Generative AI for Marketing Content Creation: New Rules for an Old Game

Abstract

Finding the right content determines success in diverse activities and in various marketing functions such as advertising, public relations, social media marketing, customer relationship management, inbound marketing or personal sales. High-quality content is needed when communicating brand purpose, responding to a social media firestorm or investing in high-reach channels such as traditional TV advertising. Top-of-funnel marketing activities such as blog or social media posts, search engine marketing, press releases, e-mail marketing, cold outreach in sales or the creation of landing pages for inbound marketing all benefit from more efficient creation of more content. According to estimates from communication giant WPP, savings can be as much as 10 to 20 times.

Headnote

KEYWORDS

GenAI, Marketing Content Creation,Marketing Productivity, Content Quality

Content creation is essential to marketing x A striking image, a catchy claim, a funny video: High-quality content builds brands, excites audiences and persuades them to purchase. Finding the right content determines success in diverse activities and in various marketing functions such as advertising, public relations, social media marketing, customer relationship management, inbound marketing or personal sales. What makes generative AI (GenAI) truly revolutionary is that the same technologies promise to impact the entire value system of the content of marketing management. We have learned that GenAI can help produce high-quality content of literally any modality in terms of text, images or certain types of video. In a recent survey of 600 B2C marketers, about 50% report regular usage with plentiful applications as Fabian Buder and colleagues report in this issue (p.50). Given the versality of GenAI and the plethora of its potential applications, how should marketers make sense of the scope of change driven by GenAI?

Marketing trade-offs between content quality and content quantity >< It is useful to take a step back and consider the fundamental marketing trade-offs that GenAI is breaking up. Any piece of content created in marketing always involves the fundamental decision on how much time and effort to invest in content creation. Very simply: The longer one takes, the less one creates.

With faster pace and proliferating channels, more content is being required from marketing. Capitalizing on market opportunities as well as responding to competitive moves or societal developments requires efficient means of content creation. On the other hand, careful reflection is sometimes warranted. Any successful sales representative carefully analyzes prospective high-value clients before choosing what to show and in which way. High-quality content is needed when communicating brand purpose, responding to a social media firestorm or investing in high-reach channels such as traditional TV advertising. Marketers sometimes simply need to get quality right. This relates to ultimately winning the hearts and minds of customers. Building deep connections in this way necessitates deep customer insights. These take time to obtain and digest and limit the amount of content marketers can produce.

These limitations mean that marketers can communicate with many customers superficially and with some in depth. They can consider many alternative communication strategies and perfect one in full detail. At every step of the way, trade-offs between quantity and quality are being made. It always comes down to decisions of when to perfect and when to cut corners.

Leveraging the new economics of content creation in times of GenAI x The trade-off between quantity and quality defines the productivity frontier of marketing. Every organization has limits of how much content it can produce. Expanding quantity comes at the expense of quality, and vice versa. With GenAI, this trade-off suddenly breaks down. Three main avenues emerge: driving quantity without comprising quality, driving quality without reducing quantity and driving both quantity and quality simultaneously (see Figure 1).

* Driving the quantity of marketing content with GenAI x More efficient content creation is the most obvious purpose for GenAI in the realm of marketing. Most prominently, ChatGPT and associated Microsoft services, Google's Gemini or Meta's open-source Llama help with ideation and the blank page problem, assist in finding alternative forms of expressions or adapt messages to specific audiences or channels. Top-of-funnel marketing activities such as blog or social media posts, search engine marketing, press releases, e-mail marketing, cold outreach in sales or the creation of landing pages for inbound marketing all benefit from more efficient creation of more content. Specialized solutions such as Jasper.ai or Copy.ai have been optimized for marketing purposes and can efficiently build on brand voice, product knowledge or website information to take context into account.

When it comes to images, even more profound efficiency gains are attainable. Useful content can be created without photographers travelling the globe, searching for suitable locations or waiting for the right weather. Dall E, Midjourney or Stable Diffusion can create a sunset at the beach as quickly as it can a snow-covered alpine village. Developments such as in Firefly from Adobe allow the adding of one's own images to guide creation, and image-to-text conversion can assist in finding suitable prompts to expand on.

