How PR is moving from experimenting to implementing with generative AI

Firms are no longer just testing the waters — they’re upping their investments, preparing for disclosure and copyright issues and future-proofing comms skills.

Coca-Cola’s Masterpiece campaign uses generative AI to have paintings come to life and pass around a bottle of Coke. (Photo credit: The Coca-Cola Company, used with permission)

PR pros are sensing a shift in the air as questions around generative AI change from “How does this work?”’ to “How are we going to use this to change our business?”

Publicis Groupe, in an unprecedented move, revealed its Q4 and full-year earnings two weeks early as it announced a $326 million investment in AI over the next three years.

Arthur Sadoun, CEO; Carla Serrano, chief strategy officer; Sam Levine Archer, chief solutions officer; Dave Penski, CEO of Publicis Media and Nigel Vaz, CEO of Publicis Sapient all appeared in an hour-long video to tell investors about CoreAI, a new tool that will combine trillions of data points to quicken content output.

Meanwhile, analysts are calling 2024 the year of AI implementation.

The PR industry is antsy about AI — both from excitement to put things out into the world faster than ever before and from nervousness about automation replacing human workers, technology exacerbating racial disparities, untested legalities surrounding copyright and the spread of misinforming propaganda.

AI spend on comms and marketing  

All signs point to holding companies cranking up their investments in AI across both PR and advertising. According to Forrester’s 2024 prediction report, investments in AI platforms will triple while generative AI software sees a 36% compound annual growth rate from 2023 to 2030. On January 25, Publicis Groupe made its big move.

“It’s not surprising to hear that,” says Nicole Greene, VP and analyst at consulting firm Gartner. “One of the big things that’s been happening in the last few months is announcements from a lot of holding companies and larger agencies about partnerships with technology providers and brands to bring forward the creative capabilities of this technology.”

Analysts can’t predict just how much of these holding company-level investments PR firms will see. After Publicis Groupe’s recent announcement of a €300 million investment in AI over the next three years, MSL for 2024 alone “anticipates an investment of €100 million, with 50% on technology through licenses, IT software and cloud infrastructure,” according to an MSL/Publicis Groupe spokesperson.

However, 66% of PR executives expect generative AI to have a moderate to significant impact on communications, according to Gartner’s 2023 Communications Future Strategic Priorities Poll. And 94% think content creation is the comms function generative AI will impact the most. 

Gartner also predicts that, by 2036, organizations using AI will increase spending on products, services and hybrid offerings using the tech to more than $50 trillion.


Guideline-setting to prepare for business overhauls

Highwire PR created a framework for safely integrating generative AI into PR to guide the industry through these concerns in August.

The guide touches on four pillars PR pros should consider before using AI: legal and contractual issues, copyright and ownership, ethics and quality of work. It delves into 14 topics within those pillars including regulation and confidentiality of client information, registration and plagiarism, bias, discrimination and originality.

It’s designed to give employees and clients confidence the firm is using AI properly and with consistency. It also includes risk-assessment guidelines for common scenarios to help employees mitigate AI misuse.

If firms walk away from the guide understanding any one thing, James Holland, EVP of digital at Highwire and author of the framework, wants it to be that they must understand the legal risk they’re posing to clients when using AI.

“You’ve got to think for the client, and not just for yourself or the agency,” he says. “There are several layers of ownership attached to anything we make.”

For example, a track called Heart on My Sleeve containing AI-generated vocals that sound like Drake and The Weeknd went viral in April. Spotify removed the track as Universal Music Group was able to copyright claim a sample at the start of the track despite not owning the original track.

The New York Times also sued OpenAI and Microsoft late last year over claims the companies are profiting from stolen copyrighted content used to train their large language models. OpenAI has set up licensing deals with publishers such as the Associated Press, but the case is expected to set precedents about how generative AI is allowed to interact with copyrighted written material.

The PR Council created the first iteration of its guidelines for using generative AI in April. The guide initially focused on text, given that most businesses were interested in chatbots. In February, the PR agency trade body updated its guidelines to include feedback from chief creative officers who are dabbling in multi-modal prompting, which is the process of feeding the same AI model datasets from text, images and audio.

“We heard a ton of enthusiasm from creatives,” says Kim Sample, president of the PR Council. “There’s this whole new flexibility to create visuals using generative AI.”

As creatives find new ways to put AI-generated content in front of clients and consumers, PR pros need to be on top of making sure the source of that content is properly disclosed, Sample advises.

