What Forbes, Dotdash Meredith and BuzzFeed, among other publishers, are saying about AI by 2025

by Catherine the Wolf and Lu * January 28, 2025 *

Ivy Liu.

In the last year, publishers have explored how AI technology can streamline their operations. This is both to create internal workflow efficiency and to produce content for consumers. Publishers use AI internally for a variety of applications, including article classification, content recommendations, data analysis, and ad-targeting. Publishers have halted the use of generative AI for news articles when it comes to consumer-facing AI. They are experimenting with generative AI in search functions, article translating, quizzes, and games. Digiday+ Research interviewed executives from five publishers to learn about their current AI use and future investments. The executives included:

• Business Insider,
• BuzzFeed,
• Dotdash Meredith,
• Dow Jones,
• Forbes,

Instead, companies use LLM application programming APIs to create tools either in-house or through a partnership with a third party vendor. According to Digiday+ Research’s survey of 119 agency and brand professionals, cited in the report on how marketers use AI, most respondents said they primarily rely on third-party vendors to create AI tools. Digiday’s discussions with publishing executives revealed that many publishers work with third-party companies to build AI tools. Those with larger budgets, however, often use a combination of in-house vendors and third-party suppliers. According to executives, a company’s budget and size can have a major impact on whether or not building in-house software is an option.

• Jon Roberts chief innovation officer at Dotdash Meredith : The actual infrastructure of these LLMs are extremely expensive. Many LLMs are similar, so we live in a world of commodities. We don’t have to build commodity technology ourselves as a publisher. We want to use one of the best ones out there and then we can benefit from all the innovation that these companies create as they compete against each other… Next, you have second-tier companies who build tools on top of AIs but are not the creators of AI. It’s still early in the game, so many of our use-cases are ones that we have prototypes for. Ten startups will inevitably pop up to address that use-case. It’s a constant evaluation of which piece is specific enough for our use case to build it ourselves or, if speed is important for an edge, do we want a partner? The integration of technology into our user experience could ultimately benefit our users. In that sense, we have tried to do as much in-house as possible, especially when it comes to integrating AI APIs across our products. Jessica Probus is the publisher and general manager of BuzzFeed. She said: “Compared to other publishing houses, we probably lean more towards the in-house side of things.” It’s not because AI is involved, but rather because it’s the way we work. We have a team for internal work, a team for machine learning, a team for custom tech, and an innovation team. The innovation team has focused on AI more in the last four-year period. On the tech and editorial sides, we build most of our products using platforms and LLM APIs. We use LLMs that are provided by our partners to build models on top. This allows us to control both the output and the input. From a client’s perspective, we use a similar approach. We don’t force them to build a specific model. Instead, we offer several services to make it easier for clients to integrate our content with whatever model they already have. Microsoft Copilot was introduced. We have AI assistants to help us with our IT and help desk problems. We build most of the products that users use on forbes.com ourselves. We use LLMs – specifically, Google’s Gemini. The products we build on them are our own. “We know our audience better than anyone else, so it makes sense to build these journeys ourselves.”

How publishers are incorporating AI in their workflows

The executives Digiday spoke to told us that they are integrating AI into both internal and external workflows. While their use of AI for consumer-facing material varies, the internal use of AI appears to follow a consistent pattern aimed at reducing repetitive tasks. There was also a consensus across publications that using AI internally poses less risk than using it to create public content. Here are some ways publishers have told us that they use AI internally and externally.

• Forbes’ Supitskiy “We use tools for testing, such as Microsoft GitHub Copilot for our developers. There are transcripts in the newsroom. Our team may be working on images for augmentation. So, there’s a lot to implement, but it’s also a lot to do on a daily basis and make things more efficient.

• Dotdash Meredith Roberts: This is in many places: text, images, videos, research, and data analysis. We’ve looked at coding completion in depth when evaluating tools to help with repetitive coding. I think that the confidence that AI will make things faster and easier has proven to be true. The process is a bit more customized than initially thought, but there are still plenty of processes we are tackling in this manner.
• Hope for Business Insider: We segment into three buckets – internal use cases, where we improve workflows, or internal tools such as our CMS, or tasks that employees perform on a daily basis. The second bucket is based on the actual end-user product experience. Can we apply some of the advancements in generative AI to products to solve previously unsolvable problems? Or, we can provide users with a better experience than the previous generation of technology. The last bucket is macro-environment. This includes syndication, and how our content reaches readers. AI has a huge impact on the entire technology ecosystem. Google’s working methods are changing. They are adopting AI and this is changing the dynamics of how many publishers are perceived on search. Reddit and other social platforms are also trying new things. The third bucket is to be aware of the industry and its impact on us.
  

