How we use AI

WeNepal is AI-assisted and human-reviewed. This page explains what that means in plain language: where stories come from, what the software does, what it is not allowed to do, and what the labels on every story tell you.

• Software reads trusted outlets and official sources and writes a short summary in its own words.
• Every fact in that summary is checked against the sources before anyone sees it.
• Routine stories that pass every check can publish on their own, and say so.
• Sensitive stories always wait for a human editor.
• Every story names its sources and carries a label that tells you how it was made.
• When we get something wrong, we correct it in public.

Our engine follows a list of trusted Nepali and international news outlets and primary sources: Nepal Rastra Bank, NEPSE, the Election Commission, the Supreme Court, the Department of Hydrology and Meteorology, the Nepal Gazette and ministry press pages. Each source has a trust score that goes down if it is repeatedly wrong and up if it holds. State-owned media is labelled as such.

The engine only reads what sources publish. It does not go looking on social media, and a social post counts as a source only when the account is verified as the person or institution.

When two or more independent outlets report the same event, or one primary source does, the engine writes a summary of 80 to 120 words, three bullet points, a "why it matters" line and a source list. It writes the Nepali and English editions separately, in each language's natural order, rather than translating one into the other.

The drafting rules are strict:

1. Use only facts found in the sources provided.
2. Every number, name, date and quote must appear word for word in at least one source.
3. Write in your own words; never copy a sentence.
4. Always end with the credit line naming every source.

Every draft goes through a set of checks before a person or a reader sees it. A draft that fails a check is rewritten once; if it fails again, it is held for an editor.

Stories fall into two groups.

Routine stories in categories such as markets, weather, sports results and scheduled government announcements can publish automatically when they have at least two independent sources, pass every check and raise no sensitivity flag. These carry the label "Auto-published · routine". A desk editor reviews a sample every day.

Everything else waits for an editor, and these stories always do: elections, court cases, deaths and violence, communal tension, named individuals facing allegations, health emergencies, national security, anything involving minors, and anything the checks flagged. An editor can approve, edit, reject or ask for a rewrite. These carry the label "AI-assisted summary · Reviewed by WeNepal".

Any editor or engineer can pause the entire newsroom at any time.

If a story is wrong, we correct it and say so on the story, dated. Major corrections are also posted on the channels where the story appeared and listed on the public corrections page. A story that should not have been published is retracted, not quietly deleted. Sources that turn out to be wrong lose trust; sources that are wrong repeatedly are dropped.

• Invent a quote, a number, a name or a date.
• Publish a story in a sensitive category without an editor.
• Generate an image of a real person, or clone a real person's voice.
• Copy another outlet's paragraphs, headlines, photos or video.
• Publish without naming its sources.
• Hide that it was involved.

Write to hello@wenepal.com. To report an error, use corrections@wenepal.com or the contact form. Our full rules are in the editorial policy.