How humans and AI work here

Who decides what gets covered. The people on the masthead choose which public bodies, channels and record systems the newsroom follows, and add or retire them by hand. The tool does not pick its own beats.

Who reads a draft, and what that involves.An editor reads the draft against its source: the recording, the filing, the document. They correct names, figures and attributions, cut what the source does not support, and add local context the record leaves out. Every edit is kept in the article’s revision history, and the article names the editor. When an editor has read the whole piece and stands behind it, the article says reviewed by, with their name. When no one has, it says that instead.

Who talks to people. Editors contact sources, officials and residents, and what they learn goes into the article under their name. The tool never corresponds with anyone on our behalf. If you write to this newsroom, a person answers.

What happens when a reader challenges a claim. A person checks the claim against the source, corrects the article on the page where it appeared, and records the correction. See corrections.

Which articles publish without a person reading them first. Some do, and each one says so on the article. That is a choice about volume we would rather be honest about than hide, and it is the reason every article links its source: the record is there for you to check us against.

The line at the top of this page is accurate, and it is useful. AI is not a separate mind that appeared on its own. It is an artifact: code, data, hardware and training loops designed by people, trained on human-produced records of the world, and pointed at problems humans choose.

What it extends:

• Scale. A person holds a few ideas in working memory. A model can search, compare and recombine far larger volumes of text, code, images or measurements in seconds.
• Speed and consistency. It does not tire of looking at the same class of pattern. That is valuable for drafting, translation, summarization, code scaffolding and first-pass analysis.
• Coverage. It can surface connections across domains that a single specialist would not have time to keep current.

AI is a high-leverage instrument. Used well, it multiplies the reach of human reason, memory and craft. Used poorly, it multiplies error, slop and concentrated power. The difference is still in how people specify the job, check the output, and decide what is worth amplifying.

That is how we use it at Alaska News. Alaska’s public bodies hold more meetings and publish more records than any newsroom our size could sit through or read. We point the tool at that: every recorded meeting is transcribed, every filing and release is read, and a first draft is written from the source. The tool extends our reach. It does not replace our judgment, and it does not stand in for a name. Why a five-person newsroom took this on, and the numbers behind it, are on the about page.

Every AI-assisted article says so at its foot, with a link to this page, so you do not have to go looking for it. When an article credits one of our AI personas, the persona’s profile says it is an AI and not a person, and the article’s structured data credits Alaska News as the author rather than presenting the persona as a person.

An article is drafted from something: a recorded public meeting, a filing, an agency release, a document. That source is named on the article and linked where the link is publishable, for signed-in readers and anonymous ones alike. Where we recorded the meeting ourselves, the full transcript is published alongside it.

This is the check that matters most.An article has to represent its source faithfully: what the council voted, what the filing says, what the trooper report claims. If our summary and the source disagree, the article is wrong. A source can be wrong too, and an agency release is a claim, not a finding, so we say who is asserting what and keep a source’s claims distinct from what is independently established. Tell us at corrections.

How review works at Alaska News:

• Every AI-assisted article is marked on the page and links here.
• Every AI draft goes through an automated review before it can publish.
• Some stories are held for an editor before they publish.
• Editors edit or review nearly every article, most after it goes live.
• When an editor completes a full review, the article credits them by name.

We describe it here once rather than on every article, and no article claims a review that did not happen. This is a change we made to ourselves: the footer once told every reader an article was “reviewed by editors” regardless.

No model is responsible for what appears here. Responsibility sits with the people named on the masthead, whether or not a human read a given piece before it went out. “The AI wrote it” is not a defence we will offer.

No invented people. Our AI personas are labelled as AI. We do not give them fabricated biographies, invented credentials, or generated headshots presented as photographs of a person.

No invented facts, quotes or sources. A quote in an article is a quote from the source material. Where a claim cannot be traced to a source, it does not run.

No AI-generated images presented as photographs of a real event, place or person.

No AI correspondence with sources. If a person emails this newsroom, a person answers. We will not let anyone believe they are corresponding with a human when they are not.

AI drafting is fast and confident, and confidence is not accuracy. The failure we watch for is a summary that reads well and misstates what was decided in a room. That is why the source is on every article: not as a formality, but because it is the thing that lets you catch us.

How the pipeline works, end to end, is on how it works.