Introducing the AI Article of the Future
A Frontiers Prototype to Rethink the News Experience with AI
Imagine a more personalized reading experience that provides a 60-second summary for the sprinter, an engaging audio narrative for the stroller, or an interactive chat for the studier. It's now possible with generative AI.
Personalize with AI
Hear it, break it down, or ask — get more from this article, your way.
The way we communicate has always been constrained by the current technology of the time. For the better part of 600 years, the de facto medium has been text.
Storytelling is crucial to our culture. Voice matters. We would gather around a fire to listen to tales, we scaled that to radio, broadcast television—but then you had to be there at that specific time and in range of the radio wave, connected to a power source. The newspaper and the scale of printing presses and their delivery networks allowed for consumption without the constraints of time and connection to power, but was still constrained to column inches and text and later image.
We've carried this format into the digital age. Web access was limited to a desktop computer, plugged into slow dial-up where text was the only real option. Even with the advent of faster internet and smartphones—where audio, video, streaming, and messaging have all been unlocked—we've had another constraint: SEO and advertising. This has forced publishers to play a game of cat and mouse with Google, having to build all their tooling and workflows around creating text content for search rankings.
But people have different expectations for consumption. The New York Times' product team understood this years ago when they used the analogy of a museum designed for three types of visitors: "The sprinter will just fly through the museum, the stroller will putter around, and the studier will read every little thing next to every piece of art."
Just as museum visitors engage differently with exhibits, news readers have distinct consumption patterns. The sprinter seeks quick, impactful summaries. The stroller enjoys moderately paced, engaging content experiences. The studier wants in-depth, thorough content exploration.
Someone checking their phone between meetings needs different information density than someone settling in for a long read. Someone walking their dog wants to listen, not stare at a screen. AI has unlocked the possibility for us to scale the methods we have always wanted.
What if publishers could create formats that best suited all three types of readers at scale?
The experiment
We built our site as a testing ground, with our articles as the content. That content has multiple ways to experience it. Three core modalities: listening, summarization, and conversation.
Our content management system of choice is Sanity, which allows for the level of customization we need and makes connecting to various AI services and additional tooling straightforward. The technical foundation uses Gemini Flash for lighter tasks like summaries, and Gemini Pro for deeper conversational elements and voice. We wrapped everything in Mastra AI and the Vercel AI SDK to make model swapping simple. This lets us optimize for speed and cost without rebuilding the experience.
Listening: returning to our roots
Audio doesn't have to be just narration of the article. We generate a script that transforms written content into a podcast, giving editors the script for refinement and review before audio generation.
This takes us back to those fireside stories, but without the constraints of time and place. People walk, commute, exercise. Traditional articles lock you to a screen. Audio versions return to our natural preference for spoken narrative while giving you the freedom to consume when and where you want.
Summarization: choice of depth
Not everyone has 15 minutes. We created three consumption levels:
Quick read gives you an overview in a few bullet points. Perfect for deciding whether to invest more time.
Key takeaways provides a 1-2 minute read covering the essential content. More than scanning, less than full commitment.
Full article remains available for deep engagement.
We aren't trying to simplify the content. We're acknowledging that context matters. While we're doing this dynamically, there's no reason we couldn't have this as specific tooling within our CMS to generate a first pass and let the authors tailor it.
Conversation: articles that respond
The chat modality lets readers explore ideas rather than just consume them. Ask questions, challenge points, dive deeper into specific sections. The AI understands the article's context and can explain concepts further.
We added voice input and audio responses. Speak your questions instead of typing. Listen to explanations while multitasking. This creates a more natural, conversational relationship with written content.
The technical challenge here is interruption. Real conversations let you cut in when you disagree or need clarification. We solved this with immediate response stopping—the audio pauses instantly when you start speaking, processes your input, then responds. No button pressing required, though privacy controls let users disable voice detection.
