What is an AI news briefing, and how is it different from a newsletter?

An AI news briefing is a short, written summary of what happened in a subject you chose, put together by software that searched for it, read it and decided what was worth your time. The difference from a newsletter is who it is written for. A newsletter is written once for thousands of readers. A briefing is written for you, about the thing you described.

How an AI news briefing is made

The tools differ, but the good ones follow the same four steps.

• Search. Your description is turned into a set of searches, run across the news and the wider web.
• Read. The promising results are opened and read in full, not judged by their headlines.
• Judge. Each story is weighed against what you said you care about. Most are dropped.
• Write. What is left is written up: a headline, what happened, why it matters to you, and the sources.

In Relay, the thing doing this work is called an agent. You give it a name and a brief, which is a few sentences about what to watch for, and it gathers on the days and at the time you set. A run produces a handful of stories, rarely more than ten.

What it replaces

Most people who set one up are replacing a habit, not a product: a round of tabs every morning, a folder of newsletters they skim, or a set of keyword alerts that send links without saying which ones matter.

A briefing differs from each of those in the same way. It reads the article for you and tells you why it is in front of you. If nothing worth knowing happened, it says so, and that is a result too.

When a newsletter is still better

A good newsletter has an editor with taste and sources of their own. If someone already writes a newsletter about exactly your subject, keep reading it. A briefing earns its place when your subject is too narrow for anyone to write a newsletter about: one regulator, one market in one country, four competitors, a single piece of legislation.

What makes one good

Three things decide the quality of a personalized news digest, and none of them is the model.

• The description. Vague in, vague out. Naming organisations, places and products gives the search something to hold.
• Sources you can check. Every story should link to where it came from.
• A way to correct it. In Relay you save a story, ask for more like it, or say you are not interested, and the agent adjusts.

Trying one

Relay is in private beta. The guide walks through setting up an agent, and the public briefs show what the result reads like before you make one of your own.