The State of AI: 2025’s guide for marketers

Another factor in the AI picture that will carry into the new year is the prospect of AI-generated or altered content being labeled as such. Marketers are hesitant to see “Made with AI” labels slapped across their creative campaigns, as the label doesn’t differentiate between content completely generated by AI or content in which AI tools were simply used to help during the creation process — despite the fact that those are two different things.

Sarah Mehler, co-founder and CEO at Left Field Labs, a creative tech agency with a focus on applied AI, said a good starting point for many agencies is to integrate internal applications for AI, while being mindful of how those applications are put into play. “Looking at internal frameworks for optimization and efficiency has been an entry point for us and many clients,” Mehler said. “So, first looking at best practices and frameworks. How does everybody use these [AI] tools in a consistent way that is also in alignment with security and privacy needs – compliance. It’s all very top of mind and highly magnified than any other tech tool.”

Across the board, marketers’ adoption of AI technology has steadily increased over the last three years. According to Glossy+ Research surveys conducted in 2022, 2023 and 2024, marketers’ use of AI increased from 44% of survey respondents in 2022 to 57% of respondents in 2023, and jumped again to a whopping 77% of respondents in 2024.

Nick Coronges, global CTO at R/GA, a global creative innovation company and digital agency, said the agency saw the potential use for AI in advertising a decade ago and began investing accordingly. “We launched R/GA Ventures about 10 years ago, where some of the first investments that we made were in AI companies,” Coronges said. “For example, we made an investment in a company called Reply.ai that did conversational interfaces using NLP before the ChatGPT explosion and generative AI. More recently, we invested in companies that were doing computer vision.”

Left Field Labs’ Mehler said that she expects the use of AI technology to keep growing at a rapid pace as marketers and the general public become more comfortable with its applications. “The cross-media synergy and the alchemy of these mediums, AI being a huge amplifier and enabler of that, is going to continue to advance in ways and speed I can only imagine at this point,” Mehler said. “When we think about cross-media from immersive, experiential, digital to physical, how does AI amplify and enhance each one of these things and really work as a strong driving component?”

Methodology

Glossy+ Research surveyed 119 brand and agency respondents about their current and upcoming AI investments and usages to map out marketers’ applications of the technology. Glossy+ Research also conducted individual interviews with marketing and technology executives responsible for AI investments and applications development. They included executives from:  

• Huge
• IPG Media Lab
• Left Field Labs
• M7 Innovations
• R/GA
• WPP

The majority of marketers continue to implement AI technology into their workflows by collaborating with third-party vendors, rather than building AI systems in-house. Since 2022, more than half of respondents to Glossy’s annual emerging technologies surveys have consistently said that they use a third-party vendor to build AI solutions — 53% of respondents said this in 2022, 62% of respondents said the same in 2023 and 58% of respondents said so in 2024.

Marketers’ reliance on third-party vendors to build AI solutions is logical given the high learning curve associated with building and implementing AI tech. Additionally, the cost of building AI solutions in-house, which generally includes establishing a team of workers dedicated to the task, is cost prohibitive for many companies. 

Matt Maher, founder of M7 Innovations, an independent research and development firm focused on media and technology, explained to Glossy that while many companies were initially eager to build their AI solutions in-house, the financial reality of doing so has tempered some of those early aspirations. 

“It was these grand ambitions upfront of feeling like we need to build [AI] in-house, and I think that’s because a lot of clients and brands viewed this wave of AI like the internet in 2000, going from zero to one,” Maher said. “[They believed] the outsized advantage would come in building it in-house, and creating your own advantage by having it inside your infrastructure. I think everyone quickly started to realize it is very different than the internet. We already have the internet, and it’s just a layer on top of it.” 

“The true headstarts lived in the Amazons, the Microsoft with the OpenAIs, Googles, the Metas,” Maher added. “They had such a massive headstart because they’ve been using this technology as part of their infrastructure. … To build it yourself, you need owned server farms, to have the expertise of data scientists. It was just an exorbitant cost. Instead, you could take this nice layer built by a tech giant and then build your custom layer on top, and still have the same benefits at an absolute fraction of the cost.”

Currently, the majority of marketers are using existing LLMs (large-language models) from major tech companies, such as Google’s Gemini, OpenAI’s GPT and Meta’s Llama, and they are using the tech companies’ APIs to create their AI applications, rather than building the LLMs or APIs themselves. However, even with the availability of major tech companies’ tools, many marketers still lack an internal tech team capable of utilizing those tools to build their AI applications. As a result, they often seek out third-party vendors or agencies to build their AI tools. 

