ChatGPT-4 for Strategists (Part 1)

I know, another ChatGPT article, but this one is actually helpful

Nobody knows how AI works. Isn't that ridiculous?

“If we open up ChatGPT or a system like it and look inside, you just see millions of numbers flipping around a few hundred times a second,” says AI scientist Sam Bowman. “And we just have no idea what any of it means.”

With AI technology moving at the speed of light, so much so that the people who created it aren't really sure what it's doing, it's very easy to feel like you're getting left behind. It's no surprise then if you hadn't heard: ChatGPT had a major restructure, and this article aims to explore the various new and improved practical applications available to you, so that you don't have to.

We'll focus on key features, their use cases for Strategists and some prompts you can use to best leverage them. I imagine that by the time I've finished writing this article AI will be serving you breakfast and tucking you into bed, but at this moment in time, these are the most usable functionalities based on my own personal experience.

GPT-4's Time to Shine

The reason why I'm here inside your brain talking to you today is because users of ChatGPT's free version (which, let's be honest, is most of us) now have access to GPT-4 instead of GPT-3.5. This is Open AI's "smartest and most capable model" and comes armed with a whole host of additional capabilities. There's millions of articles you can read on these differences, but here are the major ones we'll be benefitting from:

• Upload and communicate with images and documents.
• Create diagrams.
• Generate Images.
• Internet browsing
• Noticeably better understanding and informed responses.
• Broader knowledge set.

Before a LinkedIn bro passively aggressively comments, it's worth noting that the most advanced model available is GPT-4o (where the 'o' stands for 'omni'). This sparkly new model, currently in beta and available for paid subscribers, seems to include more backend refinements - only really noticeable for people building AI applications - rather than key features, so for the sake of this article we're focusing on the differences between GPT-3.5 to GPT-4.

Three is the Magic Number

I know what you're thinking, "But Jonny, how am I supposed to ingest such an extreme amount of wholesome goodness and supreme intellect in only one sitting? It's simply too much to handle". The good news is that you don't have to.

Since the best things come in threes - blind mice, wise men, little pigs - so too will this article series. Each one will dive into a different type of interaction with our soon-to-be robot overlord, with the order as follows:

1. Part One: Images
2. Part Two: Documents
3. Part Three: Text

Without further ado, please, come on in:

Images

The ability to upload and communicate with images in ChatGPT is a huge advancement, and it's as easy as it sounds - attach an image along with your prompt to give ChatGPT additional context. Below, we'll focus on the most useful applications as a Strategist, but it's arguably even more of a gamechanger for Creatives (think checking for unconscious biases in ads, ensuring a piece of creative is adhering to certain brand guidelines, suggesting improvements to enhance diversity and inclusion, etc. etc.).

#1 Content Audit:

Content audits come in many shapes and sizes, but in this example we're talking social content. Importantly, you don't need to be working on a social media project to benefit from a social content audit. Social insights offers some of the most dynamic, real-time, and consumer-centric insights available. Where else can you find a neatly organized, chronological arrangement of a brand's entire catalog of communications?

However, continually scrolling through Instagram feeds to extract brand and creative insights can be incredibly time-consuming. This makes it a perfect use case for ChatGPT - take a few screenshots and attach them with a prompt like so:

Prompt:

"I am about to upload screenshots of the Instagram feeds of Nike, Adidas, Reebok and New Balance. Once I have finished attaching them, please summarize, compare and contrast the content style and artistic direction for each brand, making sure to pick out key insights or differentiators"

Result:

What would have taken hours now takes less than a minute, and the output is both accurate and articulate. In the past I have often pondered on how to write up creative observations in a style that resonates with our Creative friends, but I'm pondering no more.

It's worth noting that I chose not to request request strengths, weaknesses, opportunities and threats, which is typically a core aspect of a content audit. I prefer to do that part my self; as a general rule of thumb I'll use ChatGPT for the heavy lifting and my own humble brain for the more critical thinking tasks. That's not to say ChatGPT couldn't handle a SWOT analysis—I'm confident it would do a great job, however, I believe it's important to stay anchored in the research. It's a very slippery slope if you start employing ChatGPT from start to finish.

