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

Over the past year, publishers have been exploring how AI technology can help them streamline their operations, both to create internal workflow efficiencies and to produce consumer-facing content. Internally, publishers are using AI for a range of applications, from article classification and content recommendations to data analysis and ad targeting. When it comes to consumer-facing uses of AI, publishers have put the brakes on using generative AI to write news articles. However they are experimenting with generative AI for search functions, article translation, and quizzes and games. 

As media organizations’ adoption of AI is only expected to continue growing, Digiday+ Research spoke with executives at five publishers about their current use of AI and future investments in the technology. They included executives from:

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

Overall, most companies that incorporate AI into their business practices do not build their own large language models (LLMs), with the exception of tech giants such as Google, Meta and OpenAI. Instead, companies are using LLM application programming interfaces (APIs) to create tools in-house or by partnering with a third-party vendor. According to Digiday+ Research’s survey of 119 brand and agency professionals, cited in our report on how marketers are using AI, the majority of marketer respondents said that they primarily use third-party vendors to build AI tools. During Digiday’s conversations with publishing executives, several noted that many publishers partner with third-party vendors to build AI tools, while those with larger budgets often rely on a blend of in-house and third-party vendors. A company’s size and budget can greatly affect whether building in-house tools is even an option, according to the executives.

• Jon Roberts, chief innovation officer at Dotdash Meredith: “The actual infrastructure of these LLMs is hugely expensive. Many of the LLMs are pretty comparable, which means we now live in a world where they are rapidly becoming a commodity. As a publisher, we generally don’t need to build commodity tech ourselves. We want to use the best one out there, and then we get the benefit of all the innovation as these companies compete with each other… Then you go to second-tier companies, which are companies building tools on top of those AIs, who are not creators of AI. … Because it’s so early, many of our use cases are ones we have prototypes for and, inevitably, there are 10 startups that pop up for that use case. It’s a continued evaluation of which piece of tech is specific enough to our use case that we build it ourselves, or if speed is so important for an advantage, do we want to partner with a third-party company.”

• Harry Hope, CTO at Business Insider: “At Business Insider, … we have an established product and engineering team here, and AI is an area that we want to explore thoroughly. The ways of bringing the technology into our user experiences could benefit our users at the end of the day. So, in that sense, we’ve tried to do as much as we can in-house, especially in terms of integrating AI APIs into our products across the board.”

• Jessica Probus, publisher and general manager at BuzzFeed: “Compared to other publishers, we’re probably leaning more on the in-house side for a lot of things. It’s not necessarily because of AI, but because that’s how we operate. … We have an internal team, a machine learning team, a custom tech team and an innovation team. But the innovation team over the last four years has been focused more on AI in particular. On the tech side and the editorial side, most of what we’re building is using the platforms and using the LLM APIs to build things in-house.”

• Ingrid Verschuren, evp of data and AI at Dow Jones: “There’s two ways to look at it. From an internal perspective, we are using LLMs provided by partners and we build a model on top of it because it allows us to control the output and input. From a client perspective, we take a similar approach. Rather than forcing them to build a model, we actually offer several services that make it easier for them to integrate our content into whatever model they have.”
• Vadim Supitskiy, CTO at Forbes: “For internal tools and improving productivity, we mostly use third parties. We introduced Microsoft Copilot. We partner to have AI assistants for our help desk and IT problems. But when it comes to building products for users on forbes.com, we mostly build it ourselves. We use LLMs, specifically, we mostly use Google’s Gemini. But the products that we build on top of them are in-house. We know our audience best, and it makes sense for us to build those journeys ourselves.”

How publishers are adding AI into their workflows

The publishing executives Digiday spoke with told us they are incorporating AI into both external and internal workflows. While the ways in which they use AI for consumer-facing content varies, their internal usage of AI seems to follow a consistent pattern focused on reducing repetitive work. Additionally, across publications there was a clear consensus that using AI for internal applications poses less of a risk than using it for public-facing content. Here are some of the ways in which publishers told us they are using AI internally and externally:

• Forbes’ Supitskiy: “We use tools for development, like Microsoft GitHub Copilot, for our developers to test. There are transcripts for the newsroom. Our team might be working on images to augment. So, there is a lot of implementation, but it’s a lot of day-to-day tasks and making things more efficient.”