In terms of content distribution, advertising platforms such as Meta provide tools to adapt creative assets, generate new backgrounds or expand images to adjust to different aspect ratios, making distribution more efficient. Through Taboola, marketers can create entire online advertisements including headlines, descriptions and image content based on their landing pages and directly execute distribution.

The efficiency and content quantity rendered by these applications are particularly useful when marketers want to continuously engage audiences with new content, quickly capitalize on recent trends or promptly flush out new products or when many channels and languages need to be managed. Heinz ketchup, for example, has leveraged the hype around GenAI with various ketchup-related prompts, demonstrating that most images resemble the iconic Heinz bottle. Mattel reports generating four times as many Hot Wheels toy car concepts with GenAI. Other marketers attempt to engage communities to further increase the variety of content. Coca Cola has invited artists to join its Create Real Magic digital platform to produce images for display on billboards in New York and London. All these developments are possible due to the dramatic reductions in content production costs. According to estimates from communication giant WPP, savings can be as much as 10 to 20 times. But these gains in efficiency do not free marketers from making informed decisions about what to communicate. While off-the-shelf models provide impressive content and are quick to implement, they are not optimized for specific marketing objectives. Establishing a clear strategy and operational measures remains essential, in addition to developing the right prompts, as Oguz Azar argues in his article (p. 18).

Driving the quality of marketing content with GenAI x The efficiency gains of GenAI are an intuitive starting point for improving the decision-making of marketers. With GenAI, marketers can explore a broader solution space than previously conceivable. Proponents of agile marketing have long urged marketers to quickly test realistic concepts, collect feedback and iteratively improve. Such measures have been limited by the number of concepts marketers could communicate and the level of quality with which those concepts could be represented. With the ability to produce better visuals, marketers can leap more directly to concept illustrations that feel more real to customers, enabling better informed feedback. Taking a broader scope of alternatives into account and communicating these further increases the chances of finding the best possible solutions for the task at hand.

These developments require marketers to rethink the way they assess the alternatives. Think of creating advertising copy. Traditional market research and consumer surveys center on collecting many responses to just a few alternatives from a large number of consumers. This allows for the collection of detailed and nuanced feedback on each individual alternative. Testing more alternatives necessarily limits the sample sizes per alternative and the number of diagnostic questions. Online survey solutions such as Qualtrics, quantilope or QuestionPro are integrating GenAI into their systems to assist rapid survey development. However, one needs to be aware that the fewer performance indicators collected, the more essential it becomes to find the right ones.

To scale further, one can combine predictive Al with generative Al to assess even more options without the need to collect feedback. In their article (p. 36), Christian Scheier and Dirk Held discuss how integrating different Al applications can help in finding a more manageable solution space. Another approach is to utilize GenAI directly for assessment. With its ability to reflect on different types of input, GenAI can simulate individual customers. In pricing, even off-the-shelf ChatGPT can provide reasonable inferences on price response for mainstream products. Providing better training data on the actual market will likely further improve such predictions. While the actual potential has yet to be explored, it can offer exciting options to flexibly simulate various possible market conditions. The ability to dig deeper wherever marketers see potential is essential to obtaining insights that have impact. GenAI is capable of not only simulating "what if" scenarios but also providing rationale and relevant visuals with the click of a button.

Sales is another area abound with highly variable and unstructured information, such as email chains and audio and video of personal interactions. Services like Microsoft Copilot for Sales or Salesforce Einstein GPT digest this information and can combine it with other sources, including publicly available data. GenAI excels at summarizing such content. It allows sales agents to drill down to any needed level of detail. It can consider the subtle signals and language nuances both within a prospective customer and across customers. This ensures that no important details are missed. It allows for quick snapshots of opportunities and more effective time allocation. In their article (p. 30), Xueming Luo and his colleagues show these effects empirically while also pointing out that freeing sales agents from mundane tasks makes their jobs more challenging.

