California Management Review
California Management Review is a premier academic management journal published at UC Berkeley
by JungYun Han and Jelena Tiu
Image Credit | zongyi
The rapid emergence of Generative artificial intelligence (AI), referring to “machines performing cognitive functions that are usually associated with human minds, such as learning, interacting, and problem solving”1 quickly changes our lives. These algorithms, trained on vast datasets, can now perform many parts of human-like functions 2. However, an interesting but underexplored question is whether generative AI can be effectively employed in human-centric work, such as creative industries where creativity has traditionally been considered immune to technological advancements3. This is because in these industries, creators require both cognitive and emotional capabilities4,5 with feeling capability being innate to human nature 6. Furthermore, generative AI falls short in understanding human motivations, and capturing the nuanced context of consumer needs7, thus resulting in less authentic content8.
Vegard Kolbjørnsrud, “Designing the Intelligent Organization: Six Principles For Human-AI collaboration,” California Management Review 66/2 (2024): 44-64
Roland Rust, Ming-Hui Huang, Vojislav Maksimovic, “The Feeling Economy: Managing in the Next Generation of Artificial intelligence (AI),” California Management Review, 61/4 (2019): 43-65.
Thus, we pose a frontier question in creative industries: How can creators leverage generative AI to generate, elaborate on ideas, and even co-create content?
To address this question, we explore the case of the leading media advertising agency in the Philippines, which has recently started creating digital advertisements with generative AI (referred to as “Stillwater Agency” for confidentiality). The advertising industry provides an ideal context for examining our research question because the advertising industry is among the most commercially driven and technologically advanced sectors within the creative industries. In addition, as this industry utilizes a wide range of content formats, including text-based (e.g., Google Ads, social media ads), visual (e.g., TV commercials, logos, photography), and audio (e.g., radio ads, podcasts), such diversity provides a rare opportunity to compare the use of different generative AI tools for various types of content.
In the following section, we explain how Stillwater Agency’s Content/Creatives department has recently used generative AI in advertisement creation. This department is responsible for producing advertising materials and consists of two independent sub-teams: the art team, which handles visual content and communication, and the copy team, which focuses on written advertising and marketing messages.
Our case study reveals distinctive patterns in the use of generative AI for creating digital advertisements based on the agency’s goals (innovativeness vs. efficiency) and the types of content produced (visual vs. text-based). First, amid intense competition for creative content and mounting pressure to deliver advertisements within tight deadlines, creators often face a trade-off between innovativeness—producing novel and original content that resonates emotionally with customers and inspires purchasing and efficiency—creating content quickly. Second, we compare the application of AI tools in visual-based and text-based advertisements, which require different types of technologies. In the current AI landscape, both visual (e.g., Midjourney, Firefly) and text-based (e.g., ChatGPT, Gemini) tools are powered by machine learning, though their technical modalities differ.
In 2023, the art team got a new assignment for creating visual materials for a top food and beverage company based in the US. The main task was to create visual material with an AI-generated chef that users can ask questions regarding recipes that users can easily follow at home. Once the visual team decided on the concept for the material after background research with AI, they used Midjourney (AI tool for visual content generation) and entered the first prompts that described the AI-chef, “Filipino chef lady, in a rustic kitchen, Pixar style”. Then, the team discussed the quality of the AI-generated image, followed by giving feedback to Midjourney (i.e., keep revising the prompt) until the team was able to get the image that they wanted. According to the senior manager in the art team, the team believed that an AI-created visual, such as the AI-chef, would capture market attention and help customers navigate the company’s product catalogue in a creative and engaging way. Although it was their first attempt at using AI for ad creation, they saw this experiment as a worthwhile opportunity, especially given the rising trend of AI in advertising.
The senior manager recalled that after numerous rounds of feedback and revisions with Midjourney, the team was satisfied with the quality of the image. They continued tweaking their prompts until the final product met both their and the client’s standards. The team presented the AI-chef image to the client, and after receiving approval, they moved forward with the completion of the campaign materials. During this final stage of refinement, however, the team relied less on AI and incorporated more input from team members to add a human touch and integrate their creative ideas.
While using Midjourney significantly shortened the initial phases of image production itself, the overall production time involving numerous iterations, discussions, and revisions did not mean a dramatic time-saving. Instead, the primary benefit of using AI in this case was to generate locally relevant and MZ-friendly images of the AI-chef, which aligned well with the team’s goals. Ultimately, when the team reported the final output to the F&B client, the collaboration between AI tools and the creative team met the client’s expectations for authenticity, technical quality, and usefulness.