“We believe content that is AI-generated, text or visual, should be disclosed to the client,” she adds. “We don’t think you can present work that’s 100% generated by AI. Even when you’re enhancing, there needs to be a level of disclosure to build trust.”

Comms tools that revolve around AI

Comms vendors such as Signal AI aren’t just integrating the technology into their offerings — they’re revolving their entire suites around it.

Signal AI, in particular, sells clients data analysis tools that rely on discriminative AI, which is a model used for classifying content. That tech scours licensed media partners’ articles and podcasts as well as social media to find common through lines in how companies communicate with the public.

For example, if Nike wanted to promote an innovation effort, Signal AI could look across industries to see how Pfizer has talked about breakthrough vaccines and extract lessons from that messaging.

“We’re flexible and deep enough with this huge dataset that we can look at innovation in pharma and biotech and map the subtopics within vaccines, supply chains and advanced manufacturing for molecules,” says Dan Gaynor, head of strategic solutions at Signal AI. “We can then provide that same level of depth for Nike that may want to look at carbon fiber running shoes, marathon times and athletic performance.”

Signal AI’s other flagship service is reputational threat analysis. The vendor trains its discriminative AI to recognize people, companies, industries and related entities in all three categories to analyze sentiment around a topic. It then combines that knowledge with an ability to recognize how often an entity has appeared online and in what context to determine if weighing in on a topic will provide that entity a reputational boost or risk.

For example, Signal AI worked with a major streaming platform to see if global expansion announcements were risky given its relevance to the SAG-AFTRA strike.

Most comms vendors are using AI for this kind of media monitoring. In October, Muck Rack began offering PR pros lists of recommended journalists based on their current list. According to Muck Rack’s blog post, communicators added 36% of journalists that it recommended to their media lists. Stagwell’s PRophet offers influencer discovery and campaign management, media alerts and press release distribution powered by AI.


How to prompt

“A good prompt is the one that returns what you want,” says Kyle Monson, cofounder of WE’s tech firm Codeword.

Monson’s best prompts range from “summarize this article from The New York Times” to “what are some of the top smartphone innovations of the past decade” — an insight he found necessary before working for a smartphone client. His best advice is to start with a general prompt, then get more specific with an additional one or two follow-ups rather than being as specific as possible on the first attempt.

But as language learning models get better at understanding the natural way humans communicate, Monson says there’s no need to be an expert prompting engineer.

“There’s a lot of talk about prompt engineering and learning to speak the language of LLMs, and a lot of that is bunk and our industry is making something hard so we can claim specialty in it,” he says. “Part of our job is making everything seem really hard so our clients know they need us.”

He added that most PR practitioners can put in simple requests such as asking for the top AI reporters at big publications and will receive satisfactory answers, so long as they’ve already put in some work to familiarize themselves with that industry.

Lawrence Edmondson, founder of tech consultancy Eat Big Digital, agreed prompt engineering likely won’t be sought after in a few years. However, he says it’s worth learning as it’s highly sought after right now.

“Prompt engineering is something I would encourage folks who may not have been trained in data science, especially Black and brown people who may not have a college degree or trained in computer science, data science or data engineering, to get into that field,” he said. “For now, it’s super hot.”

Macaela VanderMost, founder of Newfangled Studios, recommends going through several prompts before being satisfied with an answer.

“Treat it like a brainstorm,” she says. “Spend a good half-hour or 45 minutes talking to that AI before you feel like you’ve got the answer.”

She also recommends taking advantage of multi-modal prompting to learn how AI talks. By asking a chatbot to describe an image, she’s gotten a good sense of how AI communicates.

Generative AI as a creative tool

Despite generative AI looming over every conversation the marketing industry has, most creatives can count the number of AI-generated campaigns they know on one hand.

Coca-Cola’s Masterpiece campaign stands out as the leading example of what generative AI can do to augment creativity. The ad depicts students sketching art in a gallery before several paintings come to life and pass around a bottle of Coke, courtesy of AI.

“It’s pretty creative and the quality of the work is really high,” says Michael Olaye, SVP and managing director of strategy and innovation at R/GA. “The hours and money they saved being able to create with the tools they had is pretty cool.”

In a somewhat controversial move, Coca-Cola released an NFT collection based on the artwork in August.

For the 2023 Lunar New Year, McDonald’s released the first ad to use an AI-powered piece of technology called neural radiance fields to create 3D scenes.

