AI Writes News Better Than It Finds It
Artificial intelligence has become proficient at drafting news stories from structured data. It struggles far more with the harder problem: spotting which stories matter in the first place.
A recent Wall Street Journal report highlighted a Fortune contributor using AI to produce six to seven news pieces daily from public documents. The automation freed the journalist from routine work-translating press releases, compiling weather reports, recording stock movements. But this efficiency masks a fundamental limitation.
AI systems excel at processing vast datasets faster than humans can. Yet they remain poor at identifying genuinely newsworthy developments from thousands of daily candidates.
The Signal Problem
The proliferation of AI-generated content has created what researchers call a signal-to-noise problem. An estimated half of all new writing on the web is now AI-generated. In academic publishing alone, the volume of AI-powered submissions has become unmanageable-even covering AI-related research comprehensively on platforms like Arxiv now exceeds what any single person can do.
This is where AI should theoretically excel. It can iterate through datasets humans cannot process, finding outliers in seconds. Instead, AI-driven newsworthiness detection remains stuck identifying stories based on what made headlines before.
The Backward-Looking Problem
Researchers at Stanford and Northwestern have made progress training AI systems to recognize newsworthy stories. These systems work best with structured data-court filings, state bills, city council meeting minutes. They identify patterns in word distributions that correlate with past news coverage.
The problem is fundamental: they look backward. They find more stories like the ones that succeeded before. Real newsbreaks rarely follow patterns.
Consider the Edward Snowden revelations. No algorithm would have predicted that a former CIA contractor would become a major news source. After his disclosures, one might imagine monitoring "recently departed CIA employees" as a strategy. But no public API tracks such movements. LinkedIn and similar platforms have closed their data to AI scraping. Even if the data existed, the human dimension-the decision to come forward-cannot be automated.
Where Algorithms Hit a Wall
AI-powered newsworthiness detection depends on formalized structures: JSON outputs from APIs, press releases from known organizations, RSS feeds with rigid hierarchies. These work for templated content-weather alerts, stock movements, municipal announcements.
They fail for everything else. A "security alert" posted on GitHub or X might be routine or might signal a critical vulnerability. The difference depends on context, community knowledge, and technical expertise. No keyword matching can parse the difference reliably.
Scientific papers present a specific challenge. Researchers frequently bury negative results or downplay failures in tables and figures. AI systems struggle to interpret visualizations. More critically, determining whether a paper represents genuine novelty requires understanding prior work-often scattered across years of citations in formats AI cannot easily process.
The Trust Problem
Journalists balance competing constraints: time, access, credibility, audience priorities. They develop instincts about which sources are reliable and which have hidden agendas. They know when to trust a tip and when to verify independently.
Any AI system claiming to filter newsworthy stories requires journalists to trust that discarded articles are genuinely not worth their time. That trust breaks the moment a major story emerges from a source the algorithm deemed unimportant.
A crowdsourced study training AI on research paper newsworthiness achieved about 80% agreement with expert judgment on top candidates. Agreement on broader selections was only moderate. The system missed factors like how a story would resonate with specific audiences.
Cultural shifts-political upheaval, wars, sudden crises-can upend all baseline assumptions. A finely-tuned AI system would need near-complete retraining when the world changes.
What This Means for Writers
The practical effect of AI's limitations in story discovery may be counterintuitive. As major publishers restrict data access to combat AI scraping, well-funded news organizations with resources for manual research gain relative advantage. Smaller outlets and independent journalists lose ground.
The technology meant to democratize news production is having the opposite effect. "Gut feeling" remains irreplaceable-not because journalists have mystical instincts, but because evaluating newsworthiness requires understanding context, community, credibility, and consequence in ways that resist automation.
AI has proven useful for the mechanical work of journalism: writing from data, translating information into prose, catching routine stories. The work that requires judgment-knowing what matters and why-remains distinctly human.
For writers, this suggests a bifurcation: AI handles volume and speed on known-structure stories. The competitive advantage goes to journalists who can identify stories no algorithm would find. That skill depends on the very thing AI cannot replicate: being embedded in communities, understanding significance, and knowing which sources matter before they become obvious.
Your membership also unlocks:
