If you work in a newsroom, a marketing department, or a comms shop, you’ve heard the AI pitch a hundred times by now: It’ll save you time. Story ideas, research, production, headlines, social copy, and even, to some extent, writing. All of those use cases are about doing basically the same thing, only faster.
There’s nothing wrong with that. Efficiency is a worthy goal, and artificial intelligence can help immensely in advancing it. But AI can also open up novel ways of achieving the ultimate goal: connecting and engaging with audiences who find your information valuable. My former colleague NLW calls this “opportunity AI,” and he dedicated a recent episode of his AI Daily Brief podcast to it. He was talking to a broad audience. I want to work out what it means specifically for media.
This also feels like the right moment to have that conversation, because the efficiency gains haven’t exactly won over the goodwill of readers. Some newsrooms are seeing impressive numbers in terms of output: one African digital publisher told an industry conference that a single reporter’s weekly output jumped more than 150% after adopting its in-house AI tools. But the experience for readers hasn’t changed at all. It’s still an article, except now they’re checking it for AI tells. Opportunity AI in media means treating the article as just the beginning.
Your archive is a product now
The first place most newsrooms went looking for something new was the chatbot. All kinds of publications experimented with them, from The Washington Post to my own, but the ones that are still around point to the smarter way to think about it: not chatbot first, but expertise turned into a product.
That approach favors niche titles especially, and Nursing Times, a trade publication for nurses, is the clearest example around. It launched its answer engine, which draws only from its clinical and news archive, in February 2024. According to Robin Booth, the publication’s managing director, it had fielded more than 200,000 questions by mid-2025, with about 100 subscriptions directly attributable to the feature. Here’s the number that’s revealing: 80% to 90% of that usage came from suggested questions embedded right in articles, not an open chat box. Readers don’t want to interrogate an archive. They want the next question answered right where they already are.
Similarly, Skift’s “Ask Skift” feature has existed since 2023. It’s built to answer travel professionals’ questions, fueled by the publication’s more than 11 years of reporting, research reports, travel companies’ financial filings, and more. It’s meant to turn Skift’s most engaged readers into users, CEO Rafat Ali wrote.
But a chatbot bolted onto a website is basically a 2023 idea. Lenny’s Newsletter, a deeply reported newsletter for builders, has updated it for a 2026 lens. It’s offering “Lenny’s Data,” its archive of 370 posts and 317 podcast transcripts accessed via an MCP server that paid subscribers can connect to their personal AI, whether that’s Claude, ChatGPT, or something else. That way the reader can pull that expertise whenever they want, or their AI can reach for it on its own when it’s relevant. Lenny’s Data is probably better described as a creator business than a newsroom, but it’s the cleanest illustration yet of an archive becoming a product rather than a destination.
Local news has its own version of the same idea. The Texas Tribune turned its ongoing coverage of school vouchers in the state into something readers could ask about, with their questions sometimes influencing future coverage. Every time the chatbot was forced to answer “I don’t know,” the question was forwarded to the reporting team, which could turn it into a new story. The first question it couldn’t answer, about how vouchers would affect the state’s teacher retirement system, became exactly that.
Go further afield and you get Agência Mural in São Paulo, which worked out how to take public weather and flood data, add its own editorial judgment, and deliver personalized WhatsApp alerts to five underserved communities. This is service journalism for the age of AI: not a stream of stories, but a continuous decision service.
The video desk you can build yourself
Video is where this gets interesting if you don’t have a big team behind you. Creating a sustainable video strategy has been something of a holy grail in digital media organizations. Even a modest operation typically requires a team of highly skilled individuals to create, manage, and distribute the content. To make the reporter-narrated short-form videos that are becoming more common, publications like The New York Times and The Economist can hire staff. Everybody else has usually had to settle for a clipping tool.
That’s starting to change. AI models and tools have become sophisticated enough to edit video on their own. In theory, a reporter could shoot a few minutes of crude video on a smartphone, and a custom pipeline will cut it to 60 seconds, clean up the audio, and add captions and branding. Once approved, the AI can publish and share it across social networks. NLW is already doing a version of this: The AI Daily Brief’s clips run through a pipeline he built himself with Claude rather than a commercial tool, and he isn’t a coder.
Short video isn’t a new idea, but using it this way still counts as an opportunity rather than an efficiency, because for most publications simply weren’t doing it. There was no workflow to speed up. With AI handling the bulk of production, reporters can become visible in a format audiences are moving toward while the newsroom retains ownership of the whole pipeline.
Many will fail and that’s fine
A pipeline, though, isn’t the same thing as an audience showing up, and betting on new AI opportunities also means some of those bets won’t land.
Case in point: a research center at the University of North Carolina at Chapel Hill partnered with four local outlets to build chatbots with software that cost about $40 a month. Over 45 days, the bots got just 185 queries total, and about a third of the conversations included a question the bot couldn’t answer. Three of the four outlets dropped theirs when the program ended.
Politico’s venture into opportunity AI was much more expensive. Its custom-report tool could build bespoke publications based on the detailed reporting of its Politico Pro library. However, it sometimes created material that included hallucinations, so the company agreed to shut it down in May as part of its arbitration with its staff’s union.
So what actually separates the experiments that survive from the ones that don’t? A few things line up: the cost, and with it the risk, tends to stay low. The tools get handed a narrower job than an open “ask us anything” chatbot dropped in front of readers. They live somewhere readers already have a reason to ask something specific, a WhatsApp thread or an article page, and there’s a feedback loop back to the staff, so the tool has a chance to get better.
The opportunity AI test
If you’re weighing a use case and can’t quite tell whether it’s opportunity AI or efficiency wearing a new outfit, run it through four questions:
Do readers get something they could not get from you before, or the same thing faster? If the honest answer is “faster,” it’s not opportunity AI.
Could you have afforded to try this a year ago? If yes, it isn’t opportunity AI. It’s a project you never got around to prioritizing.
Does it make your reporters more visible or less? A new video pipeline for first-person short videos? Yes. A chatbot that brings new ideas to the staff? Yes, indirectly. A rewrite AI where the reporter shares the byline with a bot? No.
Is the subscription still the unit of payment? Almost every example that works so far bundles the new product into an existing subscription or experience. Nobody has proven readers will buy “AI” on its own.
None of this means abandoning the article. It means admitting the article was never really the point. The point was always the reader who needed something, and for most of the media’s history, the article was the cheapest, easiest way to fulfill that need by a mile. That’s no longer true. A nurse can get an answer, a local resident can get a flood warning on a messaging app, a subscriber can put a decade of reporting to work inside a personal toolset, and a reporter can show up on a phone screen without a production crew.
The newsrooms that figure this out won’t be the ones with the biggest AI budgets. They’ll be the ones that stopped asking how much faster they could make the old thing, and started asking what the audience would do with the new one.
A version of this column appears in Fast Company.