• Dotdash Meredith’s Roberts: “Team by team, finding the big, expensive, slow bottlenecks, and then seeing where and how LLMs can apply to that. That’s in multiple places: text, image, video, research and data analysis. Coding completion is something we’ve taken a deep look at in the evaluation of tools to help with repetitive bits of coding. I think the confidence that there will be places that can be made faster and quicker with AI has turned out to be fairly true. Everything’s a little more custom than people claimed it would be to start with, but we’re not short of processes that we are attacking in this way.”
• Business Insider’s Hope: “We segment into three buckets — internal use cases where we either improve workflows or improve internal tools, such as within our CMS or tasks that an employee undertakes from day to day. The second bucket  is the actual end user product experience. Can we take some of the advances in generative AI, and apply that to products to solve problems that were unsolvable before? Or, we can give users a much better experience than we could with the last generation of technology. And the last bucket is the macro-environment, especially in terms of syndication and how our content gets in front of readers. That’s tremendously impacted by AI across the technology ecosystem as a whole. The way that Google works is changing. They’re adopting AI, and that’s changing dynamics of how a lot of publishers are seen on search. Reddit and social platforms are trying different things as well too. Having awareness of the industry as a whole, and how that’s impacting us is the third bucket.”

AI with ‘strict human oversight’

Publishers sit at an interesting crossroads when it comes to incorporating AI within their external workflows. Due to journalistic integrity and ethical standards, AI is often not utilized in certain parts of the workflow, particularly for writing news content. And every publishing executive whom Digiday spoke with for this report made it clear that their publications do not publish content completely written by AI. However, while many publishers have strict rules about not publishing AI-generated articles, some have started creating other types of consumer-facing content using the assistance of AI, and guided by stringent human oversight. BuzzFeed, for example, uses AI to build quizzes and games, while Dow Jones uses AI for article translation and content summaries. 

• Dow Jones’ Verschuren: “One of the markets that we were not operating in was South Korea, and we’ve developed an assisted auto-translation service that is translated by using an LLM model. It was efficient because it was done by machines with strict human oversight. It allowed us to go into the market and to offer people in Korea [an option] to read the news.”
• Verschuren: “Part of Dow Jones’ business is Factiva, which is a database of business information services that a lot of analysts use to make their workflow more efficient. One of the things that we launched recently was Factiva Smart Summary. Rather than having to read through a whole bunch of news articles, they can go to the summary. They can then still trace back to the original article as well.”
• Verschuren: “[A tool we tried] is called a Joannabot. Joanna Stern is one of our tech columnists and every year when a new iPhone comes out, she writes a review about the new iPhone.  … She wanted to have a chatbot where users could ask questions. She used all of her previous columns about iPhone reviews as her model. It was a very cool experiment. It showed a lot of engagement from a reader perspective. However, even though it was a controlled content set that went into the model if you wanted to, there was still a possibility to break the model. If you asked one question, it wouldn’t hallucinate [give an incorrect answer], but if you try to break it, you could. Again, an example of why human oversight is so important.”

The Wall Street Journal’s AI-powered Joannabot

• BuzzFeed’s Probus: “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 that turned celebrities into Shrek. It was an example of letting our audience participate in the content creation. … That was one way of creating something that was co-created with the audience and editorial staff. What sets our thinking apart is that we’re not trying to use AI only. It’s the idea of having humans, whether it’s our users or editorial staff, partner with AI and make interesting things.”
• Probus: “Another example is for [our] Tasty [app]. We have a new function where you can personalize a recipe on Tasty in the app. Say you’re just obsessed with garlic, you could use the AI to add 10 times more garlic to every recipe. Being able to tweak it to be more personalized for what you want. A lot of [the applications] are ways of either contributing new things that didn’t exist before or being able to personalize the content that we’re already creating in a more infinite and engaging way.”