Marc Maleh, global CTO at digital agency Huge, said one advantage of big tech companies providing open APIs attached to LLMs is that AI tech has become more readily available for agencies and their clients. “A lot of the big players built a suite of services. The beauty of a lot of them is they’re all API accessible. You no longer need to home-grow some of the models or the computing infrastructure to do certain things, but there’s a lot of room for innovation and products that agencies are building themselves,” Maleh said. “It’s API accessible now. We can leverage and build into tools and experiences that we’re using, not only for our talent but also in the way that we creatively think about work for our clients.”

Despite marketers’ overarching reliance on third-party vendors to build their AI applications, nearly a quarter of respondents to Glossy’s survey (23%) said in 2024 they are still using a combination of third-party vendors and their own in-house resources to build their AI solutions. 

Adam Simon, svp, managing director and U.S. head of innovation at IPG Media Lab, a research and strategy firm focused on emerging technologies, said that whether an advertiser exclusively uses a third-party vendor to build its AI solutions, or uses combination of a third-party vendor and in-house resources, depends on whether that advertiser is a brand or an agency. 

“On the brand side, it is mostly third-party, and the agency side is a little bit more of a blended approach,” Simon said. “Some brands have large technology teams, but agencies are increasingly becoming technology companies. So, agencies have bigger in-house technology and a ton of data.” 

“Thinking about the agency-brand relationship and what [agencies] can offer so that [brands] don’t have to go out and do that work themselves — in terms of in-house tools, we’ve been investing in building layers on top of broader, more general-purpose market tools,” Simon added. “And also building AI tools for our employees to use in-house as part of their brainstorming and activation.”

As marketers’ appetite for AI technology grows, the number of tech vendors devoted to AI solutions has continued to increase at a rapid pace as well. Forbes, this year, released its sixth annual “AI 50” list of global AI startups and noted that the number of applications it received for the 2024 list more than doubled from the prior year — 1,932 submissions in 2024 vs. 796 submissions in 2023. And, Forbes noted, the companies that made this year’s top 50 list jointly raised $34.7 billion in funding, underscoring the overall size of the AI tech market. 

As more AI vendors enter the market, advertisers may find themselves facing higher vendor prices in the not-too-distant future, however. “With marketing tech providers, it’s pretty safe to say they’re not going to be reducing their costs — especially as they started to create AI and other tools that … require a ton of processing power. That’s not cheap,” Gartner marketing analyst Mike Froggatt said when speaking about insight from a 2024 Gartner digital advertising report.

Generative AI continues its ascent in marketing

In 2023, chatbot technology was marketers’ most-used AI application. However, that all changed last year when copy generation overtook chatbots as marketers’ most common application of AI tech. In 2024, 65% of respondents to Glossy’s survey selected copy generation as the top NLP or AI technology their company uses, followed by chatbots at 55% of respondents. Image generation was marketers’ third-most common use of the technology with close to half of respondents (45%) selecting this AI application. 

It’s important to note here that all three applications are forms of generative AI. One of the main reasons for the shift in marketers’ usage of the applications could be because of recent customer trends. Consumers are now much more familiar than even a year ago with how sophisticated chatbots can be, especially with the rise in popularity of ChatGPT and other tools like Anthorpic’s Claude. Advancements in chatbot technology have given consumers a taste of what’s possible beyond older chatbots, such as the original Facebook-powered chatbots on websites, and consumers have increased their use of newer models like ChatGPT. 

M7 Innovation’s Maher said that, until recently, chatbots have suffered from negative consumer perceptions and a knowledge gap regarding the technology behind older chatbots and newer GPT models. “[The word] chatbot has this negative stigma or connotation,” Maher said. “There hasn’t always been a large language model [LLM] behind them. It was initially NLP, natural language processing, all these stacks-to-voice tech. Now, the LLM is the layer underneath that can power a lot of these chatbots.” 

“Because they have a stigma, when you think of chatbots on Facebook Messenger, they can only do three things and they get them all wrong,” Maher added. “But ChatGPT has this different feeling. It’s this thought partner. I can ask it all these things. There is that bifurcation of ‘I don’t want to build a chatbot for my business because people hate chatbots, but we seem to be okay with GPT and Google AI.’” 

