This State of the Industry Report, sponsored by Arc XP, explores how publishers are incorporating AI into their newsrooms beyond the typical workplace efficiency avenues, how far they’ve gotten with those implementations, the challenges they’ve encountered and how they’re overcoming them.
As AI permeates every industry, publishers and broadcasters are no strangers to the technology. AI is no longer experimental but a useful tool for everyday tasks that can drastically improve how organizations function and deliver experiences for their readers. While many businesses have incorporated AI into their daily operations, publishers have a unique opportunity to bolster their success by leaning on AI throughout their newsrooms.
However, misinformation still surrounds AI and its usage. Additionally, consumers aren’t always trusting of the technology, nor are those working closely with AI, due to the noise around it coming for their jobs. Publishers and broadcasters are strategically navigating these challenges while implementing the technology into their newsrooms.
In this State of the Industry report, Digiday and Arc XP surveyed 108 publishers and broadcasters to understand how they’re incorporating AI into their newsrooms, aside from the usual workplace efficiency tactics, how far they’ve gotten with their AI initiatives, what this free time has enabled them to do and how they’re overcoming implementation challenges.
Three-quarters of our respondents attribute the bulk of their AI spending to editorial/content creation, followed by advertising and marketing, with 10%.
No one selected “we aren’t using AI,” signaling that publishers and broadcasters understand the value of this technology and are actively putting it into practice.
“We know all our customers have a different approach, so they may generally say they’re using AI, but I don’t think it means we’re going to have Chat GPT write every article,” said Joey Marburger, vp of content intelligence at Arc XP. “I think it’s more aligned with the approach that Arc XP has taken around workflow, so not using it to write a whole story or purely generate content, but for additive things like copy edit or figuring out some headlines, summaries, tags, translations and things that are more operational.
The top three teams handling AI initiatives as far as budget, allocation and execution go, align with where respondents indicate the majority of their spend is being allocated — editorial/content creation (93%), followed by a dedicated AI-focused team that collaborates with multiple departments (90%) and subscriptions and customer experience (89%).
The three at the bottom, an executive team managing across departments (23%), operations (13%) and advertising and marketing (9%), likely signal that publishers are working toward putting control in the hands of the departments that are using the technology, rather than introducing a middleman that could hinder productivity or innovation.
“You can’t do things in a vacuum and just give it to a data science team,” Marburger said. “Because generative AI is a little less data science in application, and it really comes down to what you want to do with it — it’s a tool like anything else. If you do predictions and more siloed product development with it, it could fall flat because it’s so new and changing so rapidly that if you’re not using it regularly or for real, not just experiments, then you don’t know what you want.”
While AI involves a large amount of data, Marburger underscores the importance of involving the teams that plan to utilize the data, tool, etc., rather than keeping things fragmented.
“If we built something we think is amazing and all the newsrooms rejected it, then it’s a complete failure, like any tool or product,” said Marburger. “So, when I was on The Washington Post side running product, that was how we developed things so quickly. When it was a nugget of an idea, you would go straight to an individual reporter, or team, or editor in the newsroom and spitball the idea, then start working on it. Not, ‘hey, we built this whole thing and didn’t include you,’ and it could just not work. I think AI is in that rapid cycle right now.”
None of our respondents indicated their budgets for AI are decreasing — nearly all (97%) said they’re increasing.
“Not many media organizations were prepared for this big influx of AI — it’s almost like when the internet happened, who had web developers already on staff?” said Marburger. “They didn’t exist, but luckily, AI specialists do. But it’s more about where that money is going and what it means. Now they can jump a few steps ahead and use their people to focus on making their business successful without investing as much in translation or distribution.
“Investment, especially for news organizations, has to be cautious: Should I hire a data scientist or another reporter?” he continued. “Well, the reporter would probably further my bottom line more immediately, but data science is maybe more multi-year, so that’s tough. I’m surprised there hasn’t been as much investment, but it’s trending up, so within the next couple of years, the level of investment in AI or AI adjacent things will probably be close to a significant chunk of the pure newsroom investment.”
While there have been many viral hype moments in AI, many media executives are now investing in the technology, knowing they may not see returns for a few years. To optimize savings in the meantime, they can use AI to open up to new markets, access new audiences and explore translation.
“Translation is more than just this English word to this German word, for example, it’s the context and management of it, and the resonance of your content in that geographic area,” Marburger said.
The focus resource-wise has been on infrastructure (93%), research and development (89%) and innovation labs and dedicated teams (75%).
