How we report

Happened in AI exists to help you understand the important AI developments in a few minutes, and to show you the evidence behind each one. These are the rules we work by.

Every story answers five questions

What happened?
Why does it matter?
What is confirmed, and what is still a claim?
Where is the original evidence?
When was this published, and when was it last updated?

If we cannot answer the first four, we do not publish the story yet.

Sources

We monitor company newsrooms, research feeds, regulators, and independent news organizations. We treat them differently:

A company announcement tells us what a company says it has done. It is not independent proof of how well a product works or what effect it will have.
Independent reporting and documents (court filings, regulatory records, financial reports, datasets) are how we test significant claims.
Preprints are research papers shared before peer review. We label them as preprints every time.
For medicine, we keep four things apart: a preprint, a peer-reviewed result, a regulatory authorization, and evidence that patients actually benefit. One does not imply the next.

We link to the original source wherever we can, so you can check our work. Each source on a story is labelled with its type.

Verification

Our verification standard is to open primary documents and check figures, dates, availability and limitations. A story’s preparation and human-review status are separate from the evidence it cites. Consequential claims need a second, independent source. When we cannot confirm something, we say so in the story under Claimed, not independently verified or What we don’t know yet.

Stories with the Happened in AI Automated Desk byline are researched and published by software without a human review of each story. Before publication, the workflow checks that at least one primary source and an independent report or public document are present, that short quoted evidence passages appear on the linked pages, and that the story passes the site's content and build checks. These checks cannot guarantee that every interpretation is correct. If the sources cannot be reached or the checks fail, the workflow skips the story. We invite readers to open the linked evidence and report errors.

Stories with the Happened in AI AI Desk byline are AI-prepared drafts from an owner-directed editorial batch. Their visible review label states whether human review is pending. A shared Editors byline on older coverage does not establish that a human review was recorded.

Labels

News reports what happened.
Analysis interprets verified facts. Its conclusions are the writer’s.
Explainer gives background you need to follow the news.
Research watch covers new papers and results, with their stage of review.
Opinion is an argument, and is labelled so you can tell it apart from reporting.
Developing (in amber) marks a story where the facts are still coming in.

Corrections

When we get something wrong, we fix it, note the correction at the bottom of the story with the time we made it, and add it to our corrections log. We do not quietly change facts. Small fixes to spelling or style that do not change meaning are not logged.

How we use AI

We use AI tools to help monitor sources and to suggest draft summaries from the documents we have collected. AI tools never invent quotes, results, or citations for us, and they do not decide what is true. AI-prepared coverage is labelled explicitly. An AI-prepared item remains an editorial draft until a person reviews it, even if it has been made publicly available. A “human review pending” label means no completed human verification is claimed. Human review and corrections are required to finalise this work; public availability or an editorial byline must not be treated as proof that review occurred.

AI tools help monitor sources and suggest draft summaries. The Automated Desk uses a scheduled ChatGPT Work cloud task and an image generation tool to research, write, and illustrate eligible stories for direct publication under its own byline. The workflow checks source passages and site rules before publishing, but automated checks can miss errors. It does not claim that a human editor reviewed those stories. The publication owner is responsible for the workflow and its corrections.

Images

We prefer original diagrams and charts made from verified facts, with sources in the caption. When we use a generated illustration, its caption and alternative text say AI-generated illustration, and we never present it as documentary evidence. Photographs are used only under a licence that covers our use, and we record who made each image and under what terms. We do not take images from other news articles. Automated Desk stories use newly generated illustrations, not photos from source sites.

Writing

We aim to be clear, restrained, and curious. We avoid hype, unsupported predictions, and headlines that claim more than the evidence shows. We write our own summaries and link to other publishers rather than republishing their work.

Contact

To report an error, reply to any newsletter issue. A dedicated corrections address will be listed here before launch.