Don't miss ADWEEK House at Cannes, June 16-19. Join us as we celebrate our 45th anniversary and explore the industry's now and next. RSVP.
In the past nine months, ad agencies have been inundated with pitches from new generative AI companies, as well as incumbent tech giants like Meta and Google.
However, as agencies delve deeper into the practicalities of existing generative AI tools, a growing sense of skepticism is taking hold based on frustrations with where the tools still need to improve, as well as a worrying homogenization of work that uses AI, tempering the mass excitement of the tech, according to eight agencies.
“Not a lot of attention [is] paid to [the] fit and finish of these tools,” said Adam Buhler, svp of creative technology at Digitas. “Everyone is trying to get their feature into the market before others and the implementation is not fully baked in.”
Keeping up with the flurry of new tech is just one challenge. There are over 360 generative AI companies in 2023, according to CB Insights, and funding for the sector has increased fivefold in the first half of 2023 compared to the full year of 2022.
These companies offer generative AI solutions to revamp sometimes dreary marketing tasks like ad copywriting, generating ad visuals at scale and query-based campaign recommendations like media planning.
Take digital marketing agency Winclap, which received 40 new pitches for generative AI tools across various use cases in Q3 alone. After testing, the agency discarded over 60% of them due to issues like hallucinations, according to Leandro Santos, head of creative studio, Winclap.
Performance agency Tinuiti’s practice lead, emerging tech, Nirish Parsad, reviews every new generative AI pitch at least three times. Despite initial variations, subsequent evaluations reveal uncanny similarities in creative output across all demos, including similar product backgrounds and ad copy, regardless of the brand category.
For some agencies, demos appear effective in providing marketers with recommendations to improve campaign performance, but when the tools are queried multiple times, these tools often produce hallucinatory results.
“Over the past year, we’ve seen a shift from back-room testing and caution to a bit of a tech arms race to launch new AI features and new AI tools,” said Brian Yamada, chief innovation officer at agency VMLY&R. “Many of the tools come with a warning sign that they won’t always get things right and are looking for feedback to improve.”
Generative AI’s awkward reality
An agency executive who requested to speak on background, tested Microsoft’s GitHub Copilot to assist the agency’s engineers and developers. During these tests, when Copilot was tasked with generating code for a specific brand website function, it occasionally produced code directly derived from real-world examples on which it had been trained.
“It will give you somebody else’s code and that’s not good,” the executive said. “This is the issue with all generative AI tools including Dall-E.”
“GitHub Copilot generates unique code suggestions using context provided by the user. Our previous research shows that in rare situations (less than 1%), a suggestion may match code in GitHub’s public repositories, and these matches most often occur when a user has provided little or no context to Copilot,” a GitHub spokesperson told Adweek.
Meanwhile, Digitas is assessing a generative AI tool designed to assist in brainstorming and revising copy for copywriters. It found that the tool frequently generates generic copy and instead has added time to the copywriters’ workflow to verify the accuracy and validity of the generated content, said Digitas’ Buhler.
Human-AI teamwork
The reality of generative AI’s limitations highlights the need for ongoing training.
In a recent study conducted by technology company SOCi, polling over 300 digital marketers, 70% expressed feeling overwhelmed by the rapid evolution of AI and its integration into their marketing strategies, while 42% reported not receiving any formal training on AI and its marketing applications. To address this gap, some agencies are training employees on how to refine their prompts into these tools for desired outcomes.
Brainlabs, Media.Monks, Tinuiti, and Digitas are training their employees through prompt engineering, crafting inputs for generative AI tools to ensure the generation of optimal outputs.
“Generative AI will be a big part of our process to speed up the briefing and script development,” said Brainlabs’ Julia Amorim, global svp of creative. “It’s essential that our creative directors are well versed in that scale, and it becomes part of their overall toolkit.”
Meanwhile, Digitas uses an internal tool to screen generative AI outputs for plagiarism.
Still, if these AI systems become ubiquitous, there’s a looming danger of brands all having highly similar creativity, which leads to a homogenous brand landscape, said Tinuiti’s Parsad.