Meanwhile, staff training (6%) and ethics and governance (8%) aren’t of focus, likely due to widespread challenges regarding establishing AI standards and guidelines. With AI’s tendency toward bias, and other ethical concerns, it’s likely publishers are pouring efforts into research and development, for example, to train or build LLMs that have reduced bias, increased standards and more.
Regarding how publishers are approaching the development of their AI infrastructure, nearly all are utilizing a hybrid approach at 96%.
“For news organizations, you pretty much have to do a hybrid model, unless you can get your hands on some GPUs, which is hard and very expensive,” Marburger said. “I don’t know if it’s the right time to do that because that’s just the models, but at the pace of how models are going, by the time you spend $100,000 or more, it’ll probably be irrelevant. You also have to have good customer traction. For example, a newsroom can create an account on OpenAI and use that in a way that scales with their usage. It’s definitely a hybrid time, although I anticipate that within the next three to five years, that barrier to entry to do more cost-intensive investments will go down.”
The AI tools publishers rely on vary, but, unsurprisingly, workflow automation tops the list at 93%, followed by video and image optimization (87%) and automated content creation (82%).
Ethics and sentiment analysis and fact-checking and verification are at the bottom of the AI tools publishers are using at 5% and 4%, respectively. Workflow automation and automated content creation are common uses for AI among our respondents throughout this survey, but, interestingly, while fact-checking and verification are only at 4%, 84% of respondents indicate they use AI for automated fact-checking as a consumer-facing initiative. This may signal an in-house initiative rather than specific tools or software.
“When you start getting into more advanced techniques, like fact checking, depending on the timing, some of the tools aren’t there yet without some pretty advanced knowledge, but that’s changing,” Marburger said. “We know some of the most advanced models will make things up sometimes, so you almost have to fact-check the fact-checking, which becomes extremely inefficient. But, people should be exploring it, and that’s probably the disconnect, because they’re looking into it, but it’s not good yet — it’s not up to our standards or they realize how much work it is and want to wait until they can scale it and make it fully trustworthy.”
One example of some challenges arising from current automated fact-checking tools was demonstrated through a study by Indiana University, which found many cases of the technology increasing the belief in false headlines and decreasing the belief in true headlines mislabeled as false — another reason our respondents may be testing this technology but not very happy with its performance.
As for the tools that aren’t working well for publishers, video and image optimization (90%), content recommendation and distribution (84%) and real-time data and alerts (73%) claim the top spots.
While 93% are using AI workflow automation tools, 45% aren’t satisfied with those tools. Nearly everyone using video and image optimization tools isn’t happy with the results they’re receiving, as are those using AI real-time data and alert tools.
“We haven’t done much with photo or video yet because it’s a standards thing, although models keep getting better and better,” Marburger said.
About the AI image generators like Midjourney, DALL-E and more, he added, “There’s the running joke of seven fingers and asking for something and getting text that’s all jumbled. There are hundreds of models out there with different levels of quality, and then there’s the ethical and copyright sides. If you’re missing visual assets, then is using AI better than not having anything? Or, there’s dynamic clipping, so creating a 30-second version of this 5-minute video because it has original content to work from. When you’re pulling it out of thin air, it’s tough to keep it consistent, and that’s something Arc provides — consistency in your prompting at scale, so it’s not everyone typing different things into ChatGPT and getting wildly different responses.”
More than half of respondents are in the adoption for routine tasks stage (65%), with nearly one-quarter integrating AI into their core workflows (24%). No one surveyed said they were in the planning only, strategically deploying across the organization or fully scaled across the organization stages.
Marburger cites an early maturity curve, with generative AI only having moved from prototype to commercially usable models in late 2022, for the likely lower percentage of publishers having fully integrated AI in their newsrooms.
“I wasn’t surprised to see such a steep drop-off between using AI for routine tasks (the 65%) and fully integrated (1%),” Marburger said. “Most newsrooms spent 2023–2024 experimenting with safe, low-stakes use-cases like transcription, translation, tagging and A/B testing headlines. Moving from pilot automations to cross-function orchestration, where AI touches every stage from newsgathering to commercial ops, requires multi-year platform work, new governance structures and staff retraining. Even now, 87% of newsroom leaders say generative AI is already transforming their organizations, yet few claim it’s embedded end-to-end.
“Integration demands hard, often invisible prerequisites,” he continued. “Outliers such as Reuters, where bespoke tools like Fact Genie and AVISTA are now live in production, had already invested heavily in data pipelines, unified content APIs and a ‘human-in-the-loop’ culture before generative AI arrived. Most publishers don’t yet have that foundation.”
As for the ways publishers and broadcasters are using AI so far, improving workflow efficiency tops the list (94%), followed by automating scheduling and distribution of content (82%) and automating video editing (i.e., social clips, adding subtitles) (82%).