BuzzFeed’s Shrek-ify generator

AI advances publishers’ search functions

With more AI applications becoming accessible to the general public, consumers can now find AI-powered search functions to help them with their online queries. Google has implemented AI into its search results to answer questions and provide users with more information. Social media platforms are also starting to adopt similar AI-powered search features. As this trend continues, publishers have felt pressure to advance their AI-search functions as well, but they’re not incorporating the tech just because it’s en vogue.

• Forbes’ Supitskiy: “One goal that we have is [we] don’t use technology just because it’s cool. Make sure that you’re solving a problem. We quickly realized that search hasn’t changed in a long time. It’s kind of static on publisher sides, and we saw an opportunity to engage people. Our approach [with Forbes’ generative AI search tool Adelaide, which is powered by Google Cloud] was, let’s test it on a small percentage of the audience. And we got a lot of really great information from users, what they liked and didn’t. We saw the engagement go up considerably, [people were spending] 20% longer on [our] site and [having] four times the interaction with the page. People were not just getting answers there, but they were diving deeper into these topics or clicking on profile links, article links, or going back and asking a follow-up question.”
• Business Insider’s Hope: “We released an AI-powered search experience on our website. It’s the first time that we delivered a user-facing feature on Business Insider that directly and visibly uses AI. … We started with a problem that a lot of our users were facing, which is that our on-site search was not that great. In the past, a lot of people just said, well, we can just go to Google to search your site, so it’s not a big deal. And for a long time, publishers and Google used to have this symbiotic relationship, so it wasn’t a big problem. Ironically, with the advent of AI and with Google changing how it works, becoming a little bit more of a walled garden, introducing AI summarization where users are less incentivized to click off to a publisher, those dynamics are changing. It’s not acceptable for us anymore to have a subpar on-site search experience. We want to provide the best possible experience for most loyal readers. Because, in a lot of ways, search incorporating AI is almost becoming table stakes now. You’re seeing every major player in some way incorporate it.”

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

Publishers’ use of AI for advertising moves beyond the ads

Another significant area in which publishers have been using AI is their ads business. Possibly spurred by Google’s previous plan to sunset third-party cookies in the Chrome browser, many publishers looked for alternative ID options to incorporate into their ads businesses. By the time Google announced that it no longer planned to deprecate cookies in Chrome, many publishers had already built out AI-driven ID alternatives. And some of the publishing executives Digiday spoke with said they found that the new alternatives had better results. Others are using AI for article classification, content recommendations and subscriptions analysis.

• Dotdash Meredith’s Roberts: “Used correctly, AI keys off of words. It’s a language model, not a user profiling model. We’re not training this on historic cookie data. We’re training it on, and tying it to, content you’re reading. … We can build all kinds of metadata on top of the page in a fully privacy compliant way that starts with language. … By adding LLMs, we get a much higher resolution and higher fidelity. [For example,] because the D/Cipher tool that we built has been upgraded to have OpenAI baked into its core, we know that people who are planting seeds in their garden are also interested in long-term, buy-and-hold savings accounts …  and they’re interested in travel to Europe. Because we have combined all the data that we got from the site’s readers, we’re able to make these inferences of what you’re likely to do next. … By making the extrapolation not just from word to product, but the user’s needs to wants, without using any identity data, that massively increases the aperture with which we understand our user behavior in richer real time.”

• BuzzFeed’s Probus: “On the ad front, it is how to use AI while accounting for the increased privacy regulations from Apple and other places and also to do a better job of ad targeting. We’re able to scrape the content and understand the connections between different content behaviors without having to have any of that [private] data. We also have a huge amount of first-party data that we own, just because we have a lot of logged-in users and a lot of loyal users, so that gives us an advantage. I think we’re seeing the opportunity to use AI, not to get around the regulations, but to find opportunities for not violating them.”

