GazetAI is an AI news product. A language model writes every summary and every layer of analysis you read here. That is a reasonable thing to be sceptical about, so this page describes exactly how a story reaches you — and, in more detail than the rest, what this product does not do.

Where the news comes from

We do not employ reporters and we do not publish original reporting. Stories are drawn from Event Registry, a global index of more than 50,000 news sources that groups articles about the same real-world event together. Our pipeline runs on a schedule through the day rather than continuously, so a story can reach your feed some hours after the outlet that reported it.

How one story per category is chosen

For each of the 11 categories we take the events available in that period and score the articles within them, keeping exactly one. Nothing here is a human editorial judgement, and nothing here is a judgement about whether a story is true. The signals are:

How highly the index ranks the outlet
How much article text there is to work from
How recently it was published
Whether it carries an image
Whether it carries a named author rather than only an outlet name
How closely it matches the centre of the event's cluster of articles

What the AI writes

The selected article is read by a language model, which produces the summary, the key takeaway and the analysis layers — sentiment, world impact, pros and cons, market read, watch points, historical parallel and the rest, 19 in total — and then translates all of it into the nine languages the app supports. Everything in that list is generated. None of it is quoted from the original article, and none of it is checked against another source.

What GazetAI does not do

These are the limits that matter most when you decide how much weight to give what you read here. They are structural, not temporary shortcomings we are about to fix.

We do not fact-check

No claim in any article is verified against a fact-checking database or a second source. Nothing on this site should be read as a statement that a reported claim is true — only that it was reported.

The summary comes from one article, not many

Even when dozens of outlets covered an event, the text you read is written from exactly one of them. It carries that outlet’s framing and its errors. Synthesising across sources is a different and much larger job than the one this product currently does.

We do not rate outlets for bias or reliability

You will not find a political-lean label or a trust score anywhere in the product, because nothing in our pipeline can source one. A model guessing at an outlet’s bias and presenting it as an assessment would be worse than saying nothing.

Confidence and sentiment are the model judging its own output

Where the app shows a sentiment or a confidence signal, that is the language model’s own assessment of the text it just wrote. It is not an external measurement and it is not evidence of accuracy.

Where we show how many outlets covered a story, it is a floor

That number counts English-language outlets only, is capped at 100 rows, counts wire copy republished by many outlets as many outlets, applies no quality filter, and is measured once when the story is ingested rather than kept up to date. It describes how widely an event was reported. It is not a count of sources behind our text.

We do not cover everything

One story per category per run is a deliberately narrow selection. A significant event can be missed entirely because a different one scored higher in the same period.

Corrections

Every article in the app and on this site carries a control to report a problem with it — a factual error, a mistranslation, or analysis that misrepresents the story. Reports reach a person, not a queue that nobody reads. If an article is wrong we remove or correct it; we do not silently edit a story and leave the original date on it. If a correction affects what this page describes, this page changes too.

AI disclosure

Every summary and every analysis layer in GazetAI is generated by a language model. This is disclosed in the app on each article, and it is stated here. We do not present generated text as human-written reporting anywhere in the product.

Changes to this page

This page describes how the product actually works today. When the pipeline changes, this page is meant to change with it, and its revision date above is meant to move. If you find something here that no longer matches what the product does, tell us — that is a defect in the same way a broken feature is.

Contact: support@gazetai.com

Editorial Standards