Turning a Public Data Feed Into Editor-Grade Articles at Scale
Dry public filings in, original news articles out. A two-stage system watches a public filings feed, lets only records with a real story through a 10-filter quality gate, then routes each one across 26 filing types and 7 research paths to a writer that produces angle-led articles a human editor would sign off.
Before: the stories were there, but nobody could find them fast enough.
Public data feeds are full of publishable news: a notable appointment, a pattern of resignations, a change of ownership. The problem is volume. Raw filings are dry, structured and endless, and almost all of them are not stories.
Doing this by hand means someone reading record after record to find the one worth covering, then researching the people and companies involved, then writing it up. It is slow, it is tedious, and the best stories are spotted late or not at all.
The operator, a digital publishing business, wanted a system that could watch the feed, pick out the records that make a story and turn each into an original, angle-led article good enough to publish, without a journalist writing it and without it reading like a machine wrote it.
What we built: finding separated from writing.
Stage one: the poller and the gate
The first workflow scans the data source continuously and pushes every record through a 10-filter quality gate. Only records with real story potential reach the writing stage. That keeps the site from drowning in noise.
Stage two: routing and research
The second workflow routes each qualifying record by filing type, across 26 types, down one of 7 research paths. Each type has its own story template, so an appointment story and a resignation story are structured and angled differently, the way a real news desk would handle them.
The hard part: voice
Early output was accurate but gave itself away through forced keyword insertion. The SEO seams showed through the prose. The fix was structural. Instead of writing clean copy and then jamming a keyword in afterwards, the writer now chooses a natural, human-sounding keyword up front and writes with it already in place. The rule is simple: if you would not say it out loud, pick a different keyword. The writer was also taught to include the context a human editor would expect, such as board composition, prior company names and other directorships, so the articles carry real depth rather than restating the filing.
A writer prompt refined over many versions
The writer prompt is now at v7+. It carries: A downstream step verifies the keyword is present and adds outbound links to legitimate sources. It acts as a safety net, not as the writer.
What changed: a steady supply of real stories.
The business turns a stream of dry public records into a steady supply of original, angle-led articles that read like they came off an editor's desk, with context, natural keyword placement and credible sourcing.
The 10-filter gate means only real stories get written. The routing across 26 filing types means each story is handled in its own shape. The result is a repeatable way to find publishable news in a public data feed at a volume no newsroom could match by hand.
Day to day now.
The poller runs in the background. Most records are filtered out silently. When a director with a notable track record joins a company, the record clears the gate, gets routed to the appointments path, picks up context about the person's other roles and comes out as a publishable article with a natural headline and sourced links. Pieces that fail the kill check never reach the site. The team's time goes on tuning the templates and the gate, not on reading filings.
The same pattern, in your business.
Most businesses sit on a feed of structured data that contains useful signals, if someone had time to read it all. The pattern is the same: filter hard, route by type, act with a template for each.
- If you run a recruitment firm, it watches public appointment and company data for leadership changes and drafts outreach to the right contact the same day.
- If you run a property business, it reads planning or transaction data for your target areas and produces a weekly market note in your voice.
- If you run a B2B sales team, it flags new companies, ownership changes and expansions in your sector and writes a short brief on each for the account owner.
Not sure which of your data sources are worth mining, or how strict the gate should be? That is exactly what an AI Consultation works out before anything is built.
AI Automation
This is AI Automation applied to data: raw records turned into accurate, publishable output on repeat. We point the same pipeline at the data that matters to your customers.