GREENTECH

How the generative A.I. boom could forever change online advertising

In this article

Sebastien Bozon | AFP | Getty Images
Shortly after ChatGPT hit the market last year and instantly captured headlines for its ability to appear human in answering user queries, digital marketing veteran Shane Rasnak began experimenting.

As someone who had built a career in creating online ad campaigns for clients, Rasnak saw how generative artificial intelligence could transform his industry. Whether it was coming up with headlines for Facebook ads or short blurbs of ad copy, Rasnak said, jobs that would have taken him 30 minutes to an hour are now 15-minute projects.

And that’s just the beginning.

Rasnak is also playing with generative AI tools such as Midjourney, which turns text-based prompts into images, as he tries to dream up compelling visuals to accompany Facebook ads. The software is particularly handy for someone without a graphic design background, Rasnak said, and can help alongside popular graphic-editing tools from Canva and Adobe’s Photoshop.

While it’s all still brand new, Rasnak said generative AI is “like the advent of social media” in terms of its impact on the digital ad industry. Facebook and Twitter made it possible for advertisers to target consumers based on their likes, friends and interests, and generative AI now gives them the ability to create tailored messaging and visuals in building and polishing campaigns.

“In terms of how we market our work, the output, the quality and the volume that they’re able to put out, and how personalized you can get as a result of that, that just completely changes everything,” Rasnak said.

Rasnak is far from alone on the hype train.

Meta, Alphabet and Amazon, the leaders in online advertising, are all betting generative AI will eventually be core to their businesses. They’ve each recently debuted products or announced plans to develop various tools to help companies more easily create messages, images and even videos for their respective platforms.

Their products are mostly still in trial phases and, in some cases, have been criticized for being rushed to market, but ad experts told CNBC that, taken as a whole, generative AI represents the next logical step in targeted online advertising.

“This is going to have a seismic impact on digital advertising,” said Cristina Lawrence, executive vice president of consumer and content experience at Razorfish, a digital marketing agency that’s part of the ad giant Publicis Groupe.

In May, Meta announced its AI Sandbox testing suite for companies to more easily use generative AI software to create background images and experiment with different advertising copy. The company also introduced updates to its Meta Advantage service, which uses machine learning to improve the efficiency of ads running on its various social apps.

Meta has been pitching the Advantage suite as a way for companies to get better performance from their campaigns after Apple’s 2021 iOS privacy update limited their ability to track users across the internet.

‘Personalization at scale’

As these new offerings improve over time, a bicycle company, for example, could theoretically target Facebook users in Utah by showing AI-generated graphics of people cycling through desert canyons, while users in San Francisco could be shown cyclists cruising over the Golden Gate Bridge, ad experts predict. The text of the ad could be tailored based on the person’s age and interests.

“You can be using it for that sort of personalization at scale,” Lawrence said.

Meta’s Advantage service has been gaining traction with retailers using it for automated shopping ads, according to data shared with CNBC by online marketing firm Varos.

In May 2023, roughly 2,100 companies spent $47 million, or about 27.5% of their combined total monthly Meta advertising budgets on Advantage+, the Varos data showed. A month earlier, those companies directed 26.6% of their budget, or $44.9 million, to Advantage+.

Last August, when Meta formally debuted its Advantage+ automated shopping ads, companies put less than 1% of their Meta ad spend into the offering.

Meta Platforms CEO Mark Zuckerberg speaks at Georgetown University in Washington, Oct. 17, 2019.
Andrew Caballero-Reynolds | AFP | Getty Images

Varos CEO Yarden Shaked said the increase shows Facebook is having some success in persuading advertisers to rely on its automated ad technology. However, Shaked said he’s “not sold on the creative piece yet,” regarding Meta’s nascent foray into providing generative AI tools for advertisers.

Similarly, Rasnak said Midjourney’s tool isn’t “quite there yet” when it comes to producing realistic imagery that could be incorporated into an online ad, but is effective at generating “cartoony designs” that resonate with some smaller clients.

Jay Pattisall, an analyst at Forrester, said several major hurdles prevent generative AI from having a major immediate impact on the online ad industry.

One is brand safety. Companies are uncomfortable outsourcing campaigns to generative AI, which can generate visuals and phrases that reflect certain biases or are otherwise offensive and can be inaccurate.

Earlier this year, Bloomberg News found that AI-created imagery from the popular Stable Diffusion tool produced visuals that reflected a number of stereotypes, generating images of people with darker skin tones when fed prompts such as “fast-food worker” or “social worker” and associating lighter skin tones with high-paying jobs.

There are also potential legal issues when it comes to using generative AI powered by models trained on data that’s “scraped from the internet,” Pattisall said. Reddit, Twitter and Stack Overflow have said they will charge AI companies for use of the mounds of data on their platforms.

Scott McKelvey, a longtime marketing writer and consultant, cited other limitations surrounding the quality of the output. Based on his limited experience with ChatGPT, the AI chatbot created by OpenAI, McKelvey said the technology fails to produce the kind of long-form content that companies could find useful as promotional copy.

“It can provide fairly generic content, pulling from information that’s already out there,” McKelvey said. “But there’s no distinctive voice or point of view, and while some tools claim to be able to learn your brand voice based on your prompts and your inputs, I haven’t seen that yet.”

An OpenAI spokesperson declined to comment.

A spokesperson for Meta said in an email that the company has done extensive research to try to mitigate bias in its AI systems. Additionally, the company said it has brand-safety tools intended to give advertisers more control over where their ads appear online and it will remove any AI-generated content that’s in violation of its rules.

“We are actively monitoring any new trends in AI-generated content,” the email said. “If the substance of the content, regardless of its creation mechanism, violates our Community Standards or Ads Standards, we remove the content. We are in the process of reviewing our public-facing policies to ensure that this standard is clear.”

The Meta spokesperson added that as new chatbots and other automated tools come to market, “the industry will need to find ways to meet novel challenges for responsible deployment of AI in production” and “Meta intends to remain at the forefront of that work.”

Stacy Reed, an online advertising and Facebook ads consultant, is currently incorporating generative AI into her daily work. She’s using the software to come up with variations of Facebook advertising headlines and short copy, and said it’s been helpful in a world where it’s more difficult to track users online.

Reed described generative AI as a good “starting point,” but said companies and marketers still need to hone their own brand messaging strategy and not rely on generic content. Generative AI doesn’t “think” like a human strategist when producing content and often relies on a series of prompts to refine the text, she explained.  

Thus, companies shouldn’t simply rely on the technology to do the big picture thinking of knowing what themes resonate with different audiences or how to execute major campaigns across multiple platforms.

“I’m dealing with large brands that are struggling, because they’ve been so disconnected from the average customer that they’re no longer speaking their language,” Reed said.

For now, major ad agencies and big companies are using generative AI mostly for pilot projects while waiting for the technology to develop, industry experts said.

Earlier this year, Mint Mobile aired an ad featuring actor and co-owner Ryan Reynolds reading a script that he said was generated from ChatGPT. He asked the program to write the ad in his voice and use a joke, a curse word and to let the audience know that the promotion is still going.

After reading the AI-created text, Reynolds said, “That is mildly terrifying, but compelling.”

Watch: Social media showdown: Instagram to launch direct competitor to Twitter