While not as effective in terms of pausing or detecting user tone, this technique is cost effective compared to voice-to-voice LLM implementations.
The real insight: technology finally matches intent
As John Hashimoto who is consulting as an AI product strategist for the American Journalism Project puts it:
At the American Journalism Project, I work with newsrooms wrestling with what AI means for their mission, their workflows, and their communities. The Article of the Future is about unlocking that value: using AI not to replace the reporter's voice, but to extend it—across formats, audiences, and time.
Users seek diverse content consumption methods, prioritizing relevance to their current circumstances and preferences.
Traditional publishing couldn't afford this variety because the technology didn't allow it. Creating multiple formats manually for every article would be impossible at scale. We've been constrained by the medium, not by choice.
AI changes this equation. As Matt Brown from our team explains:
What's most exciting is the ability to reformulate core journalistic content—instantly and intelligently—for different platforms, formats, and audience modalities. We can take a single reported story and adapt it into a concise summary, a narrated audio experience, a chat interface, or a visual timeline—without reinventing the wheel each time and at great cost.
One source article becomes multiple experiences without the production overhead that previously made this impossible. We can finally deliver what we've always wanted to—stories that adapt to how people want to receive them.
Tooling for podcasts
Part of this process is building the tools to augment the editorial workflow. We built a custom tool for podcast generation. The tool connects to our CMS (this could be any data source) and allows the editor to create a new podcast with the text as the content. The pipeline then allows them to set a length and format: interview, conversation, news, along with additional directions on the audience the podcast is for. A transcript is then written, passed through multiple agents to research, draft an outline, format into a script, be reviewed for consistency and finally presented back to the editor for review. At this point the editor can make changes, decide to create another version, or go ahead with the final audio generation. It's important to have these stages and the ability to create multiple versions—like any creative process, you need drafts until you get to the final version.
Once the transcript is good to go, the text goes off to Google TTS service and renders out the audio, streaming that up to Mux for storage and playback. The editor can then listen to the audio before inserting it within our CMS.
What this means for design
Building these interfaces breaks the usual assumption that everyone wants the same reading experience. We're not optimizing single user flows anymore. People switch between wanting quick summaries and deep dives within the same article. They might start listening while walking, then switch to reading when they sit down.
This shift is happening across the industry. From Hearst to Ippen Digital, publishers are challenging the limits of the 1,000-word article. Microsoft's CEO even "listens" to podcasts by uploading transcripts to Copilot and chatting with it. As one European newsroom put it, "The article is dying—not because people don't want information, but because they want it their way."
This pushes us toward thinking like a service rather than a product. Instead of prescribing the user journey, we provide tools and let people create their own path through the content.
The measure of success isn't time on page or completion rates for individual formats. It's whether people find value in having choice, and whether that choice leads to deeper engagement with ideas.
We have implemented this across all articles published on our website. Our test case, "The AI Revolution in Newsrooms: 3 paths to the future," showcases how a single piece of journalism can live simultaneously as a podcast, a chatbot, a personalized brief, and a learning tool.
This initial rollout helps us understand how readers interact with more responsive articles. We're observing user behavior and gathering feedback to refine our tooling and enhance the experience. As we gain insights into engagement patterns, we'll continue iterating on what's possible in digital publishing.
Learn more about LLM content tooling
Our goal is to create an ecosystem where content isn't just consumed, but actively experienced. We encourage anyone interested in transforming their content workflows to reach out to us.
More
- AI and Deepfakes
As AI continues to reshape media at warp speed in 2025, I’m excited to bring you the first in a series of AI and Media Radar Reports from Frontiers, our newly launched AI Product Studio.
- The AI Revolution in Newsrooms
Two years ago Open AI’s ChatGPT made its "What can I help with?" debut, sparking a whirlwind of reactions across the media industry.
- Introducing Frontiers
In today's “ground is shaking” technology landscape, the media industry faces a paradoxical choice.
Return to articles