An automotive brand based in China requested collaterals, a TV commercial, and a digital video for their new car launch with a short lead time of only three to four days as well as a tight budget. They did not ask for AI-generated output, but the art department was forced to use AI given the tight lead time. According to the project manager, “Client wasn’t looking for it on the on-set, nor were they informed of the working process behind the material. It was done out of necessity, given the workload from other brands coming in, and the struggle to integrate assets from regular stock photo sites into a working Car key visual.”
Unlike the previous case with the AI-chef (Scenario 1), the art team in this instance prioritized speed over novelty in producing materials for the car promotion. Stillwater Agency has extensive experience with car promotions, which tend to follow a more standardized format. After agreeing on the overall concept and direction, the team quickly entered prompts into Midjourney to generate images depicting the desired car visuals, repeating the process only a few times to fine-tune the results. As compared to Scenario 1, the revision process here was much quicker, with minimal human feedback. The creators selected the best image after some rounds of prompt adjustments, then polished the final advertisement. In this case, the objective for using Midjourney was straightforward: meet the tight deadline while maintaining an acceptable quality of work. From interviews, we learned that for many projects with constrained budgets, agencies have a high hope for this approach to save time and minimize the need for extensive human involvement.
Interestingly, when interviewing copywriters to know their use of AI tools, until now, we could not find any case that ChatGPT or Gemini helps them to write catchy and memorable advertisement phrases. Copywriters consistently expressed skepticism about AI’s ability to produce writing that matches human-level creativity. Instead, these tools are primarily used for researching and benchmarking existing content. Copywriters told us about their experiment with AI; some suggestions were generic and also found it difficult to make the communication more localized and culture and demographic-specific towards the Philippine market. Given that they produce text-based advertisements in the local Philippine language, currently available AI tools may not have the same quantity and quality of language data as compared to English.
A Filipino multinational food and beverage company wanted to create a Philippine-hosted app and website for foodies and home cooks, featuring recipes and articles with the brand’s products as the main ingredients. In this project, the copy team was required to churn out four articles per week associated with the brand. Text-based AI helped in searching relevant articles, providing a summary of the articles and reference links for certain recipes. However, the copywriters had to make some tweaks to the overall body of the articles to generate the required number of articles for publishing per week. The team greatly valued the time saved through the use of AI, as it managed most of the routine tasks that did not require advanced skills or creativity. By leveraging the AI’s capabilities for these straightforward tasks, the team could focus their efforts on refining the content and adding humor, thus efficiently meeting the weekly article production requirements.
Figure 1. Four Scenarios for AI adoption in creating advertisements
Client consent: Our interviews revealed that client consent for using AI tools is a prerequisite for working on highly innovative and novel content creation (Scenario 1). The senior manager in the art team noted that when they proposed purchasing Adobe Firefly licenses for all art directors to streamline the process and produce image content under tight deadlines, some clients declined the suggestion, saying “They didn’t think AI was reliable enough.” This indicates that, while creators may be open to using generative AI tools, the endorsement from clients is a critical factor for the proactive use of AI in generating high quality output. For this reason, the F&B client’s explicit request for generative AI for innovative advertising provided a new opportunity for experimenting with AI in visual content creation.
This finding raises the question of why explicit client endorsement wasn’t required for visual content creation when the main purpose of AI usage was to meet time constraints (Scenario 2). As mentioned earlier, most clients still remain hesitant about AI-assisted advertisements. This reluctance has led creators to avoid discussing the specifics of their work process in detail with clients, if possible. In particular, since this type of tasks follow standardized formats and expressions, clients are unlikely to distinguish between AI-generated work and human-created designs unless visual designers in art team specifically explain their use of AI. Furthermore, the art team perceives no need to disclose AI assistance to clients, especially for mock-up (interim output) for confirming the designs and conceptual directions with the clients. Instead, it simply speeds up production. In their view, whether the images are drawn manually or generated by AI, the outcome remains identical.
Legitimization: In the advertisement market, generative AI has rapidly evolved from being a secret weapon to a legitimate collaborator in just a few years. When tools like Midjourney (launched in July 2022) and ChatGPT (released in November 2022) were first introduced, advertising agents initially used them mainly for internal purposes such as background research and brainstorming. Their use was primarily limited to personal experimentation to evaluate the quality of the outputs. Creative directors noted that there was a reluctance to openly acknowledge the use of these tools due to concerns that clients might perceive this as a lack of ability to generate novel ideas independently.
