How Edition Uses AI to Write Your Morning Paper

At four in the morning, most cities are quiet. But somewhere in that stillness, an AI newspaper is already being made. Edition is pulling from over fifty sources, reading thousands of articles, and beginning the work of deciding what matters. all before anyone has reached for coffee. This is the story of how that works, and what the machine does and does not do on its way to producing a finished broadsheet.

The raw material

Everything begins with sources. Edition monitors a curated list. wire services, national papers, specialist publications, a handful of independents. across politics, technology, business, science, culture, health, and international affairs. These are not chosen by algorithm. They are chosen by editorial judgment: does this outlet produce reliable, substantive reporting? The list is maintained the way a newspaper editor maintains a contact list. Sources that go dark, drop in quality, or wall off their reporting get reviewed and sometimes removed.

Throughout the night, the system checks each source for new material, captures the full text, and files it away. By the early hours, the pool is deep. hundreds of fresh articles from dozens of publications, spanning every major beat. This is the raw material. None of it has been touched yet.

Reading the room

The next thing that happens is classification. Every article in the pool gets read. not by a person, but by a language model making the same judgment a section editor makes forty times before breakfast: what is this story actually about?

A story about a tech company facing an antitrust hearing could belong in technology, or in politics, or both. A medical study with economic implications sits at the border of science and business. The model reads the piece and tags it, sometimes with multiple categories, because stories rarely fit in a single box. This is not a keyword search. It is a judgment about what the story means, made by reading it.

Sorting signal from noise

When a major story breaks, every newsroom covers it. A single event can produce thirty articles across different outlets. all reporting essentially the same facts, with minor variations in emphasis or sourcing. If the system simply picked the most recent stories, a reader's newspaper would contain five versions of one event and miss four different stories entirely.

So the system clusters. It groups articles about the same event or development and treats them as one story with multiple tellings. When it comes time to select, it picks the strongest version from each cluster. the one with the deepest reporting, the most complete facts, the best source material. and moves on. One story, told well, instead of three told thinly.

The personal cut

This is where each reader's newspaper diverges from every other. You have told Edition what you care about. technology and foreign policy, say, but not sports. Maybe you have weighted business higher than culture. Maybe you have excluded celebrity news entirely. The scoring system takes the full pool of clustered, classified articles and evaluates each one against your profile.

Several signals factor in. Editorial importance: is this a story everyone should know, regardless of personal preference? Topic relevance: does it match your interests? Source diversity: has the system already leaned too heavily on one publication? Freshness: when was this reported? Story trajectory: is this a developing situation that has been building for days, or a one-day item?

The result is a ranked list. your list, different from anyone else's. The top stories make the cut. How many depends on the size of your edition: a two-page broadsheet holds fewer stories than a four-page one, and the selection accounts for that. Nothing is chosen for its ability to hold your attention or provoke a click. The only question is whether you need to know it.

How AI journalism actually works here: the rewrite

This is the step most people do not expect, and the one that defines the reading experience. Every article selected for your newspaper is rewritten.

Not summarized. Not trimmed. Rewritten from the original reporting into a consistent editorial voice, at a precise word count, for a specific slot in the broadsheet layout.

The reasons are practical. When you pull articles from fifty different publications, you get fifty different writing styles. A formal Reuters dispatch, then a conversational tech blog post, then a dense policy analysis. Reading them back to back is jarring. your brain has to re-calibrate with every story. Rewriting brings everything to a single standard. Your newspaper reads like one publication, not a scrapbook.

Then there is the question of fit. A broadsheet has slots of specific sizes. The lead story might call for 350 words. A secondary piece gets 200. A brief gets 100. Original articles almost never arrive at those lengths. The rewrite tells each story completely and coherently at the exact count the layout needs. No story ends mid-thought. No story wastes its first paragraph on context that made sense on the original website but does not belong in a newspaper.

And every rewritten piece credits its source. The original publication, the original reporting. always attributed, always linked. Edition does not do original journalism. It curates and rewrites, the same function newspaper editors have performed for over a century, applied across a broader range of sources than any single masthead covers.

Typesetting a newspaper at 5 AM

With the articles rewritten, the system does something that would have occupied a compositing room full of people a generation ago: it typesets a newspaper. Real columns. Weighted headlines. Proper leading and kerning. Section breaks. Page numbers. A masthead with a date.

The layout engine handles the quiet, essential work. fitting variable-length stories into fixed page slots, ensuring headlines do not collide, maintaining column widths, managing hyphenation. The result is a print-ready broadsheet. It can be read on a screen, or printed and read on paper. Either way, it has the spatial logic of a real newspaper. the kind of layout where you can take in a full page at once and understand the hierarchy of stories before reading a single word.

Then it waits. Each reader has set a delivery time. The paper is generated in advance so it is ready at the exact minute. When that minute arrives, the finished newspaper arrives. complete, designed, with a last page.

What the machine does and does not do

The machine classifies, clusters, scores, rewrites, and typesets. It handles the production work that would otherwise require a team of editors and designers working through the night.

The machine does not report. It does not verify facts beyond what appears in the source material. It does not invent information. It does not editorialize past the boundaries of the original article. If the original reporting is wrong, the rewritten version will carry that error forward. Edition's quality depends entirely on the quality of its sources, which is why source selection and monitoring are treated as infrastructure, not an afterthought.

The machine also does not watch you read. There are no read-time metrics feeding back into story selection. There is no behavioral loop. no "you spent forty seconds on this, so here are more like it." Personalization comes from what you explicitly tell the system you care about. Nothing more.

The invisible craft

Newspaper production was always invisible. Readers picked up the morning paper without thinking about the overnight editors, the wire service agreements, the compositing room, the delivery trucks. The machinery was hidden behind the masthead.

Edition rebuilds that machinery for a different era. The sources are digital. The editing is automated. The delivery is electronic. But the output is what it has always been: a finite, designed, editorially coherent newspaper that arrives in the morning and has a last page.

The craft is interesting. But the newspaper is the point.