How we know what mattered

Every morning we scan the AI newsletters people actually subscribe to and the technology press underneath them, and rank the day by how many of them covered the same story. Every ranking here is a count, not an opinion.

The newsletters people actually subscribe to

The roster is the list someone new to AI would end up on: the daily roundups that reach the most people, alongside the analysts and researchers those roundups quote. Every one of them is credited by name on every story it covered, and linked to its own signup.

The technology press underneath them

The newsletters tell us what was picked up. The press tells us what happened, in the words of the desk that reported it, alongside the labs' own announcements. Every story here is grounded in that reporting and links back to it.

We count how many of them ran the same story

Every morning each edition is scanned, the stories are matched across sources, and the ranking is how many independent newsletters landed on the same event. That makes the order a measurement. Every other list of the day's biggest AI stories is somebody's judgment.

We show where they disagree, not just that they agree

Two newsletters covering one acquisition rarely frame it the same way. Each story page quotes how each source put it and states the split, because the disagreement is usually the informative part and averaging it away destroys it.

We record who had it first

Each source keeps the date it ran a story, so the archive knows who broke it and how long the others took. 166 stories so far had a source ahead of the rest. No newsletter publishes its own lead time, and no public archive of these newsletters exists to work it out afterwards.

Two agents do the reading. A person checks the work.

One agent reads each newsletter as it arrives and pulls out the stories it covered. A second writes the brief from what all of them said, and quotes how each source framed it.The ranking itself is not theirs. Matching stories across sources and counting how many ran each one is ordinary code, so the order of the page is arithmetic rather than a model's judgment. A person checks the edition before it stands.

Nothing is deleted

Every story keeps its URL permanently. The archive is the point: coverage patterns only mean something across months, and a story that resurfaces reclaims its original page rather than becoming a second copy.

Limits

Three things the number on every card is not, worth knowing before you quote it.

The count measures attention, not importance

It records how many editors picked a story up. Newsletters follow each other and the same wires, so attention clusters, and a story every one of them ran is a story that was hard to miss rather than necessarily the one that mattered most.

The roster is 30 of hundreds

The roster is a choice, not a census. A story nobody on this list covered does not appear here at all, and will not show up as a low count. Absence on this site is evidence about our 30, not about the world.

Within a single day, nobody is first

Sources that ran a story on the same date are treated as tied, because this data records the day something published and not the hour. When we say who had it first, we mean first by a day or more.

Think we are missing one?

The roster is not fixed. If a newsletter or a publication belongs here, tell us and we will add it. Every addition makes the count mean more, so we would rather hear about the ones we have missed than defend the list we have.

Suggest a source