How I've embraced AI tools as a Creative Leader — Case Studies in AI for Marketing.

I like tools. Tools are great, they allow us humans to do all sorts of things, faster, easier, or with less pain than before we had them. This applies equally to hammers as it does with software like Adobe Illustrator. Both tools have made life easier for professionals in their respective fields. However, it's important to note that these tools don't replace the professionals themselves. Instead, the professions evolved.

I believe the same principle applies to artificial intelligence (AI) tools. While they are more sophisticated and can be more advanced or creative, I see them as tools for humans to use in completing nuanced tasks. The keyword here is "nuanced." As a creative leader, I don't feel threatened by AI tools. Instead, I view them as shortcuts that reduce friction in achieving desired outcomes. The key lies in how we utilize these tools.

Instagram vs Reality.

Whenever AI tools are introduced, there tends to be a frenzy around using them to make quick profits. This often leads to scams, with promises that AI will do all the work and deliver precisely what people want to see. In my experience, most tools, whether AI-powered or not, often overpromise and underdeliver. I don't attribute this to software developers or marketing teams. Instead, it stems from the assumptions we make about what we hope the tool will do, and then feel disappointed when it falls short of our expectations.

This is particularly true in the realm of AI tools designed for marketing copy creation, brand image generation, or complex automation setups. These tools lack the context and true intuitive understanding of human trends and emotions to be fully effective on their own.

I have experimented with many of these tools and attempted to incorporate them into workflows and various projects. The key factor that determines success is having a specific objective in mind when utilizing an AI tool. I've come across numerous videos of people asking ChatGPT to write a business plan that will generate six figures in a year. While it's not impossible, what ChatGPT doesn't do is actually start and run a business for you. Some might argue that it's just a matter of time, but I believe it highlights a misconception about the amount of effort required to achieve meaningful outcomes that truly matter… to humans.

It's essential to acknowledge that certain stages of a workflow can indeed be performed more efficiently by AI than by humans. However, the instructions given to the AI are even more crucial than the actual scope of work. There's a delicate balance between investing human time and effort in forcing a tool to perform tasks it's fundamentally not adept at. Similar to human workers, AI tools and bots have their own specialties, pitfalls, advantages, and at times, personalities—often, it seems, manifesting as stubbornness.

Understanding Workflow

Before delving into the various tools and projects I've worked on, let's briefly discuss workflow. Every successful project can be broken down into a series of stages or tasks that are interconnected, resulting in a desirable outcome, whether it's a website, collateral materials, or even larger endeavors like campaigns or closed deals. I have yet to come across an AI tool that can completely replace my job and give me credit. However, I have found that attempting to replace a single step in a workflow with an AI tool tends to be more successful and efficient. What do I mean by this?

Is it worth it?

The input required by an AI bot needs to be balanced. The time it takes to explain to the bot what you want it to do shouldn't exceed half or even a quarter of the time it would take you to perform the task yourself.

For example, asking MidJourney to create a Monet-style oil painting of you and your extended family from a photo is likely a better use of your time than asking ChatGPT to write a personal love letter to your wife. This is primarily because ChatGPT lacks knowledge of you, and your wife, and would require extensive input anyway: just write it yourself.

I have chosen to document a series of case studies involving projects in which AI tools were either at the core, used as assistants, or replaced specific stages in a workflow. I approach these tools with a positive mindset, believing that there is room for all of them. It's important to recognize that we are still in the early stages of AI, and as we continue to discover better applications for these tools, they will become more adept at handling complex tasks.

AI Blog Content. Great right?

Let's begin with The Stacked Blog, a part of Stacked.camp. I want to focus on how AI tools, specifically ChatGPT, have been utilized for copywriting in the marketing world.

Stacked.camp is an open-source map that allows users to locate firewood for camping across the United States and Canada. As part of our content strategy, we have developed various branches, with one of them being a blog. When ChatGPT became a viable option, we started exploring its potential for writing our articles. We often hear about the ability to instruct ChatGPT to generate a finished article just by telling it what we want. While it is impressive that ChatGPT can write blog posts, we realized that it lacked depth and the quality of the content was relatively low. Crafting the blog's tone, positioning, format, and style of information required detailed instructions and a significant amount of work to ensure its relevance to our readers.

Initially, I asked ChatGPT to write a blog post about camping in Utah. While it did a decent job, upon reading the actual post, I noticed a few issues that needed adjustment. First, the length of the blog post was only around 300 words, which was insufficient from an SEO perspective. Second, despite discussing camping in Utah, the content did not align with our content marketing strategy, which focuses on firewood and encourages people to find firewood with Stacked. When I tried this approach with different states, the tone varied significantly. It became clear that for this method to be a legitimate way of generating copy in our brand's voice and effectively communicating our message, we needed more controls in place.