Maher noted that brands that are resistant to incorporating chatbots into their businesses may want to reconsider, as the technology will only continue evolving. “[Brands] don’t want to build a chatbot in 2024. But at the same time, I’d argue now is the perfect time if you do it the right way,” Maher said. “We talk a lot about large language models, but I think what we’ll soon start to talk about more is not LLMs, but SLMs, small language models.”

Brands and agencies are also focusing on other areas in which AI applications can have a greater impact. Two areas in particular that have seen recent growth in marketer usage are copy and image generation — the applications increased in usage year over year between 2023 and 2024 by 22 percentage points and 26 percentage points respectively. 

However, despite the popularity of generative AI, not all consumers have reacted well to some audience-facing applications of the technology. Coca-Cola’s recent use of generative AI to assist in the creation of its 2024 holiday ad is one example. The beverage company used an AI image generation application to create portions of its “Holidays are Coming” film, along with a human production team. However the final commercial faced consumer pushback. 

“Reaction is polarized,” Pratik Thakar, global head of generative AI at The Coca-Cola Company, said in a November interview with Glossy. “What we feel is, it’s one sector of people, they immediately jumped into saying that we are just creating something very quick and it’s like pressing the button and the ad comes out vs. how much work goes in.”  

“And that’s why, this type of conversation, we need to do more,” Thakar added. “The way we are all using CGI and Adobe, the same way people will be using AI, and there’s nothing wrong. As a company, as I mentioned, ethical prompting, right balance of humans driving the tool. And AI is a tool. It’s a tool made for humans.”

Despite the consumer pushback on Coca-Cola’s ad, industry experts still see a bright future for generative AI’s use in video production, especially for its potential to eliminate repetitive workflows. R/GA’s Coronges said the digital agency has incorporated generative AI into its video creation process. “The most interesting thing is you still have a creative team architecting the story, but you’re using AI applications like [those from AI application company] Runway to bring content into your story,” Coronges said. 

“Both of these things will happen in parallel,” he added. “You’ll see AI operations that are getting the basics, making sure that you’re not wasting time — for example, doing storyboards by hand when your goal is to get a very quick representation of your idea in front of a client. … Those are interesting for efficiency and to get people’s heads out of grunt work and into what they’re actually trying to do.”

While several generative AI applications in Glossy’s survey saw increased usage in 2024, social media listening saw a 29 percentage point decrease in usage since 2022. This was a surprising change considering one of the main tasks AI applications are capable of is sifting through enormous data sets. 

Eric Lee, CTO of Left Field Labs, said marketers’ use of social media listening applications could be decreasing because of instances in which generative AI programs essentially wind up talking to each other on social platforms. “We have seen a lot of discussion around when AI is putting out content on social channels and you have other AI that is listening and responding to that content,” Lee said. “The quality of the conversation in general – what is the value to a brand when the engagement is purely about volume, vs. when you actually have people who are doing it thoughtfully and with more of a targeted direction?”

While Glossy’s survey results showed a decrease in marketers’ use of social media listening applications, not all industry experts have observed the same trend. R/GA’s Coronges explained that the decline Glossy noted could be attributed to a change in users’ definition of AI and an increasing comfort level with the technology. “I don’t think you’re seeing a decline in AI being used for data analysis,” Coronges said. “It’s probably just the semantics of what AI is. AI basically means the new things you can do with software. That’s what people are using the term AI to mean. So, all the tech that is machine learning, is just now part of data. We no longer call it AI.” 

“What AI now means is generative AI,” Coronges added. “Which is, neural nets, LLMs, image models and transformers. What [data] Instagram or TikTok uses to determine pieces of content you see next is data analysis and that’s a more traditional sort of machine learning.”

Marketers weigh different margins of error when using AI for internal vs. external content

The shift in marketers’ primary usage of generative AI from building chatbots to generating copy and images, as seen in Glossy’s survey results, could be an indication of an industry-wide trend of brands and agencies moving away from using external-facing applications of AI to using internal ones. Chatbots typically interact with users on an external, customer-facing side. Copy and image generation is generally used internally to test and create content, or to speed up processes within a workflow. This shift in marketers’ focus is particularly noteworthy as using AI for copy and image generation potentially has different use cases than chatbots. 

Stephan Pretorius, CTO at agency company WPP, whose holdings range from creative ad agencies like Oglivy to media agency networks like GroupM, said deciding where in the workflow to place AI applications is an ongoing process. “It’s a fundamental question in terms of the interaction of humans with AI,” Pretorius said. “What is the right balance of autonomy vs. interaction, and how do you know which things to push people to do vs. what things do you push machines to do in a more automated way?”