  AI with’strict oversight by humans’

  Publishers are at a crossroads in terms of incorporating AI into their external workflows. AI is not used in certain parts due to journalistic standards and ethical standards. This is especially true for news content. Digiday’s interviews with publishing executives for this report revealed that their publications don’t publish content entirely written by AI. While many publishers have strict guidelines about not publishing AI generated articles, others have begun creating consumer-facing content with the help of AI and under strict human supervision. Dow Jones, for instance, uses AI for article summaries and translation. BuzzFeed uses AI to create quizzes and other games. Dow Jones Verschuren : “One of the markets we did not operate in was South Korea. We’ve developed a assisted auto-translation that is translated using an LLM Model. It was efficient, because it was done with strict human supervision by machines. It allowed us to enter the market and offer people in Korea to read the news.” We launched Factiva Smart Summary recently. They can read the summary instead of reading through a bunch of articles. They can still go back to the original article.”

• Verschuren: “[A tool we tried] is called a JoannabotJoanna Stern, one of our columnists in the tech section, reviews every new iPhone that comes out. She wanted a chatbot that users could use to ask questions. She used her previous columns on iPhone reviews as a model. It was an interesting experiment. It showed a high level of engagement on the part of readers. Even though the model was built with a set of controlled content, if you so desired, it was still possible to break it. If you asked a question, the model wouldn’t hallucinate, but if it was broken, it could. Another example of why human supervision is so important.”

  The Wall Street Journal’s AI-powered Joannabot

  • Buzzfeed’s bully: “In terms of the company strategy, most of what we’re trying to do is build things that are consumer-facing. The reason is that we’re focused on using AI to do things that weren’t possible before. That’s our guiding principle for how we’re thinking about the applications. For example, one of the things we built very early on was an infinite quiz. We have BuzzFeed quizzes, and we plugged in the API to have an infinite number of answers for every quiz that we put through this framework. That’s not something that would have been possible without AI. You can have a huge team of editorial people and you’re not going to be able to do anything like that. That was the initial framework, and we’ve been building on that since then. A lot of the things we’re building are not just in our existing workflows. They’re creating new workflows that we didn’t have before.”
  • Probus: “We also have what we call ‘generators,’ where users can create images within an editorial framework. The one that people seem to resonate with is a Generator which turned celebrities into Shrek. This was an example of how we can involve our audience in the creation of content. This was a way to create something that was cocreated with our audience and editorial staff. Our thinking is different because we don’t just use AI. It’s about partnering humans with AI, whether they’re our users or editorial staff. You can now personalize recipes in the Tasty app. You could, for example, use the AI to increase the amount of garlic in every recipe by 10 times. You can customize it to your liking. Many [the applications] can be used to either create new content that didn’t exist previously or personalize existing content in a more engaging and infinite way.

    BuzzFeed’s Shrek-ify generator

    AI improves search functions for publishers

    As more AI applications become available to the public, consumers will now be able to find AI-powered searches to assist them with their online queries. Google has integrated AI into its search results in order to answer users’ questions and provide them with more information. Similar AI-powered search functions are being adopted by social media platforms. Publishers are also feeling pressure to improve their AI-search features, but not because it’s trendy. Forbes’ Supitskiy : “One goal we have is to [we] not use technology just because its cool. Be sure you’re solving the problem. We realized that search had not changed for a long time. We saw an opportunity to engage the people. Our approach[with Forbes] Let’s test the generative AI search tool Adelaide, powered by Google Cloud, on a small portion of the audience. We got some really good feedback from users about what they liked and disliked. We saw a significant increase in engagement, [people were spending] a 20% longer stay on the [our] website and [having] a four-fold increase in interaction with the page. People were not only getting answers, but they were also diving deeper into the topics, clicking on profile links, articles links, or asking a subsequent question. Search experiencein our website. This is the first time we have delivered a feature to Business Insider users that uses AI directly and visibly. We started by addressing a problem faced by many of our users, namely that our on-site searching was not very good. In the past, many people said that they could just use Google to search for your site. It wasn’t a big issue. For a long time Google and publishers had a symbiotic partnership, so there wasn’t much of a problem. Ironically, these dynamics are changing with the advent and use of AI, as well as Google’s changes in how it operates, including becoming more of a “walled garden” and introducing AI summarization, where users are less motivated to click through to a publisher. We can’t accept a sub-par on-site experience. We want to give our loyal readers the best experience possible. In many ways, AI-powered search is now almost a given. You’re seeing it in every major player.”

  Forbes’ and Business Insider’s AI-powered, on-site search engines

  Publishers’ use AI for advertising goes beyond ads

  A significant area where publishers have been utilizing AI is their ads businesses. Many publishers were prompted to look for alternative ID options by Google’s plan to sunset third party cookies in the Chrome web browser. Many publishers had already developed AI-driven ID options by the time Google announced it would no longer deprecate cookies on Chrome. Digiday spoke to some publishing executives who said that they were more satisfied with the results of the new alternatives. Other companies are using AI to classify articles, recommend content and analyze subscriptions.

