Notes on sourced writing, from a newsroom pipeline that writes about itself
Every post here is drafted in Brenzuri from published reporting, then reviewed by a person. The brief, the source list and the confidence report are published with the text.
AI-written articles and SEO: what Google's guidance actually requires
Search engines do not penalise text for being machine-written. They penalise text that exists only to rank. The difference is sourcing, structure and whether a reader gets an answer.
"Explainer for content teams: does AI-generated content hurt search rankings? Use Google's own documentation as the primary source, cover the helpful-content guidance, and end with a pre-publish checklist. 1,200 words, neutral."
Why a single-source claim is the most useful thing a draft can tell you
Deleting an unconfirmed sentence hides information. Underlining it puts the decision where it belongs, with the editor, and records what they chose.
One brief, two languages: the EU privacy article, source by source
The sample article on our landing page came from a 900-word brief and five sources. What each source contributed, which figures only one outlet carried, and what changed in the Spanish version.
This post was written by the tool it describes. Here is the brief.
A walk through one article, from the six fields in the brief to the approved second version, with the pipeline trace and the two sentences the editor rewrote.
What an editorial agent should be allowed to do, and what it should not
Brief, generate, read sources, fix flags: yes. Approve and publish: never. The reasoning behind the MCP permission set, with the cases that shaped it.
A budget, not a balance: what an agent can spend and what it never sees
An agent working in Brenzuri is told whether the next run is affordable and nothing else. Why the credit balance is hidden from it, how hold and capture keep a failed run free, and what the version history shows.
Readers say they want AI-assisted news labelled. Most publishers still do not.
What the survey data says about disclosure, what publishers actually print, and why a label derived from an approval record is harder to get wrong than a checkbox.
Same story or different event? How the pipeline keeps sources apart
Two outlets can use the same names and dates and describe different things. Clustering before writing is what stops a draft from merging them.
The brief is the article: six fields that decide the quality of a draft
Topic, angle, audience, language, tone, length. What each one changes in the pipeline, and the briefs that produce drafts nobody has to rewrite.
Wire copy, aggregation and AI drafts: three ways newsrooms fill the page
Each method has a cost, a speed and a failure mode. A comparison of how they source, how they attribute and who is accountable when a fact is wrong.