Pretorius said WPP has instituted a guiding principle, called the “empathy gradient,” that informs its product strategy. “Basically what it is, on one end you’ve got the need for low empathy, but higher automation, and vice versa. When you think about all tasks as being somewhere on that empathy gradient, it helps you make decisions about what you automate vs. what not,” Pretorius said. “In general, we’re trying to push down and automate as many of the menial, repetitive, nonjudgment-requiring tasks and spend more human time on the things where we need judgment, critical thinking and discernment. At the same time, we want to empower people to make some of those decisions themselves.”

Marketers’ interest in using AI to automate repetitive tasks is reflected in Glossy’s survey results. The percentage of survey respondents who said they use AI to generate editorial or consumer-facing copy decreased by 6 percentage points in 2024 vs. 2023. Conversely, the amount of respondents who said they use AI to generate sales communication copy and internal or operational copy, which many would consider to fall into the category of repetitive tasks, increased by 24 percentage points and 21 percentage points respectively year over year.

One type of AI application that some companies and agencies are using internally is voice-to-text software, particularly for translation services. M7 Innovation’s Maher said he uses voice-to-text translation when interacting with clients who may not be native English speakers. “I’m constantly talking about very complex technologies, often to people who don’t speak English as a first language,” Maher said. “So, if I’m going into a deep-dive on blockchain, Web3 or generative AI, I will have that translated into the person’s native language – a lightweight touch. Maybe the translation isn’t perfect, but it is a step change. It’s better than me just speaking in English.” 

Maher noted that there is a difference in how his company uses AI applications for B2B and B2C clients. “With B2B, I see more leniency with things that you can do [with AI],” Maher said. “Whereas, in B2C, brands are careful about what they put out to consumers because that artifact, especially in social [media], can get clamped onto. With B2B, there’s a larger swath of ability to try out different things to see if they work, because a lot of B2B relationships are always looking for those efficiencies.”

As marketers work to find the right fit for AI in their workflows, it will be important for them to consider whether functions are internal- or external-facing. Generative AI content produced for internal purposes typically can have a wider margin of error, whereas content that consumers will see would be held to a much higher standard. 

Huge’s Maleh said brands have to think about how much risk they’re willing to take when using AI externally. “I would say [room for mistakes] is not very high, generally speaking,” Maleh said. “I’ll put it into three buckets. Bucket one is brand. Is the asset on-brand? That is incredibly important. [For example,] it’s great that I can use gen AI to make an image of a watch, but if that image is put out on Instagram and it’s not an actual watch that I can go buy in the store, the internet is going to blow them up for that. That’s the zero margin for error. Now, if I use a photograph of a watch, but the background is gen AI, and I have 50 permutations of that background, I have a little more margin of error and that’s a little more acceptable.” 

“Bucket two is [the] legality,” Maher continued. “Do I have a framework in place from a legal perspective that is compliant to use a certain data set? If I’m in Europe, that is different than in the U.S. to use AI for personalization. The third bucket is the ethical side. Are you using ethical frameworks? And there’s a difference between legality and ethics, but the ethical framework side of it is, are you doing something that is bolstering a bias.”

Industry experts weigh in on the future of AI

Glossy+ Research asked the industry experts in our focus group about AI trends to keep an eye on for the future. Here’s what they had to say on topics including agentic AI, custom AI models, using AI for video generation and upcoming hurdles in AI tech development.

Several executives we interviewed brought up agentic AI, which is essentially an AI solution that can autonomously complete tasks. Agentic AI differs from tools like chatbots in that it has the ability to solve complex problems, such as multiple-step data analytics or goal setting and completion, by designing its own workflow and using available tools. 

• “Agentic AI has also been one of those hot topics for the past few months. We are going to see a lot of AI agents who [a user] can give a broader set of directions [to] and have them complete more complex tasks on your behalf. The coming years will be important for that.” — Eric Lee, CTO at Left Field Labs

• “We are building in WPP these super agents, which are effectively [AI] agents that are experts on an entire domain of knowledge. Those are slightly more complex because you have enormous data sets that you need to process in order to build those. You need lots of domain expertise, and very often, it’s not just a question of chucking a whole bunch of PDFs into the context window. You need to actually think about the logical structure of the domain. Is there a master theory or framework that becomes the starting point through which the examples get evaluated?” — Stephan Pretorius, CTO at WPP 

Some industry experts in Glossy’s focus group said increasing reliance on custom AI models is another trend that is on the rise. Custom AI models are applications that are trained on a specific set of data, typically proprietary data. These AI models are built for specific tasks and brands and, as a result, can provide much more customized outputs. But they come with some downsides. 