Interestingly, you don't even need to specify which screenshots correspond to which brands. I ran a test, asking ChatGPT whether or not it could identify which screenshots were corresponding to which brands, and it was totally accurate, breaking down in detail how it came to each conclusion:

Not only did it use the obvious visual clues (logos, products etc.) but incorporated elements of its own prior knowledge, for example identifying 'prominent sports figures associated with New Balance' which I would have struggled to do without extensive further research. The sophistication and nuance of AI's image recognition capabilities have come a long way.

#2 Break down a campaign or piece of creative.

As Strategists, we spend a lot of our time critiquing the work of other people. It could be work that your agency has created, that your client has created, or that of a competitor brand. It could be to help position your client's brand uniquely and more effectively against competitors, or it could involve analyzing highly successful campaigns to reveal what resonates with a particular target audience. Essentially, this involves a combination of looking and thinking, and this is something our mate ChatGPT is great at.

Whereas the social content audit use case above is primarily a time-saving exercise, where you could probably copy and the results exactly as they are (although I still wouldn't recommend it), the value in this case lies in stimulating your thinking and exploring paths you may not have previously considered. You can't really go wrong, as there is no 'correct answer' for ChatGPT to solve with regards to observations and insights, merely a stream of thoughts and ideas to help support your strategic ideation:

Prompt:

Result

Consider me impressed - this is a pretty stellar breakdown of an amazing ad. This use case probably offer the most value if you're pressed for time and are wanting to analyze a whole portfolio of work. You could attach 10 ads and get 10 thorough write-ups in a matter of seconds. Think of ChatGPT a bit like your own personal baker (bear with me here). Instead of having to build ten cakes yourself, from top to bottom, just so that you can pop the cherry (the insight) on top, ChatGPT lines up the pre-baked cakes in an orderly row and watches you confidently stroll along, a smug look on your face, popping the cherries (the insights) on top of the proverbial creative cakes, one by one, in record time. If you stuck with me through that then bless you.

The point I'm trying to make is that you're still using your own brain to critically evaluate the ads, you're just using ChatGPT to get a head start on the potential fundamentals of each one.

#3 Get Consumer Feedback.

Using ChatGPT for consumer feedback can significantly enhance strategy development by providing quick, cost-effective, and insightful preliminary research. While it shouldn't replace traditional in-person focus groups, it serves as really nice addition to the research toolkit. By uploading product or packaging images, ChatGPT can simulate likely consumer responses and help identify common questions consumers might ask when faced with the product.

For example, during the early stages of research, conducting extensive surveys or focus groups may not be feasible due to time and budget constraints. ChatGPT allows for the collection of predictable feedback quickly and efficiently, offering a broad understanding of potential consumer reactions. It's important to note that this approach is about identifying common trends and patterns rather than uncovering groundbreaking insights.

One of the key benefits of using ChatGPT is its ability to highlight seemingly obvious aspects that might have been overlooked. By predicting the most likely concerns and questions, it enables strategists to plan and address these issues proactively. This way, when the time comes for more detailed, in-person research, the team is better prepared and can focus on more nuanced insights. In essence, ChatGPT enhances the research process by providing a foundation of average consumer responses, helping to shape more targeted and effective strategies.

The suggestions made are practical, actionable and reasonable. The balanced critique and recommendations demonstrate how AI can complement human insights and enhance the overall quality and direction of certain projects.

#4 Charts and Visualizations.

There's a lot of quickfire use cases that fall into this category, and I'm probably only scraping the surface here, but as I see it, the main value of ChatGPT in this instance is in saving you from doing a hell of a lot of Googling and spreadsheeting (yes, I used spreadsheet as a verb, get over it). We've all been there, spending far too long perfecting a pie chart, finding sources to support a chart or cross-referencing a visualization with Google research; these are all perfect use cases for GPT-4:

(a) Redraw a Chart

Prompt: "Redraw this bar chart into a pie chart, with each country corresponding to the color primarily associated with that country. Include annotations"

Result:

Second Prompt: "Make sure the percentages are clearly visible - make the text white and more space apart from each other"

This is something that I can see myself using all the time. I could have stretched this even further, getting hyper specific with my requests about exactly how I want this chart formatted. The results are fantastic and it means no more faffing around with text boxes, color pickers or spreadsheets. Additionally, if you have a series of charts or visualizations within a deck you could ask ChatGPT to ensure the formatting and layout remains consistent.