Workflow efficiency has been the most widely promoted application of AI, so it’s no surprise that this is popular with our respondents. Tedious yet time-consuming tasks like scheduling and distributing content majorly benefit from AI assistance. The increasing need for video content can be more achievable with AI’s efficiency.
The least-selected uses for AI were crafting compelling headlines to boost engagement (10%), monitoring performance in real-time for actionable insights (7%) and creating concise article summaries (7%).
While it’s interesting that not many people are using AI for more of these small, time-consuming tasks, or for real-time performance, which could benefit future stories and engagement, it could be that they’ve found alternative ways to improve these things, or they haven’t found AI tools that have worked well in these areas.
“Our most used AI feature is recommending headlines and summaries, so that is a bit confusing,” said Marburger. “I think to implement those use cases, you have to be consistent and maintain style and tone — all things our AI features do with complete prompt customization. I think that’s why the features are so heavily used versus pasting a story into Chat GPT and hoping for consistent responses.”
Publishers could find great success if they’re able to find AI tools that work well for them. If the first ones they try aren’t delivering satisfactory results, there are so many to choose from, it’s worth trialing out alternatives to improve engagement and ultimately reduce tedious tasks.
Publisher respondents are using AI for a variety of consumer-facing initiatives, with only 3% saying they haven’t yet done so. Automated creation of accessible content (88%), personalized content delivery (85%) and automated fact-checking (84%) are at the very top, aligning with changes in consumer expectations.
While using AI for dynamic paywalls are very low on this list, they’ve been very successful for many publishers, including the Post.
“The Post has had some great successes with their smart metering, dynamic paywall, which has allowed them to more rapidly create that propensity and use cases or deliver different messages, almost more personalized generation to target that messaging,” said Marburger. “I think that’s a great use case of accelerating a business or subscription strategy they already had. I think it’s a little lower on the list because, do you have a business or subscription strategy that’s well-defined already? An LLM can’t write your business strategy and put it into the right template, it can only tell you possible paths to take, but maybe not any paths that apply to you.”
The vast amount of data that publishers and broadcasters gain when using AI is an easy way to lean into gamified experiences.
“Generative AI is really good at complex patterns, so if you have all this data that’s clean but you aren’t sure what to do with it, you could ask an LLM to define a markup chain or pattern around it to give you the power of a PhD-level scientist that you can then do other things with,” Marburger said. “But, you have to know what you’re looking for and have all that data. I think people are starting to realize they can drop a huge CSV file into an AI tool and immediately start extracting value. We’re going to see a pivot — and within a year, I think people will be saying, ‘I’m using AI more for business intelligence.’”
So, if a publisher has a large amount of readership data and wants to create a mini game to drive more engagement, they can utilize an LLM to help build such an experience.
Publishers are using AI to deepen engagement with their audiences via tools that monitor and moderate comments (88%) and predict emerging topics (85%).
Identifying keywords and topics that drive traffic, converting articles to accessible formats and hyper-personalized paywalls for optimal engagement and UX (15%, 14% and 2%, respectively) are the least used by our respondents. Notably, while only 14% said they’re using AI to convert articles to accessible formats to aid in audience engagement, 88% are using the technology for the automated creation of accessible content as a consumer-facing initiative. The difference between these could come down to conversion versus creation or viewing accessibility as a necessity rather than an engagement tool.
As AI automates tasks and works on menial items that require a lot of time, publishers are dedicating more efforts to strengthening editorial oversight (86%), spending more time connecting with local communities and audiences (85%) and focusing on innovation and experimentation (76%).
However, not as much time is being dedicated to the following: increased focus on training and upskilling (19%), amplifying underrepresented voices (18%), building revenue and growth strategies (12%) and developing niche coverage areas (5%).
“If you have more time for innovation or expanding different coverage areas, then that will probably indirectly impact that revenue side,” said Marburger. “It’s not that just more content equals revenue, but the inverse of that — less content, less revenue — is probably 100% true. I think what AI will change for reporters is that they’ll spend more time reporting, and that can’t be replaced by an LLM. Everyone is warming up to it, where they’ll have more time to do something that could be filled with work, or it could be spent taking a break, sleeping or thinking for a minute, because out of that will probably come the innovation and exploration to do other things, and AI tools can help accelerate that.
“I also love the narrative emerging around local news and community, because that’s obviously been gutted the most over the years,” he continued. “I think AI will help a lot there. What one person can do with sophisticated knowledge, could do what a thriving local newspaper team did in a 20,000-person community in the 50s.”