Other examples of AI-generated campaigns include Mint Mobile’s ad written by ChatGPT, Virgin Voyages’ digital recreation of Jennifer Lopez in Jen AI. and Audi’s Q8 E-TRON spot that translated the feeling of driving the vehicle into AI-generated art.

@cocorocha The first car I ever owned was an Audi. Since then, I’ve always been drawn to the way the brand embodies the intersection of design, technology and performance. So when @audicanada invited me to be among the first to drive their new fully electric Audi Q8 e-tron, while using powerful AI to track my feelings and emotions throughout the experience, I was excited and intrigued by the concept. In a world full of influencers, the transparency of revealing my genuine emotions on screen makes a progressive statement about how we can share more authentic reactions and connections. As someone who has strived to push the envelope and redefine what it means to be a model in the digital era, I’m honoured that Audi has included me as part of its new campaign. Look out for more on our partnership and the spectacular A.I. art generated by my personal experience with the new Audi Q8 e-tron. #ProgressYouCanFeel #ad #AudiCanada ♬ original sound - Coco Rocha

Most consumer-facing creative work has either relied on ChatGPT-generated copy or has used AI to feature trippy and surreal visuals. In time, AI experts and creatives expect it will become harder to tell what’s been made using AI, for better or worse.

Brands have also taken a stance against AI in their campaigns. In March, Dove rejected the popular Bold Glamour filter that had been downloaded on TikTok more than 16 million times, stating that it set unrealistic beauty standards and distorted young people’s self-image.

@dove No filter should tell you how to look. 80% of girls are already using filters by the age of 13. It’s no wonder their perception of beauty and their self-esteem are distorted. Help reverse the damage. #TurnYourBack on the Bold Glamour filter and digital distortion. Real beauty is bold. #RealBeautyIsBold #Dove #LetsChangeBeauty #NoDigitalDistortion #BeautyCommunity #SelfEsteem ♬ original sound - dove

Beauty brands have often used AI to recommend consumers improve their appearance. At CES, L’Oréal showcased an AI-powered adviser called Beauty Genius that will chat with a person and scan their face before recommending products to tighten their skin or get rid of the bags under their eyes.

Preparation for bias

In December, McKinsey published an article detailing how generative AI stands to disproportionately impact Black workers by replacing jobs they commonly hold with automation. While the article largely focused on gateway work that serves as a stepping stone to a higher-paying career, generative AI stands to exacerbate existing racial disparities within the comms and marketing industries as well.

More than 80% of PR practitioners identified as white in 2023, according to data from the Bureau of Labor Statistics. With so few Black people joining the industry already, an increased reliance on technology not being built for nor taught to potential or current workers stands to serve as another barrier to entry.

“What type of language are we going to use to inform this community about what is coming?” says Karen Baker, founder and president of marketing agency Boathouse. “The Black community is very uncomfortable with generative AI and what it’s going to do.”

To get that community more comfortable with using generative AI, businesses need to tap their PR teams to create awareness campaigns around the benefits AI can provide, Baker says.

For example, AI can make up for a lack of resources available to Black entrepreneurs, says R/GA’s Olaye.

“As a Black entrepreneur myself, you want to have technology that enables you to bypass any system that has systemic racism or bias in it,” he adds. “I hope as more agencies are created by Black entrepreneurs, technology such as AI enables them to match or sometimes even outperform the industries or companies they can’t get into.”

AI glossary

Chatbot — An AI program that converses back and forth with users.

Deepfake — A fake visual produced by AI that’s intended to look real.

Generative AI — A form of AI that has been fed text, audio or visuals and can create its own media based on that knowledge.

Hallucination — An AI-generated response that’s inaccurate. Hallucinations are often presented as factually correct.

Large language model — An algorithm that can generate, summarize and translate human-sounding language.

Multi-modal — An AI model that can produce and receive text, images, audio and video. For instance, a multi-modal model can describe an image it’s given with a text response.

Natural language processing — The technology that allows AI models to understand and replicate the natural way humans communicate.

Prompt — The text a user puts into a chatbot or image generator to get results.

Prompt engineering — The study of feeding an AI model optimal prompts to get the fastest, most accurate response.


Have you registered with us yet?

Register now to enjoy more articles and free email bulletins

Register
Already registered?
Sign in

Recommended for you

Recommended for you

Explore further