Digiday+ Research: Publishers favor generative AI over predictive AI

This is according to a Digiday+ Research survey conducted in the fourth quarter of 2025.

When Digiday asked publishers what type of AI they use in various workflows at their companies, more survey respondents said their companies use generative AI compared with predictive AI applications — across all workflows. Generative AI creates text or media based on a data set — for example, chatbots and image generators like Midjourney. Predictive AI creates forecasts or classifications, such as statistical modeling or category tagging.

The workflows in which publisher respondents said their companies use generative AI the most were sales (62% of respondents), creative production and design (61% of respondents), and marketing (58% of respondents). More than half of respondents also said their companies use generative AI for copy editing (57% of respondents) and for editorial research (55% of respondents).

“It’s certainly powerful in the research and ideation phase, when you’re trying to get thought starters, to collate a lot of information and to figure what direction you want to go in. That’s been pretty universally adopted across all of our teams,” Forbes’ Gould said.

Less than half of survey respondents said their company uses generative AI for editorial content creation (47% of respondents) and for content management and publishing (41% of respondents).

Although the ability of generative AI applications to complete tasks has improved over the past year, the publishing executives Digiday spoke with for this report made it clear that, while they have AI tools at their disposal, they are not using AI to create news content.

“We’ve been very public and very consistent in talking about this. We won’t publish content written by a machine,” People Inc.’s Roberts said. “It doesn’t mean that we won’t use AI in the production process, but the reason is very straightforward. LLMs create the average next sentence, paragraph. Any publisher that’s in the business of creating average content in 2025, that’s not really publishing. We’ve never been in the game to create average content.”

Roberts said that when People Inc. uses AI tools for production tasks like coding, humans still guide the process. “We’re definitely still in the world where coding has a human in the loop,” Roberts said. “For example, in data science, with Gemini 3 and the new tools from OpenAI, advanced SQL queries and data exploration are getting a lot quicker. … But, you need to know the problem you’re trying to solve and how to solve it. You need to set the code off and then check that it actually solved it the right way. There’s a lot of problem design, architecture and then QA [quality assurance] required before getting productivity out of AI tools.”

At Forbes, Gould said the publisher is actively assessing the areas where teams are integrating AI tools as AI applications continue to become more sophisticated.

“As we start to get deeper into actual work products, we’re trying to think through where AI is going to fit,” Gould said. “The engineering team is using it mostly for testing or commenting code, rather than for creating code. … Our design departments have been experimenting with how to streamline the process of creating assets and distribute them to other teams. They’re also experimenting with how these tools can unlock creativity.”

Hearst has been testing how agentic AI might improve processes for its ad sales department as part of a larger sales strategy involving generative AI. Hearst is running single-modal tests, which involve an AI agent carrying out specific requests step by step — like a digital assistant following a to-do list. The publisher has encouraged sales teams to use generative AI tools for administrative tasks like CRM updating, account research, pre-call planning and creating media proposals.

Michael McCarthy, senior director of AI, sales and business solutions at Hearst, told Digiday in May 2025 that the average amount of time it takes a salesperson to complete account research using AI is two minutes, down from 40 minutes without it.

Dow Jones’ Verschuren said, regardless of which type of AI tools employees use, maintaining oversight is key. The company recently created an AI steering committee for that purpose.

“We want everybody to pick the right tool, because it’s all different workflows. The job of a marketing person is very different from that of a journalist,” Verschuren said. “Anybody who wants to use or build a tool goes through a review so that we can decide whether the tool is acceptable, but also to make sure that the tool is deployed safely, ethically and in line with our standards.”