The initial outcomes publishers have seen within 6 months of implementing AI throughout their newsrooms are fairly significant, with improved quality content (88%), significant reduction of time spent on repetitive tasks (86%), reduced labor costs (86%) and faster content creation (85%).
The less prevalent outcomes are faster response time to breaking news (7%), expanded coverage (4%) and enhanced newsroom collaboration (3%). Only 3% of publishers say they haven’t yet seen any outcomes.
“This is totally in alignment with what we’re seeing,” Marburger said. “Our core focus with Arc XP Intelligence is to provide value through automating repetitive tasks and decreasing the time to publication. We know journalists are constantly being asked to do more production-like work with their stories, so that’s where we want AI to help the most.”
Aside from the 96% of respondents who have used AI in their newsrooms for longer than 6 months, most publishers are seeing increased operational efficiency (89%), empowerment of journalists (85%) and greater sustainability (84%).
Only 3% indicate haven’t seen any long-term outcomes yet.
As more tedious, time-consuming tasks are automated via AI, journalists can feel more empowered to cover stories they feel passionate about or simply pour more passion into their writing/coverage. The increase in efficiency can also lead to improved sustainability.
However, not many publishers are seeing tailored user experiences (4%), resilience in competitive markets (4%) and strengthened audience engagement (3%).
It could be that these types of results warrant longer implementation, need more experimentation or the current tools/methods publishers are utilizing aren’t providing the necessary results.
“It’s still early days for generative AI in the news,” said Marburger. “With any huge technology platform shift, you will always see those who lean in quickly, follow fast and lag behind. The news industry tends to be more cautious, which I think is a good thing. However, AI use is rapidly accelerating, which could be detrimental if organizations don’t at least experiment and have a strategy for adoption, even if they haven’t implemented it yet.
“AI is only one piece of the overall experience, so even though it can unlock new capabilities and deliver them quickly, there’s still a lot of non-AI development and product strategy going into it,” he continued. “For publishers to heavily invest, I think they need to see steady results before they jump into far more advanced use cases. Seeing the items at the bottom of the list doesn’t surprise me at all, but I do see that shifting quickly over the coming years.”
As AI becomes more prevalent and a larger discussion topic, it leads to fewer people being concerned about the technology coming for their jobs — as far as journalists are concerned — more frequent usage (and therefore less of a skills gap), and a deeper understanding of AI’s ability to integrate with numerous systems.
However, our publisher respondents still cite misinformation (90%), maintaining audience trust (85%) and maintaining a competitive edge (81%) as top challenges they face when incorporating AI into their newsrooms.
Integration with existing systems (5%), job displacement concerns (5%) and skills gap (5%) are at the bottom of the list of challenges, though likely due to increased familiarity with the technology.
Even with the increasing prevalence of AI, Marburger isn’t surprised to see misinformation, maintaining audience trust and maintaining a competitive edge as the main challenges our survey respondents are concerned with.
“There’s a lot of fear and excitement around AI, but publishers have to find the right strategy that works for them,” said Marburger. “If you force AI into a newsroom, there will undoubtedly be rejection and distrust. However, establishing clear guardrails and a best practices framework that allows newsrooms to build trust with AI tools will help them warm up to it. Journalists hate making mistakes, and AI can make mistakes. AI is like a junior reporter, so if you teach it new skills and give it experience, it will grow just like a human reporter.”
The top three challenges holding teams back from further AI adoption echo the top challenges they’re facing when incorporating the technology in general — maintaining a competitive edge (90%), misinformation (89%) and risk of over-reliance on AI (85%). Notably, maintaining audience trust, which was in the top three challenge-wise, comes in fourth here with 83%.
Those at the bottom are nearly identical to those at the bottom of the previous list: job displacement concerns (5%), integration with existing systems (4%) and skills gap (3%). More people indicated that maintaining a competitive edge and the risk of over-reliance were holding them back from further AI adoption than those who selected those as challenges they faced.
“I think what we’re seeing is a shift in the kind of concerns publishers are wrestling with,” Marburger said. “In the early days of adopting AI, the challenges are more tactical — how do we plug this into our workflows, who needs to be trained, that sort of thing. But once those hurdles feel manageable, people zoom out and start thinking about the bigger picture. There’s this worry that if everyone uses the same AI tools, especially the off-the-shelf models, we’ll all start sounding the same. That’s a real concern because voice, tone and editorial judgement are all things that differentiate a publisher. So, the fear isn’t just ‘Can we use AI?’ It’s, ‘if we do, are we giving away what makes us unique?’”
Regarding over-reliance on AI, Marburger emphasizes the importance of trust and control, recognizing that the challenge here is with policy, not the technology.
