The paper
How this paper is made
Yes, we use AI. Here's exactly what it does, what people do, the rules it works under, and why we think it can be good for a local paper.
Plenty of people in Brighton are wary of AI, and for decent reasons: slop, made-up facts, other people's work taken without credit, energy bills nobody talks about. We share most of those worries. So here is everything, plainly, and you can decide for yourself.
Why use AI at all
Local news has been shrinking for years. What's left often pays its way with pop-up adverts, tracking and headlines built to make you cross.
A daily paper with no adverts, no paywall and no clickbait, about what's going right in the city, doesn't pay for itself the old way. AI is what lets a very small team do it, and do it carefully. We'd rather use it well, in the open, than not have the paper at all.
Who does what
- Finding
- Each morning a script reads local news feeds, the council's news and meeting papers, and community groups.
- Sorting
- Jev, a small judgement model made by TypeSafe, answers the same narrow questions about every candidate story. Is it local? Is it useful or uplifting? Is it written to make people afraid or angry? Is it an advert in disguise, or party campaigning? Can readers take part? Code then applies our rules and drops crime, tragedy, adverts and anything that isn't local.
- Writing
- Claude, a model made by Anthropic, researches and drafts each story. It saves every source page before it writes a word, and it goes to the primary source (the council paper, the organiser, the published list) wherever there is one.
- Checking
- Machines check every sentence against those sources (more on that below).
- Deciding
- A person reads every story, settles every flag by hand, and can pull any story at any time.
The rules the AI works under
- It never invents a quote, a figure, an address or a link. If it isn't in a source, it isn't in the story.
- It never copies another newsroom's writing. When the Argus or Brighton and Hove News broke a story, we say so and link to them, and we go to the council or the organiser for the facts.
- It never makes pictures. Every photo in The Shingle is real, openly licensed and credited to the person who took it. No AI images, ever.
- It doesn't do crime, tragedy or outrage, and it doesn't campaign. We cover good work in the city and credit whoever did it: the council, any party, or none. We're progressive in what we choose to cover, never in how we report it.
The full rules are in our editorial policy. What we cover: food, culture, the sea and the Downs, community projects, local people, and the decisions that change daily life in the city. When a hard story matters, we report the problem plainly and spend most of the words on what's being done about it.
How we check
Before anything goes live, every sentence in the headline, standfirst, story and "Good to know" box is checked against the saved sources:
- Every quotation must appear word for word in a source. If it doesn't, the story can't be published.
- Every number, price, date and time must appear in a source.
- Jev reads the closest source passages and says whether they back the sentence up, contradict it, or say nothing about it. It also flags any sentence that adds a detail, or says something more strongly than the source does.
Anything the machines can't match, the editor checks by hand. The sources are listed at the foot of every story so you can check them too.
You can correct us
Every story has a "Suggest an edit" link. Spotted a mistake, a missing detail, or something that isn't fair? Tell us, and a person will read it. We fix mistakes in the open, add a dated note to the story and list it on the corrections page. You can also send us a story.
Bylines
Stories carry the byline "The Shingle desk", because each one is drafted by AI, checked by machine and approved by the editor. Putting a made-up human name on them would be a lie. The masthead says who is responsible.
The footprint
Running a paper with AI uses electricity, and you should know where it goes.
- The small part. Sorting a morning's news with Jev takes very little: on our first day it read 137 stories in about three seconds. Our pages are plain and light, with no ad scripts or trackers, and they're served from Cloudflare's network close to you.
- The bigger part. Drafting and checking stories with a large model takes real computing, in data centres we don't run.
- What we do about it. We make the paper once a day in a single batch rather than all day long. We keep stories short. We don't regenerate anything that hasn't changed.
- What we're exploring. Choosing AI and hosting providers that publish their energy and water use, running the daily batch where and when the electricity is cleaner, and publishing a short note each month on how much computing the paper used.
Money and ownership
There are no adverts, no sponsored posts and no paid placements. Nobody can pay to be in The Shingle. If the paper ever earns money, the plan is to spend it first on local people: writers, photographers and illustrators from the city.