  • Dotdash Meredith Roberts: Used correctly, AI keys words. It’s not a model for user profiling, but a model of language. We don’t train this model on historical cookie data. We train it on the content you read and tie it to it. We can build all sorts of metadata on the page, starting with language. By adding LLMs we get a higher resolution and higher accuracy. [For example,] Because the D/Cipher (19459071) tool that we built is now upgraded with OpenAI baked in. We know that people who plant seeds in their gardens are also interested long-term buy-and-hold accounts… and they are interested in traveling to Europe. We can make these inferences because we’ve combined all the data we collected from site readers. The extrapolation of user needs and wants without using any identity information, increases the aperture through which we can understand our users’ behavior in real time.
  • BuzzFeed’s Probus: We can scrape content and understand the relationships between different content behaviors, without needing any of the [private] information. We have a large amount of first-party information that we own because we have many logged-in and loyal users. This gives us an edge. I think we are seeing the opportunity to use AI to not only get around regulations, but also to find ways to not violate them.
  • The Hope of Business Insider: We’ve integrated AI into the CMS. We’ve developed a product called Saga React that takes the concept of AI generated tagging and allows for a richer classification of our articles. All of our content is run through an AI algorithm that gives us the sentiment. Is it optimistic versus pessimistic? Is it something an advertiser may not want to associate with?
  • Forbes Supitskiy : “We’ve been using AI as the core [part] [part][part][part][part][part]of our first-party[part]data platform, ForbesOne, for quite some time. It allows us analyze the audience, create segments and then connect clients and advertisers. But the platform has evolved into more than just a tool for advertising. Now it powers all recommendations on the site. It also has many propensity models like propensity for subscription. Will this user have a high tendency to subscribe? To register? To churn? We’ve developed that model to understand how to build a journey for our users and what to show them on Forbes.com.”

  What’s next?

  Digiday+ Research interviewed publishing executives to find out what excites them most about the future of AI. Here’s what the executives had to say regarding the expansion of AI for ad-targeting and how audiences have responded to various applications of this technology. Dow Jones’ Verschuren (19659052): I’m excited about the possibilities of multi-agents, and how to deploy them. Multi-agents will be helpful in the internal use case if they are combined with where there is a need for efficiency. There are not trivial scaling issues to be solved. The use case seems very simple, but the AI prompts can be quite expensive. The innovation is largely based on the infrastructure layer for efficiency and web scale. We’ve solved the problem for a Dotdash Meredith. We have extended this so that we are able to classify the open web premium, so we can understand user intent across all publishers. Because we believe there is a premium for people’s attention on the web which has not been captured in the old targeting based on cookies. There is a premium that can be built, with more value to advertisers and publishers. We can use it to ditch cookie-targeting and move to something more richer, real-time and higher fidelity.”

• Buzzfeed’s bully: “We’ll also begin to see a great deal more creative personalization.” This is what excites me the most. We have similar versions on the internet, where ads are tailored to you and follow you around. People want to be helped. The good publishers will be able combine that voice and expertise, but also better target and create engaging content that actually meet consumers’ needs, rather than spray and pray distribution to try to get your stuff to people who might not even care that thing.”

• Forbes Supitskiy : Gen AI gives us the capability to provide experiences to our audiences that they’re seeking. We want to give them the right content, in the best format, at the right time and encourage them to dig deeper. You might be interested in a summary, but you might also be interested in a specific topic. We want to be there for you and provide that content.

• Probus: “If you are just using AI as a novelty, people will not be interested. When you use AI to do something a human can do, there’s always a backlash. And a reasonable question of what the value is. When we can make something that is valuable and was not possible at the same scale or wasn’t easy or fun to do, people are less skeptical and more likely to say, “Oh, this actually is a good product.” In the last two-years, anyone willing to add AI to something to increase buzz has been able to see through it pretty quickly. This is how we think about it. If it’s a cool product [readers] doesn’t matter how you did it. Then we see people less sceptical and more interested in engaging.
• Hope for Business Insider: AI is following the adoption rate that you see when it comes to new technology. The adoption curve is usually very high, and then very low. This is because people become disillusioned or the technology does not evolve as quickly as people’s expectations. Some evidence suggests that we are heading in this direction with generative AI. With the adoption curve you reach this happy middle, where you are productive and find a good fit with the technology. You may not get everything you hoped for, but at least it is useful. AI will get there as well. The most valuable uses of AI aren’t necessarily going to be things that are super readily apparent and user facing.”

https://digiday.com/?p=566678