“Publishers are cautious, rightfully so, about handing over too much to automation,” he said. “There’s always a risk that we lean too hard on AI for speed or scale and lose sight of editorial rigor or nuance. It’s not a tech problem, it’s a governance problem. To overcome this, you have to build differentiation into your AI use. Find the best models that work for your use cases, then leverage fine-tuning and vector databases to augment the models with your content. Also, human judgment has to stay in the loop. AI can assist, but editorial oversight is what makes content meaningful and trustworthy — you need that balance.”
Publishers plan to overcome the challenges outlined here by improving data infrastructure (89%) and enhancing collaborations with tech providers (72%).
While publishers are concerned about the risk of over-reliance on AI (85% said it was holding them back from further AI implementation and 68% said it’s been a challenge when incorporating AI in general), prioritizing human oversight (5%) is last on the list here, so whatever they’re doing, to combat that, human oversight is not at the top of the list.
“Collaboration is key, but even large tech providers are in a similar situation where generative AI has been a great level-setting force among technology companies,” Marburger said. “Over-reliance on tech providers could be detrimental, similar to becoming over-reliant on social referral or search referral traffic. We all know how difficult that has been in the industry. It’s the classic build versus buy dilemma, and I think that right now is a time where publishers should do a mix — build what is critical to their business and buy what they don’t have the expertise or capital to do themselves. If you kick the can down the road, you’ll come to find there’s a very expensive toll to pay, and the can is way too expensive now.”
Looking ahead, the areas publishers most want to see expanded upon as far as AI goes include advanced, ethical AI (90%), data mining and pattern detection (88%) and cross-platform optimization (83%).
Content lifecycle management (8%), hyper-personalized chatbots (5%) and source vetting (3%) are the least chosen areas our respondents want to see innovation.
It’s unsurprising to see ethical AI at the top of this list, as maintaining audience trust and misinformation are high on respondents’ lists of challenges in AI adoption and what’s holding them back from further adoption. Ethical AI adoption is a surefire way to combat misinformation and gain and maintain audience trust.
As for those at the bottom, chatbots have already come a long way, so it doesn’t look like there’s much interest in further advancement anytime soon. While source vetting hasn’t been a priority among our respondents, 84% are using AI for fact-checking as a part of their consumer-facing initiatives. There may be a sense of distrust with AI’s ability to vet sources, but an acknowledgment that readers and consumers appreciate publications that actively, visibly fact-check.
“We have to remember that these LLMs are being trained for everybody, not just publishers or the news industry as a whole,” said Marburger. “However, they are meant for the information world, and model providers should work with publishers to understand their use cases because this is an industry where accuracy, ethics and information distribution are key. Model providers are looking for the same thing. I think the publishing industry has a huge role to play in shaping the future of AI, especially if we want that future to be ethical, transparent and aligned with public interest.
“Publishers bring something to the table that a lot of other industries don’t: a deep understanding of truth, context and responsibility,” he continued. “One of the biggest contributions we can make is pushing AI to be more accountable. We’ve spent decades building systems to vet information, check sources and uphold editorial standards. If we can help encode some of that thinking into AI models, or at least demand that those principles guide the design, we can steer AI toward being more trustworthy and less extractive.”
As publishers look ahead, Marburger recommends they prioritize more seamless AI integration into CMS tools that feel collaborative and natural. While there’s been a lot of focus on Chat GPT, it’s worthwhile to explore beyond chat-like experiences to see how publisher content can be used to offer new storytelling experiences to their readers.
“The value publishers provide is immense, but it can be overshadowed by less editorial-guided people and platforms,” Marburger said. “AI can help accelerate content production to get more real, factual information in front of people, and if we can’t just keep up, we have to lead and be at the forefront.”
Publishers are rapidly embracing AI, but not to replace journalists — to empower them. Ninety-seven percent of publishers plan to increase AI investment in 2025, with three-quarters focusing budgets on editorial tasks. AI is now routine for workflow automation, content tagging and translation, but deeper newsroom integration remains limited — just 1% say it’s fully scaled across operations.
While tools like headline generators and video editors are widely used, satisfaction varies. Many publishers say current solutions fall short, particularly in video optimization and real-time data. Still, early results are promising: 88% report higher content quality and 86% have cut time spent on repetitive tasks.
Misinformation, audience trust and the pace of innovation remain top concerns. Publishers want AI that’s ethical, adaptable and contextually aware, not just efficient. Human oversight, however, is rarely a top priority despite fears of over-reliance on automation.
The future lies in a hybrid approach that blends in-house tools with vendor solutions. For AI to truly reshape journalism, it must be shaped by journalists. As the tech evolves, the challenge is clear: Use AI to enhance editorial values, not erode them.
