The Philadelphia Inquirer built Scrape, an AI tool to surface hyperlocal news

The Philadelphia Inquirer developed Scrape, an AI-driven tool to surface hyperlocal news leads from a fragmented set of sources - municipal meeting agendas, school calendars, Facebook pages and other small outlets - after editors were spending roughly 12 hours a week just finding items for community newsletters. Instead of treating the project as a purely technical experiment, the team grounded it in existing editorial workflows: they converted a lead editor’s curated source list and hand-produced tip sheets into training and evaluation data, ran daily AI-generated digests, and had the editor annotate outputs every other day. Iterative prompt refinement moved the system away from noise (rescheduled meetings, routine road closures, recurring promotions) toward genuinely impactful items. Editorial context was encoded directly into prompts with newsletter-specific briefs - audience, in- and out-of-scope geographies, school districts and explicit exclusions - so outputs matched reader interests (for example, a Chester County brief focusing on local government and consumer-impacting developments).

Scrape matured from an experiment into critical infrastructure supporting six, soon eight, newsletters and informing nearly 20 journalists across the newsroom, reducing sourcing burden and letting editors focus on storytelling. Key lessons: newsworthiness cannot be a one-size-fits-all prompt; scaling requires editor-authored briefs; and cost-quality trade-offs for searching many small sources must be defined up front. The practical blueprint: start from a real manual workflow, turn existing outputs into training targets, establish tight human-in-the-loop feedback, encode concrete editorial judgment in prompts, and design architecture around explicit cost/search constraints.