* Driving the quality and quantity of marketing content x In advertising, the Internet has dramatically reduced content distribution costs and provided access to advertisers with smaller budgets. The volume and diversity of content and channels are already astonishing. GenAI has the potential to skyrocket this development into other dimensions. This makes it more important than ever to stand out and cater to individual target groups with the best possible content fit.

Accordingly, much of the excitement around GenAľs marketing potential centers on hyper-personalization. It seems intuitive to extend personalized content distribution to personalized content creation. An airline might display different ads and adapt landing pages depending on whether customers have previously visited a destination. While human content creators might have many intuitive ideas on how to do that, automatically adapting with GenAI is less trivial. Further complicating matters, one would ideally optimize content creation and content distribution simultaneously. However, these are two highly distinct tasks (see Interview with Taboola CEO Adam Singolda on p. 56). Full hyper-personalization therefore requires solving multiple technical challenges. To get there, one would first need to be able to create content that is optimized for market objectives without requiring human intervention.

Can GenAI accomplish the tasks previously performed by teams of market researchers, marketers and creative agencies? It is important to note that such tasks differ widely from what common GenAI applications do. They do create content, and they can even be promoted to create advertising content, so the problem might appear to be solved. But solutions such as ChatGPT or Midjourney are trained on objective data. Midjourney can produce images of a car, chair or cat, and ChatGPT mostly gets linguistic facts right. This is not the same as choosing which content to display to engage a target group. No single objective truth exists for these marketing objectives. What works for marketing changes over time and differs across individuals. It requires customized models based on unbiased, subjective training data, as Christina Schamp and colleagues explain in their article (p. 42). Box 1 and 2 illustrates how this can be done.

GenAI can be trained on other objective marketing functions beyond advertising. Similar to the results in Box 2, better performance has been reported in research on product design as well as research on search engine marketing - see the article by David Schweidel and colleagues (p. 24). These developments will allow marketers to drive both quality and quantity. In advertising, for example, automated quality optimization is the basis for actual hyper-personalization. While we have trained images on the average population, the same can be done with any market or target group. Recent advances with multimodal models (Google Gemini, OpenAI GPT4, Apple Ferret) will allow much deeper image diagnostics to understand what drives marketing performance for individual target groups. This could feed into large action models or other GenAI-based agent systems that iteratively improve marketing communication through continuous feedback collection and optimization.

GenAI developments in marketing are promising, but challenges remain >< As these various examples illustrate, the productivity frontier of content quantity and quality continues to expand. In navigating these turbulent waters, it is useful to reconsider whether driving content quantity, quality or any combination is the primary objective. These objectives come with various challenges.

* Experiment with tools and adapt organizational structure and roles x Off-the-shelf solutions provide various means of improving efficiency and can create vastly more content. Fully leveraging the efficiency gains of GenAI includes experimenting with the various tools available. Not every tool works for everyone. Prompting can be a tedious exercise. It is therefore essential to find the best partners for individual content creators. On the one hand, marketers will want a culture of openness and experimentation. On the other hand, potential biases such as racial or gender prototypes need to be managed, and digital rights as well as enough human review need to be ensured. This can require new roles. Coca Cola, for example, has appointed a head of Generative Al. Such functions will have to manage a delicate balance of the necessary bottom-up experimentation and top-down control.

* Train Al models with brand- and consumer-specific data x Driving quality involves helping people make better decisions. This includes adapting systems to the needs of individuals and helping decision-makers to more quickly learn how to better perform their traditional roles. Harvesting the true power of Al often entails exposing GenAI models to company-specific data and fine-tuning these for the desired objectives. Furthermore, asking the right questions to GenAI and drawing the right conclusions is non-trivial. Whether it's innovation, advertising, product design or sales, the new lever of GenAI will make domain knowledge even more powerful. While it is intuitive to build knowledge on proper prompting and GenAI usage, improving domain knowledge and utilizing GenAI for learning hold similar promise when it comes to driving quality.