• Business Insider’s Hope: “We’ve integrated AI into our CMS. … We’ve built a product called Saga React, which takes that concept of AI-generated tagging and allows us to classify our articles in a much richer way. We run all of our content through an AI algorithm, and it gives us the sentiment of the post. Is it optimistic versus pessimistic? Is it something that may not be brand safe for an advertiser and they wouldn’t want to associate with?”

• Forbes’ Supitskiy: “We’ve had AI as a core [part] of our first-party data platform, which we call ForbesOne, for a long time. It allows us to analyze the audience, create the segments and then connect the clients and advertisers. But that platform also grew into something more than powering the advertising business. Now it powers recommendations across the site. It also has a lot of propensity models, like propensity to subscribe. Is this user going to have a high propensity to subscribe? To register? To churn? We’ve built that model to recognize how to build the journey for our users, what to show them on forbes.com.”

What’s next?

Digiday+ Research asked the publishing executives we interviewed what they’re most excited about when it comes to the future of AI. Here’s what they had to say about expanding the use of AI for ad targeting and how audiences have reacted to various applications of the technology.

• Dow Jones’ Verschuren: “What I’m most excited about is the possibilities when it comes to multi-agents and how you can start deploying them. The internal use case, if coupled with where the need for efficiency is, I think that multi-agents are going to be helpful with that.”

• Dotdash Meredith’s Roberts: “Our OpenAI partnership to upgrade D/Cipher to take ad targeting from this finite level to this richer version opens up a lot of retail-style targeting with huge performance that we haven’t been able to touch before. There are non-trivial scaling problems to solve. … The use case feels very simple, but AI prompts are quite expensive. A lot of the innovation here is on the baseline infrastructure layer of efficiency and web scale. We’ve solved it for a Dotdash Meredith. We have now taken this and extended it so that we can classify the premium open web, so that we can understand the user intent across all publishers because we believe that there is a premium in the attention of people on the web that has not been captured by the old version of targeting based around cookies. There is a new premium to be built with more value for advertisers and publishers. We can use this to ditch cookie targeting and move to something richer, more real time and higher fidelity.”
• BuzzFeed’s Probus: “We’ll also start to see a lot more creative personalization. That’s what I’m most excited about. We have versions of that across the internet now, where things are personalized to you, such as ads for the things you bought already and follow you around. People are hungry for things that are helpful. The good publishers are going to be able to combine having that voice and that expertise, but also to better target and better create engaging content that actually meets consumers’ needs better, rather than doing the spray and pray distribution of trying to get your stuff targeted to people that might not even care about that thing.”

• Forbes’ Supitskiy: “Gen AI gives us the ability to provide the experiences to our audience that they’re looking for. We want to provide the right content in the right format, in the right time for them and encourage them to dive deeper into it. It might be a summary that you’re interested in, but it also might be a topic you want to dive into. We want to make sure that we meet you there and are providing that content.”

• Probus: “If you’re just using AI for the sake of it, people are not interested. If you’re using AI to do something that a human could do, there is consistently backlash and a reasonable questioning of what is the value. But when we’re able to make something that has inherent value and wasn’t possible at the same scale, or wasn’t as easy or fun, that’s when people are much less skeptical and are more like, oh, this is actually a good product. For the last two years, anyone has been willing to slap AI on something to increase buzz, and people see through that pretty immediately. That’s how we’re thinking about it. If it’s simply a cool product, [readers] don’t care how you did it. That is where we see people be less skeptical and interested in engaging.”
• Business Insider’s Hope: “AI is following the adoption curve that you see with a lot of new technology. The adoption curve typically goes very high and then it goes very low because a lot of people are disillusioned, or the technology doesn’t evolve as fast as people’s ambitions. There’s some evidence we might be heading there now with generative AI. But, with the adoption curve, you reach a point where it’s this nice happy middle where you’re productive and you find a good fit for the technology. It might not achieve all of the hopes and dreams, but it’s this useful thing that you’re incorpor​​ating. And AI is going to get there too. 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