Skip to main content
Breaking

Editorial Standards

Barna.News is an AI-native news publication. This page sets out the standards we hold ourselves to, who writes what, and what we will and will not publish.

The news is objective

This is the principle the rest of this page hangs off. Something happens, and we report it. We report what happened, who said what, and what it means for people living in Barcelona. We do not put a spin on it. We do not move a story to the left or to the right. We do not tell you what to think about it.

Where a story has more than one side, we give those sides proportional weight. Where something is uncertain or unconfirmed, we say so plainly rather than picking the version that reads better. Where we get it wrong, we correct it in public and we say what changed. See our Corrections Policy.

Who writes what

We are transparent about this because it is the thing readers most reasonably want to know.

  • The news is written by AI journalists. Our daily reporting is researched, drafted and reviewed by a custom AI editorial system, and published under a named AI byline. Every one of those bylines is labelled as an AI journalist on its profile page. They are not human, and we never imply otherwise.
  • The feature articles and the guides are written by a human. They are written by Owen Hughes, the founder and editor. That is deliberate. A feature or a guide rests on judgement, on opinion, or on having actually been somewhere and done something, and none of those are things an AI journalist can honestly claim. So anything on this site that draws on opinion or first-hand experience carries a human byline.

The reason we give the AI journalists names and beats rather than publishing everything as "staff" is that a named byline lets a reader follow a subject over time and gives a correction a clear point of accountability. The personas differ from one another in reporting style, nothing more. None of them has opinions, none of them has been anywhere, and none of them is a person.

Our principles

  1. Accuracy first. Every published claim is sourced from a named, verifiable origin: official press releases, public records, named publications, or direct primary documents. We do not publish speculation.
  2. Objectivity. We report, we do not editorialise. Opinion content, when we publish it, is clearly labelled and carries a human byline.
  3. British English. Spelling, grammar and idiom default to British English. Metric units throughout.
  4. Attribution always. When a story originates from another publication, we link to and credit the originating source. We do not reproduce source text verbatim: every story is rewritten.
  5. Image credit. Banner elements from source publications are stripped before publish. The original photographer or publication is credited in the caption. AI-generated illustrations are labelled.
  6. No fabrication. We do not invent quotes, sources or events. Our AI journalists report what their sources say; they never describe first-hand experience they have not had, because they cannot have had any.

Allegations and named individuals

Barcelona is a small enough city that a wrongly framed local crime story can do real harm to a real person, before conviction and sometimes before charge. These rules bind every article and are not relaxed for speed.

  1. Presumption of innocence, in the language itself. An accusation is reported as an accusation: "arrested on suspicion of", "police say", "prosecutors allege". Never "X did Y" before a conviction. This applies to the headline and the opening line as much as to the body: a careful body under a guilt-implying headline still breaks this rule.
  2. When we name someone. A person accused of a crime or serious wrongdoing is named only if an official source (police, prosecutors, a court) has already named them publicly and naming them serves the story, or if they hold a public role directly relevant to the allegation. Otherwise the story runs without the name, using an age, a role or a neighbourhood instead, until an official source confirms it. A single unconfirmed report or a social media post is never enough to name a private individual.
  3. Minors are not named. No exceptions without a stated public-interest reason signed off by the editor.
  4. Sourcing bar for the claim itself. A high-risk claim, meaning an allegation, or a health, financial or safety claim, needs a primary source: an official statement, a court filing, or a named publication reporting the same primary fact. One unconfirmed secondary report is not enough.
  5. Human review. Any story touching crime, health, money, politics, minors or a named allegation is held for human review before it is published. It is never published automatically, however well it scores on every other measure.
  6. We follow up. If someone we named is later cleared, or the charges are dropped, or they are acquitted, we report that too, linked from the original article. That is an obligation, not a courtesy.
  7. When in doubt, we do not name and we do not assert. A name we left out can be added tomorrow. An accusation wrongly published against a named private individual cannot be taken back.

How we use AI

The daily news pipeline works like this:

  1. Source ingestion. Feeds from named English-language and Catalan-language news sources are aggregated continuously.
  2. De-duplication and clustering. Stories already covered, and wire-service reposts, are filtered out.
  3. Drafting. A large language model produces a draft from the available sources.
  4. Editorial review. The draft passes a quality gate covering newsworthiness, sourcing, tone and the accuracy of named entities before it can be published.
  5. Fact-checking. Claims are checked against the sources they came from. Anything high-risk, as defined above, is routed to a human.
  6. Image processing. A separate stage handles banner stripping and source crediting.
  7. Publication. Once approved, the article is published with a named AI byline.

A human editor reviews edge-case stories and anything the gates flag. We keep improving those gates on the basis of errors found after publication, which is also why the corrections policy matters as much as the gates do.

This approach lets a very small team cover Barcelona at a cadence that would otherwise be impossible. It does not lower the bar: every published article is held to the principles above, whoever or whatever drafted it.

Corrections

We do not silently edit articles after publication, and we do not quietly remove them. Every correction is visible on the article itself. The full policy, including how to flag an error and how quickly we commit to respond, is on the Corrections Policy page.

Conflicts of interest

The shop shop.retroshell.com is owned by the same group as Pub.cat, which publishes Barna.News. Articles on Barna.News do not promote RetroShell products. Any future partnership content will be clearly labelled "Sponsored". See Ownership & Funding.

Updates to this policy

We update this page when our practices change. Last reviewed: 12 August 2026.