To address these issues, I began manually writing longer prompts that explained to ChatGPT what Stacked.camp is, why we built it, how we engage new users, and why we wanted to create a blog in the first place. These prompts were written in narrative form, specifying the desired length of the blog (between 1500 to 2000 words) and mentioning the SEO strategy, which aimed to include references to as many real-world locations as possible to increase search relevance.

After running these detailed prompts through ChatGPT, the results improved significantly. We were able to generate blog posts that included cited camping locations, actual references, and information that felt more relevant to people searching on Google for topics related to finding firewood for camping. While the posts still required some finalizing, such as copy editing, formatting adjustments, and factual refinements, not to mention MidJourney artwork, we had achieved our goal.

It is important to note that these articles are not intended to be artistic masterpieces or incredibly impressive. Instead, they are articles crafted from prompts designed to attract search engine results while providing relevant information for our users. Once we had a high-quality example of an article that ChatGPT generated, we could incorporate it into our overall prompt, explicitly telling ChatGPT that it had written this article and requesting it to do the same for different locations. Moreover, we discovered that we could include our opinion of the article as part of the prompt, further pushing the tool to create better output.

This process has been crucial in building our content strategy, style, and tone, which have resonated well with our users. It is important to recognize that using AI tools is an iterative process that requires learning and adjustment. It is far from the truth that we can simply ask a chatbot to fulfill our needs and consistently obtain useful results.

Swag For A Global Team

The next project I would like to discuss revolves around workflow and idea generation. As part of my team's responsibility at Blockdaemon, we assist our Asia-based marketing team in developing and designing their swag and sales collateral materials. The APAC market differs significantly from the United States, Canada, and European markets. The things that resonate and work, especially in the world of Blockchain, are culturally distinct. Being based in the United States, my team lacks the cultural background necessary to accurately understand, depict, and create the type of imagery requested. This is primarily due to a lack of understanding of local context and culture. Additionally, dealing with time differences and variations in working styles poses challenges for feedback and collaboration.

One of the more challenging aspects of developing well-thought-out brand swag is defining the core messaging or campaign idea. It often takes numerous iterations to come close to what the marketing team envisions from a designer. I realized that we were spending a significant amount of time assembling drafts and ideas that ultimately missed the mark for our APAC teammates.

To expedite our workflow, I introduced an image generation tool to create the initial drafts of branded swag. Replacing the manual process of idea generation proved immensely helpful in refining our concepts. We utilized MidJourney, an image generation tool, to generate 10 or 20 graphic ideas. These ideas could then be shared with our coworkers for feedback and direction.

This marked a pivotal moment for us. The tool did not replace the intellectual property and creative development work we typically undertake. Instead, it allowed us to present a multitude of different ideas within the same timeframe. Of course, there are challenges associated with it, such as the inability to easily edit the generated graphics or the level of input required to create nuanced designs. However, overall it enabled us to make significant progress in less time.

Using AI To Speed Up Production

The third project, which is slightly different from the last two, is CreativeBriefs.xyz. It is a project that exists primarily because of generative AI, although it doesn't rely on it entirely. It serves as a good example of a project, or set of tasks, that cannot be fully replicated by AI. Here's why:

The purpose of the website is to provide young designers and career-transitioning creatives with sample briefs to build mockup design projects around. Each brief represents real-world scenarios or situations that they are likely to encounter in their professional lives, with some being based on real companies or brands to make them even more realistic.

To assist in generating copy and creating contextual image assets for each brief, I utilize AI tools such as ChatGPT, Notion.AI, MidJourney, DALL-E, and StableFusion.

What sets this project apart is that while it incorporates a series of AI tools, none of them can function on their own or complete the entire job by themselves. Let's start with how I construct the briefs.

I begin by writing a draft of the brief without any AI assistance, usually based on a scenario I have recently encountered or a problem I have dealt with in the past. Then, I use ChatGPT and Notion.AI to help provide the context. This is where things can get tricky. I quickly realized that the structure of my briefs varied slightly depending on the context, which is a choice outside the control of any AI tool. To address this, I created a simple prompt, similar to The Stacked Blog example, that briefly explains the purpose and requests a specific portion of the brief, such as the context, instead of asking the AI to write the entire thing.

I then utilize portions of what the AI generated to help write the next section, adding additional bullet points myself to guide it in the right direction and provide more specific details about the various tasks assigned to the user.

