6. Social media posts

When you’re writing posts for Facebook, Twitter, or LinkedIn, or captions for Instagram and TikTok, the truth is, good content is pretty repetitive. You need to follow certain basic templates if you want to succeed.

ChatGPT and Jasper AI are pretty good at intuiting what these templates are and following them. They can help you create first draft copy, which you can then edit and post on social media.

That said, my absolute favorite use of generative AI for social media right now is the creation of conversation starters. It’s a known fact that on Twitter, LinkedIn, and Facebook groups, asking open-ended questions is a great way of starting discussion.

Now you can come up with questions on your own or source them from a list online, but you still have to spin them in a way that’s right for your target audience. At least, that’s the way it was before AI got involved.

For example, I used ChatGPT to create 20 open-ended questions to ask in a board game Facebook group. The core concepts were great, though I had to make them sound more “gamery.”

7. Audio editing

I’m a big fan of the tool Descript for audio editing. Many others I know are using it for video editing too.

After recording a podcast, such as the Weird Marketing Tales podcast, the audio is automatically transcribed. The transcription is about 95% accurate, but requires substantial cleanup. But I’m already in there editing anyway, so this is OK.

What’s very cool is that I can edit the transcript directly to edit the audio. The vast majority of the time, I can isolate certain words, remove them, and the audio sounds like the word was never there at all. It’s great for removing “uh” and “um” and other filler.

All in all, this allows me to crank out edited audio about 40% faster than before. Plus, I also get a transcript in the process, which would have previously required going to Scribie and paying someone to do it, which costs around $1 per minute of audio.

And if that weren’t enough, I can even have AI overdub my voice and make me say things I didn’t originally say. (There are some safeguards on this to discourage this from being used for evil purposes). This allows me to smooth over messy sections without pulling out my microphone and trying to screw with audio levels to make it sound like I didn’t overdub it.

8. Memes

AI can write and draw anything you want. I’ve asked it to write up a fictional Fyodor Dostoyevsky novel about a town being swallowed by a peanut butter avalanche. When asked to imagine a fictional conversation between LBJ and JFK where they wake up in purgatory and wonder how it happened, ChatGPT happily complied.

If you need to create droll or funny content, AI is great for that. So much so that posting funny ChatGPT and AI art generator outputs has become something of an art itself.

Eventually, when we become used to generative AI, this will become something I do occasionally for kicks. But while the conversation around generative AI is so hot, posting raw outputs of strange conversations with AI is a great way to get some attention online.

Warning: the following image will spoil Black Panther (2018).

We’re not even scratching the surface of what generative AI can do

Despite the thorough nature of this post, you should know that there are more AI tools out there than I have money or time to play around with. Here’s a whole list of tools on GitHub, the vast majority of which I had never heard of prior to writing this post.

It’s also worth noting that we’re all still learning how generative AI works and, by extension, how to work with it. The prompts and direction you provide to generative AI greatly change the outcomes you receive, but it’s not always clear why.

Best practices have not yet been developed. We’re all figuring this out in real-time. I would recommend, though, if you want to learn more about why generative AI works the way it does, check out this article for a primer.

The limits of generative AI

I love what generative AI can do for business. That said, it’s not without its quirks and problems, and out of a spirit of responsibility, I have to tell you what those are. You should be aware of where AI falls short before you implement it into your business processes.

1. It can’t fact-check.

For one, generative AI – bless it – is prone to hallucinations. For example, when asking for summaries of what certain software does to save me a little time when writing this post, I got an erroneous explanation of what Jasper AI does.

This is just completely wrong. Generative AI can’t tell the difference between fact and fiction yet. You’re going to save a ton of time putting together nuts-and-bolts sentences and checking your grammar. But you’re going to spend a huge percentage of that making sure ChatGPT and its peers aren’t making shit up, as they are wont to do.

2. It doesn’t have good judgment.

Language models like ChatGPT and Jasper can tell you the difference between WooCommerce and Shopify when you’re writing a compare-and-contrast style post for a client that works with eCommerce brands. But you know what it can’t tell you?

• The types of articles you need to be writing in the first place
• What the main points of your articles need to be
• Whether the writing is a good fit for your audience

These language models can put together 2,500-word college-level essays in thirty seconds. But just like mistrained undergraduate essayists, they often make well-written posts that nevertheless fail to make incisive points or share any valuable insights.

This problem isn’t limited to language models. It’s just easier to describe their issues. Midjourney cannot for its life seem to figure out how to draw hands. That’s because it has no eyes to review its own work. It can’t tell if something has “the right vibe” or not. Even the more sophisticated DALL-E 2, which allows for editing, requires a smart person at the wheel, giving good direction. And even then, you have to throw the vast majority of its outputs away before you get what you want, and that costs a lot of money!

Even the relatively simple system of Descript, which allows you to edit audio via making changes to a transcript struggles in some areas. It can’t quite tell where one word begins and another ends sometimes. In others, it can’t tell the difference between pairs of homophones such as “rows” and “rose,” which wouldn’t fool a human transcriber for more than a split second.

3. It doesn’t have opinions.

This is most pronounced in tools like ChatGPT. Because of the safety features that corporations add to chatbot AI systems, they usually won’t take a moral stance on anything. They will merely describe the pros and cons of the most common sides, and let you decide.

And look, I get it. OpenAI has to regulate ChatGPT’s behavior, as any AI company must do with its tools. Otherwise, you’ll see chatbots endorsing Donald Trump for president in 2024 or something else weird like that. And let’s be real – the world does not need to automate “hot takes.”

But still, when I asked ChatGPT whether music is better or worse than it used to be, it gave me arguments for both sides, and then said, at the end: “ultimately, whether music has gotten better or worse over time is a matter of personal opinion and perspective.”

That’s the right thing for a chatbot to say. But if a person said it, they would be copping out of making a genuinely persuasive piece.

4. It can’t quite look and sound like a human expert.

If you read a raw, unedited blog post by Jasper AI, it looks like something that came out of a medium quality content mill. In short conversations, ChatGPT sounds like a highly intelligent person, but over time you come to realize it lacks a nuanced understanding and fails to inject opinion into places where a human expert would and should.

Midjourney, Stable Diffusion, and DALL-E can make beautiful art. And yet when you look at it, it can seem soulless. It needs guidance and context in order to truly fit in.

In other words, anything you make with generative AI needs to be heavily edited if it’s ever going to get above middling quality. And middling quality doesn’t get attention on the busy internet – brilliance does.

5. It comes with copyright and ownership issues.

By nature, all generative AI tools take enormous amounts of data and use that to process it into making new creations. With tools like Descript, where you can voluntarily submit audio of your voice and give the system permission to make a synthesized version of your voice to overdub lines you don’t like, this is awesome. You’re giving them the permission to use your voice, which means they’re not violating copyright or your privacy.

But the language models and art generators? Not so much. They take other people’s work and transform them into something else, without their permission.

Some say this mimicks the true nature of the creative process. Some say it’s theft.

What I do know for sure is that if you are going to use anything an AI creates for profit, you have to edit it heavily. Otherwise, you have no idea whose text you are plagiarizing or whose art you are photobashing. The only way around this is to edit so much that you basically just end up using the AI to help you generate ideas.

Which, hey, that’s still a good outcome. Because like I said – 15-20% efficiency gains.

Final Thoughts

In its current iteration, generative AI is unlikely to replace our jobs. But it is very likely to profoundly transform them. In fact, it already has.

Pay attention to what generative AI can do and what it can’t. Go forth with a spirit of curiosity, and know that even if this doesn’t blast us into the Star Trek future, it’s still worth having open in a tab on Chrome.