(b) Get Contextual Understanding on a Standalone Chart

Prompt: "This chart, from 2022, represents where Canadian consumers expected to increase their spending over the next six months. I don't have access to the report it was taken from. Based on what you know, and other internet sources you have access to, give me some contextual understanding - cultural, financial, socioeconomic - as to the results. List your answers as bullet points and keep them succinct"

Result:

I love this. The sources ChatGPT has used (in the brackets at the end of each bullet point) are all highly respectable, industry-leading publications, so you can be assured that the facts it be spittin' are legit. However, at risk of repeating myself, please don't go and start assuming every result is 100% truthful; it's still critically important to think about the results you see and evaluate them on their own merit, but in this instance they seem spot-on.

(c) Make Sense of Your Results

This one is particularly important for all you Social Listening wizards out there. Social listening platforms, and other SaaS products, are packed with a ridiculous amount of out-of-the box analytics and visualizations. They look real sexy and can be an impressive addition to a deck, but it can be really difficult to get any true value out of them.

Having worked in the Social Listening industry for a number of years, the number one reason for a client not renewing their subscription was because they simply couldn't get value out of the charts and widgets they were seeing, and users found it difficult to implement them into their workflow. I see this as a fantastic use case for ChatGPT:

Prompt: "I have run an audience segmentation of a brand in the fintech industry. Here is the visualization of the audience segments, whereby the size of the bubble represents the relative size of that audience segment, and the proximity of the segments represents their similarity. What insights can you gather about the societal makeup of this audience, and what further research or questions should I look into to gain a better understanding of how to target them?"

Result:

The initial observations and insights might seem somewhat limited, given the information provided is primarily bubbles and segment names. However, the more valuable part of the response lies in the "Further Research and Questions" section, where ChatGPT poses practical questions that could lead to unlocking genuine insights. It's important to note that ChatGPT's strength is in accelerating the process, allowing you to focus on what truly matters. While you could generate these questions and thoughts on your own, ChatGPT offers a neatly formatted list of considerations. This allows you to quickly review and decide which ideas to pursue, saving you time and effort in the brainstorming phase.

#5 Digitizing Real-Life Sketches and Writing (nearly)

This is actually more of an example of what not to use ChatGPT for. My idea here was along the line of 'take a picture, upload it, let ChatGPT do the rest'. This 'snap and upload' methodology could have widespread implications, for example turning whiteboard notes from a brainstorming session into a nicely formatted document, taking a picture of an interesting slide from a conference and creating a version for yourself, or transcribing a sketch of a diagram you concocted into a digital version.

The issue, however, is that, as accomplished as AI image-generators are for mock-ups and sketches of things and people, they are pretty useless when it comes to generating text in an image, which is a fundamental part of the above concept. You'll see what I mean here:

As you can see, ChatGPT doesn't seem to possess the same level of logic as when dealing in text-based conversations, clearly struggling to interpret what my writing says and seemingly lacking any knowledge of how to spell or what real words are.

At first glance, it seems to have done a decent job in both instances, however the longer you look, the more you see that it's absolute useless nonsense.

The reason I include this though is that, with the speed at which technological advancements in AI are happening, it's incredibly likely that this will be a legitimate and highly practical use case in the not too distant future

#6 Create Mock-ups for Strategy Slides

We're Strategists, not creatives, but sometimes we want some pretty pictures for our slides. For client-facing presentations our Creative friends will happily throw something together for us, but there are certain occasions - brainstorms, creative briefs, workshops - where mock-ups aren't essential, but are definitely helpful.

It's in these instances where you can now make the most of ChatGPT's image generation capabilities.

This one is probably more of a nice-to-have than a need to have, but useful nonetheless, and can definitely help breathe a little bit of life into an otherwise lifeless document,

Conclusion

There we have it. We’ve journeyed through six broad use cases for ChatGPT’s image upload capabilities:

1. Content Audits
2. Breakdown a Campaign or Piece of Creative
3. Consumer Feedback
4. Charts and Visualizations
5. Digitizing Real-Life Sketches and Writing (nearly...)
6. Creative Mock-Ups

But hey, this is just the tip of the AI iceberg. There are likely hundreds more ingenious ways to leverage GPT-4, and I’m all ears.

Stay tuned for the next article where we dive into the world of document uploads in ChatGPT!