However, perceptions within advertising agencies shifted quickly. Within a year, it became more common for creators to openly admit to using AI tools as part of their process. Furthermore, the F&B client’s explicit request for AI-generated advertisement eased their pressure on ‘human creativity’ and encouraged experimenting human-AI collaboration in content creation more openly. The senior manager involved in the visual project told us, “It was the first case that our client explicitly requested to use the AI tool for their advertisement material. We were surprised by the client’s request, but soon realized that this could be our standardized work mode”. The agency is now undergoing significant changes in its adoption of AI. The F&B case in Scenario 1 implies that in spite of several hurdles (e.g. client endorsement), more advertisements may be produced through a closer collaboration between AI and creators.
No copy-creation with text-based AI: Interestingly, we could not find the evidence for the case where copywriters use text-based AI tools to produce advertising copy (Scenario 3). We asked the copy team about whether other advertisement companies use text-based AI tool for copywriting work. But, they said “As far as they know, no other agency”. This happens mainly because current tools are incapable of delivering sophisticated meanings and nuances in the local language, Pilipino. Despite this, copywriters remain optimistic about the future potential of AI in writing, too. They anticipate that as AI accumulates more data in non-English languages and enhances its machine- learning capabilities, it will eventually be able to handle nuanced text creation more effectively.
Co-creation vs. Sequential collaboration: The pursuit of innovative ideas by creators fosters a collaborative co-creation process with AI in visual advertisement (Scenario 1), whereas a focus on time-saving leads to a sequence of AI-driven production process with human selection (Scenario 2). This means that in visual content, the visual team iterates the co-creation loop by providing feedback to generative AI and revising the content until the quality work is produced. In contrast, in the latter scenario with the automotive client (Scenario 2), generative AI produces a range of potential solutions, and humans select the most suitable option. The collaboration is divided into AI creation and human selection, and the interactions are sequential.
Distinctive goals: The goals of innovativeness versus efficiency among creators are distinctly different. In the case of the F&B client, the art team’s primary objective in using AI tools was to generate more innovative and novel ideas to attract attention, rather than to save time. Although AI tools did help shorten the production time for each image, the overall production time—including numerous rounds of prompt revisions and evaluations and feedback by creators—did not necessarily result in significant time savings (Scenario 1).
Different growth rates: From the initial survey and follow-up interviews with project leaders, we learned creators anticipate fast growth of AI-assisted visual content in the advertising industry (for efficiency in Scenario 2). They mentioned the other project with FMCG client in fashion was the same with Chinese automobile case, in terms of advertisement creation process. Also, the creators in the art team told us that Scenarios 2 can be widely used, particularly when producing mock-up material as an interim output for discussion with clients, instead of producing the final advertisement because it is the hassle-free choice for the art team, saving them time from several days to a few tasks. On the other hand, in the text-based advertisement case (Scenario 4), one project leader mentioned that his team used ChatGPT for the benchmarking one client in the utility industry, in which the use of AI was limited to background research only.
Our study opens up some follow-up questions: First, does the collaboration with AI tools contribute to revenue increase? Our interviewees in Stillwater agency did not have their own answers, given that it is too early to conclude its financial impact. Yet, they hinted that as clients also know that AI adoption can reduce production time significantly, they want to cut contract fees as well. This means, an advertisement agency may face a lower fee for each advertisement, instead striving to increase production volume to maintain the revenue. Or, on the contrary, we also conjecture that at least in highly novel advertisement projects, co-creating with AI (Scenario 1), there would not be that many changes in contracts, given the creators’ efforts and time for the genuinely novel work.
Furthermore, our study focuses on the relationship between advertising agencies and industry clients (e.g., F&B companies, automobile manufacturers). However, since advertisements are ultimately consumed by end consumers, it remains unclear how these consumers perceive and respond to AI-generated advertisements. Will they accept such ads as legitimate and engaging? Or will they react negatively and harbor antagonism towards AI-driven content? While we have not yet seen the results of the AI-chef advertisement in end consumers, we conjecture that young generations are open to AI-generated content. For example, the popularity of AI singer teams in Korea implies that young generation tends to perceive the virtual singers’ performance as sufficiently touching and emotionally appealing. In sum, understanding consumer attitudes towards AI-created advertisements is crucial for assessing the broader impact and acceptance of AI in the advertising industry.
Our findings also suggest practical guidelines for using generative AI in the creative industry. Rather than viewing AI as merely a threat or boon, our comparison across four scenarios illustrates how creators can effectively utilize AI across different stages of production—from pre-production research and content creation to post-production editing. This means creators should assess their specific needs and the availability of appropriate technologies before exploring potential collaborations.
Paul M. Hirsch, “Processing Fads and Fashions: An Organization-set Analysis of Cultural Industry Systems,” The Sociology of Economic Life, (1972) 340-356. Routledge.
Jones, Candace, Mark Lorenzen, and Jonathan Sapsed, The Oxford handbook of creative industries. OUP Oxford. (2015).