• “A few things in 2025 will become more mainstream — one is custom models. Brands need control. ChatGPT, Midjourney and LLMs out of the box don’t really help tier-one brands deliver experiences for end users, because they need to control the brand identity, the voice, the behaviors, the interactions. Custom models are going to be a step in that direction, [and] that’s going to be custom language models, image models and video models that Runway and others are going to start to introduce.” — Nick Coronges, global CTO at R/GA

• “Part of the challenge with some of gen AI tooling in the past year and a half is the cost to create custom models. Amazon is now the most cost-effective platform out there because they own the data centers to do these things. I hate using the phrase ‘race to the bottom’ but to a degree, it is. [Amazon] has the computing power. Does the technology match the price? That’s going to be one data point. If I’m looking at a gen AI tool from company A vs. company B, one of them might be more cost effective, but if the output of that gen AI tool is not great or on brand, then it doesn’t matter that it’s more cost effective.” — Marc Maleh, global CTO at Huge

As marketers’ adoption of generative AI increases, so do the possibilities for future ad-focused uses of the tech. According to Glossy+ Research’s survey, copy and image generation are currently among marketers’ top uses of generative AI. But there’s another noteworthy use: video generation. 

• “The one thing that we are still going to get a lot more development on is this whole idea of creative technology and how generative AI is going to develop and get much closer to things like CGI and VFX. You’re going to see a lot more AI for creative execution and ideation, feature film grade, everything from relighting to rotoscoping. All of these things are going to become part of the filmmakers’ storytelling toolkit.” — Stephan Pretorius, CTO at WPP

• “A big behavioral one for us that we’re pushing is synthetic media — media that has had some type of generative AI influence, but not full-on AI made an entire film. We thought synthetic media would have an adverse reaction, like, ‘oh, that’s AI generated, get that out of there.’ What we found is there’s the synthetic media sweet spot. … For example, I am watching ‘Harry Potter,’ and I’d love Timothée Chalamet to be the lead. What would that be like? Or, my kids are going to watch ‘Succession,’ so clean the expletives up. That level of manipulation, and that’s a strong word, people will seek out. The big trend to look for in the future is people wanting synthetic media, and how do we live in a world where there is still human craftsmanship in media but there’s also space for synthetic media and the appetite there.” — Matt Maher, founder of M7 Innovations

Glossy+ Research has been reporting on AI and its various uses since 2017. Over the years, we’ve had the opportunity to check in with IPG Media Lab’s svp, managing director and U.S. head of innovation Adam Simon to gather his insights on the topic. When we spoke with Simon in 2017, he predicted the early stages of generative AI’s evolution, five years prior to ChatGPT’s 2022 release, and noted that not enough companies at the time were investing in AI. In 2022, Simon predicted that AI’s usefulness could be limited by the capacity to have smoothly flowing dialogue, which has played out over the past year as marketers have moved away from using traditional chatbot technology. And this year, Simon told us about upcoming hurdles in AI tech development. Here are his comments year by year:

• “Not enough brands are investing in personalization and AI or machine learning. A lot of brands are already close to the consumer and have a lot of consumer data. We’ve seen a lot of industries disrupted by direct consumer sales that we never thought were going to do direct-to-consumer. That’s going to become really important over the next 10 years.” — Simon in 2017

• “We’re finding that NLP is not advancing as fast as we want it to with things like voice assistants. Our ability to have natural conversations with them really is happening very incrementally. It seems clear that it’s going to be probably another decade before you can just start talking to Google or Siri or or Alexa the way that you would talk to a human.” — Simon in 2022

• “On a technological level, even the most advanced models are starting to hit a wall. There are some open questions about productization and long-term use of some of these models, and technological breakthroughs to allow them to keep getting better. There’s still work to be done, even if the technology around LLMs hits that ceiling in the next year or so, to take it out of just being an empty text box and push it into products. Natural language processing is a great use case. In 2025, Amazon will launch their LLM-powered Alexa. We’re looking at 2026 for Apple’s LLM-powered Siri. There’s still a roadmap for how these technologies will be integrated into products that consumers are using every day. That will continue to change and evolve even if we do sort of hit that ceiling for technological and content improvements around LLMs.” — Simon in 2024