* Define application areas and develop new KPIs x Driving both content and quality is perhaps most exciting, but it also comes with the highest challenges. It is not yet clear in which domains both higher quality and quantity can be achieved. For example, the car advertisements above are much easier to create with GenAI compared to fashion advertisements, which have a stronger human presence. Marketers will need to identify meaningful areas or applications and develop new KPIs to efficiently monitor automated marketing. Once GenAI creates even more content and marketers can optimize its effectiveness, finding useful summaries of content production will also become critical. Managing such systems requires creating new diagnostic data and will fundamentally change the roles of all involved, including internal and external content creation team members.

One of the most successful applications of GenAI is the development of the GenAI itself. In this way, GenAI is fueling its own rate of innovation. This includes tailor-made solutions for specific tasks, improved multimodality, better understanding of image and video content and better responses to user input, and perhaps including agent systems that understand marketers and act on their behalf. With open mindsets, a strong interest in learning and the ability to see beyond the horizon, marketers can benefit tremendously from the added efficiency and effectiveness GenAI can bring to the table. x

Sidebar

Marketers will need to identify meaningful areas or applications and develop new KPIs to efficiently monitor automated marketing.

The productivity frontier of content quantity and quality continues to expand.

BOX 1

Study: An example of automotive advertising

For online advertising, Tijmen Jansen, Martin Reisenbichler, David Schweidel and I have investigated whether GenAI can go the full distance and create effective ad images. We retrained open source Stable Diffusion. To do so, we collected 543 online advertisements from the automotive industry. We then ran traditional market research and collected classic mindset metrics like attention, interest or liking on conventional 7-point scales as well as click rates. Average ad performance in the eyes of consumers varied fundamentally, with the best ad achieving an average above 5.50 and the worst a little better than 2.0 on the 7-point scale.

To automatically produce well-performing ads, we picked a recently introduced Polestar 3 vehicle and collected associated marketing images. In addition, we chose two possible brand positioning objectives: ruggedness and luxury. To train Stable Diffusion on these brand associations, we collected image collages from other sources as would be done in traditional advertising.

The text prompts Stable Diffusion uses are similar in spirit to discussions in traditional ad creation where market research and marketing effectively translate visuals into language to guide content creation of agencies. Good agencies have a great deal of implicit knowledge about effective marketing communication. Stable Diffusion does not. While it knows what ads look like, it does not know their impact on consumers' minds. We therefore needed to train GenAI new mindset vocabulary and relate it to image content, using well-performing ads, images of the target product (Polestar 3) and relevant brand associations (ruggedness and luxury). Based on this training, we were able to ask Stable Diffusion to produce images for any combination of advertising objectives.

BOX 2

Can GenAI create effective ads? Results from the study

Unedited Al-generated ads outperformed actual ads x Figure 3 shows the results on the average mindset metrics. On average, the generated ads received a rating of 4.55, while the actual ads received a rating of 3.79 by consumers. It is safe to assume that the actual ads involved creative briefings, photo shootings, image editing, perhaps even market research and some form of informed ad selection. We did nothing of the kind for the Al-generated ads. Despite this, they performed more than 20% better.

Harvesting the true power of Al often entails exposing GenAI models to company-specific data and fine-tuning these for the desired objectives.

References

FURTHER READING

Brand, J., Israeli, A., & Ngwe, D. (2023). Using GPT for Market Research. Harvard Business School Marketing Unit Working Paper No. 23-062. SSRN: https://ssrn.com/abstract=4395751

Burnap, A., Hauser, J. R., & Timoshenko, A. (2023). Product Aesthetic Design: A Machine Learning Augmentation. Marketing Science, 42(6), 1029-1056.

Jansen, T., Heitmann, M., Reisenbichler, M., & Schweidel, D. A. (2023). Automated Alignment: Guiding Visual Generative Al for Brand Building and Customer Engagement. SSRN: https://ssrn.com/abstract=4656622