However, what ChatGPT cannot assist with is the nuanced aspects of the briefs themselves—the elements that truly matter to other humans and will help individuals secure career positions. It is not enough to rely solely on an AI tool to generate an output and consider it good work. Instead, we need to use AI tools to enhance different stages of our work and ultimately integrate them into a larger, more comprehensive outcome.

AI tools are perfect for speeding up the process of creating more creative briefs in less time overall, but with me acting as the creative liaison between them. The critical aspect here is the relationship between me as the creator and AI as the producer. While AI can produce as much as I want, it is me, as the creator, who controls the process and must have an understanding of what my audience or users truly desire. The AI lacks the real-world contextual feedback loop that I, as a human, possess.

Once the actual copy for the creative brief has been written and the document has been edited and restructured, I use the image generator to create contextual images that serve as references for the users. This is where MidJourney proves to be valuable because I can ask it to create non-specific items. For example, I don't need it to create a pair of Nike shoes; I simply need it to generate a pair of sneakers. This project benefits from AI by filling in details and expediting production, but I do not see it as a replacement for my own creative thinking. Without my involvement in coordinating between each tool, the project simply wouldn't come together.

Combing AI Tools & Human Authorship

The fourth project I want to mention is a digital art project called Archive 51. This project is the opposite of the first three: its entire existence revolves around AI and serves as a commentary on AI tools themselves.

Archive 51 is a series of fictional images that resemble 1980s 35mm film, supposedly taken in the Nevada desert. These images appear to document rocket launches and propulsion tests that never actually occurred. The purpose of this project is to shed light on how AI image-generating tools can be used to create fake analog images, expanding AI's capabilities beyond digital artwork and photography. Three different image-generating tools were employed to create and process each photo.

Firstly, MidJourney, a text-based image generator, was used. The process began by instructing MidJourney to create a variety of photographs that emulate 35mm film and photography techniques. Each description depicted different rocket launches, gas canisters, filling stations, scientific meetings, and crowds of onlookers, all related to these non-existent rocket launches.

After exporting the images from MidJourney, it became apparent that the variations between them were too significant to be perceived as a series of images from the same event, captured with similar cameras. The issue lay in the style, nuance, and overall appearance of the images, which did not convincingly resemble analog photography. Even with prompts that specifically mentioned film types, lens lengths, and camera bodies, the range of color filtering did not align.

To address this, I introduced a non-AI tool called HUJI Cam. This tool utilizes algorithms to distort images and create light artifacts, mimicking the inherent flaws of analog photography. It also allows for control of coloration to create a cohesive look across the series of images.

However, even with these adjustments, I still felt that the project lacked credibility as genuine analog photographs. Simply having the appearance of the front side of an analog photograph was not enough to create the illusion of authenticity.

To address this, I turned to a third tool, a handwriting tool called calligrapher.ai that enabled me to generate realistic handwriting to be added to the "back" of each image. By creating the backside of each photo with a date and caption scribbled across it, the entire project came together.

Among the plethora of AI-generated images scattered across social media, I found the ones particularly intriguing were those depicting historic events like alien invasions, presidential meetings, and scientific gatherings that may not have actually taken place. However, these images are primarily created by users instructing an AI bot to generate a scene, resulting in a standalone image. I wanted to go beyond that and create an experience using AI tools that bridged the gap between all these tools, producing a cohesive collection of believable analog material.

More Info: https://archive51.art/

This Article. Yes, it's true.

Lastly, this article. No, I didn't ask ChatGPT to write an article for me to post on LinkedIn. These are my thoughts 100%. I've read every word and edited every sentence, but the process of writing this article is drastically different from the late nights I spent sitting in front of my bulky computer in college writing a paper.

First, I started by dictating this article to my iPhone using Apple Voice-to-Text as I paced around; it just works for me. As we know, this is never perfect, so I edited each sentence to ensure it made sense and conveyed what I meant to say. Then, I took each paragraph or section of the article and asked Notion AI to "improve the writing of these, but ignore grammatical errors and structural problems to maintain my tone of voice. Use at least 90% of my original writing."

Anyone who knows me or has chatted with me over Zoom can probably hear the cadence and vocabulary they know to be me - but perhaps with a bit more sentence structure and ease of reading. What can I say, sometimes tools get replaced or upgraded, changing the profession itself. Other times, new tools appear from others, creating a new shortcut to a successful outcome. This is simply the way some articles are written.

Like I said at the beginning of this slightly AI-corrected and reconfigured article: I like tools. Tools are great for exploring what humans can do.