When you run content without a team, the temptation is to hand AI a broad prompt and hope it figures out what your audience needs. The Philadelphia Inquirer took a different approach with Scrape, an AI newsroom tool built to surface hyperlocal news leads for its newsletters. The case study, published by the Lenfest Institute on September 30, 2026, documents a project where the most important component was not the model but the existing editorial workflow it was wrapped around. For founders planning content without a content team, that distinction is the real lesson.
What Problem Was Scrape Built to Solve?
The Inquirer publishes hyperlocal newsletters covering specific communities across the Philadelphia area. The challenge was not a shortage of information; it was the opposite. Newsworthy stories were scattered across a fragmented ecosystem of small sources: municipal meetings, school calendars, Facebook pages. As Kevin Hoffman, The Lenfest Institute’s AI Fellow embedded at The Inquirer, put it, gathering news for these regions came from “a lot of really small sources, and so it’s a lot to sift through.”
The human cost was concrete. One editor was spending about 12 hours a week just finding items for newsletters. That left little time for writing or shaping the voice of the publication. This is the kind of problem many small-business owners recognize: not that content ideas don’t exist, but that finding and filtering them eats the hours you need to actually produce something useful.
How Scrape Was Built Around an Editor’s Workflow
The Inquirer’s tool, Scrape, was created to save time and raise the signal-to-noise ratio of that curation work. What made it work, according to the case study, was that it was built around The Inquirer’s existing workflows with tight feedback loops from the editor — not as a purely technical experiment.
The starting point was the manual process the lead editor had already developed. They had compiled a curated list of sources and were producing daily tip sheets by hand. Hoffman turned that into a direct AI training and evaluation loop. The hypothesis was that AI could find newsworthy information, and the test was deliberately comparative: run AI every day over the same sources the editor cared about, produce a tip sheet, and compare it against what the editor produced manually.
This matters because it reframes editorial workflow AI as an augmentation of judgment rather than a replacement for it. The features that matter for AI-assisted content planning, in this framing, are not novelty but fit: does the tool plug into the sources and standards you already use?
Starting With a Curated Source List
The Inquirer didn’t ask Scrape to discover which sources were worth reading from scratch. The editor’s curated list was the input. The AI ran over sources the editor had already vetted. That is a meaningful constraint for anyone considering AI for news lead generation or topic discovery: the tool’s output is only as good as the scope you give it.
Keeping the Output Format Familiar
The output format also stayed close to the manual workflow. The editor produced tip sheets by hand; the team used AI to produce a tip sheet and then compared the two. This kept the deliverable recognizable and directly usable, rather than producing a novel format that would need to be integrated from scratch.
Why the Feedback Loop Was the Real Test
The comparison phase ran for three to four months and was, in Hoffman’s words, “pretty intense.” Every other day, the editor gave feedback on how Scrape was doing. The team compared the AI’s output with the editor’s manual output to understand whether the AI was extracting the same insights, finding new ones the editor missed, or missing things the editor caught.
That cadence is the strongest signal in this AI journalism case study. The team did not ship a tool and check in later. They treated the evaluation as an ongoing loop in which the editor’s judgment set the standard. For founders, this maps directly to content planning: a workflow that puts human review before publication is what turns volume into usable output. The AI did not shorten the feedback loop’s importance; it made it the central test.
What the Case Study Doesn’t Tell You
The case study does not report specific outcomes beyond the initial problem. It does not say how many hours were saved, whether lead quality measurably improved, or how Scrape performed compared with the editor’s manual output at the end of the trial. It also does not detail the technical implementation cost or the engineering resources required to build and run Scrape.
That is not a criticism of the project; it is a boundary on what founders should infer from it. The three-to-four-month intensive feedback loop shows that building a custom AI tool in this way can require significant time and attention. For a founder without a content team, that is a relevant consideration before attempting a similar build.
There is also a principled limit in the Inquirer’s own AI usage policy, updated December 3, 2025: every published piece of Inquirer journalism remains under human editorial control. The policy treats AI as a tool that can expand capacity and reduce repetitive tasks, while stating clearly that the core work of reporting, storytelling, and connecting with communities is irreplaceable. The case study and the policy agree on the same point: AI may surface leads, but it cannot replace the editorial judgment needed to shape a publication’s voice and ensure trust.
Practical Lessons for Founder-Led Content Planning
Scrape’s approach translates into a few concrete practices for small-business owners whose content planning currently depends on manual trawling or a loose list of topic ideas.
Document Your Manual Process First
If your content planning currently starts from a blank page, the first step is to write down the places you already trust: forums, newsletters, competitor updates, customer questions, industry calendars. That list — not a broad query — is what you should feed into any AI tool you use. The Inquirer’s editor had already done this work, and it became the foundation for Scrape.
Run a Direct Comparison, Not a Blind Test
Ask the AI to produce a draft topic list from your curated sources on a set cadence, then set it beside the list you would have made manually. Ask three questions: Is it extracting the same insights? Is it finding things you missed? Is it missing things you would have caught? That comparison is how you learn whether the tool is actually helping.
Keep Human Review Non-Negotiable
The Inquirer’s editor reviewed Scrape’s output every other day. For founder content, the review step is where approval before publishing protects both accuracy and voice. As the Inquirer policy frames it, you remain accountable regardless of the tools used. The same principle applies whether you’re using a custom-built tool or an off-the-shelf content system.
- Start from your existing sources. Before asking AI to find topics, write down the places you already trust. Feed those to the tool rather than an open-ended query.
- Keep the output format familiar. If you currently plan content as a list of notes or a simple topic table, ask the AI to produce that same format.
- Compare AI output against your own list. Run the tool over your chosen sources on a set cadence and set the results beside what you would have produced manually.
- Make the review step non-negotiable. Approval before publishing protects both accuracy and voice. You remain accountable regardless of the tools used.
None of this requires building a custom tool from scratch. The transferable insight from this AI journalism case study is the structure of the test: a curated source list, a comparable output format, and a human standard against which the AI is measured. That structure works whether the AI you use is a newsroom tool or a content-planning aid for a niche blog.
For founders using tools like Serpio to plan content around trends in a specific niche, the same principle applies: the system can surface what’s being talked about, but you decide what your business can credibly speak to before anything goes live. The Inquirer’s lesson is not that AI replaces editorial judgment — it’s that AI works best when it’s built around the judgment you already have.
