Fact-Checking Policy
Verification is not a stage at the end of writing here, it is most of the work. Nearly every actor on this beat runs a media operation, and the first account of an event comes from somebody who wants a particular version to stick. This page is how a claim gets tested. Last updated: August 2026.
What happens before a story goes out
The writer checks their own copy against the sources rather than against memory, and files a short verification log with it: who said this, when, were they there, and what would have to be true for us to be wrong. Names, place names, titles and figures are checked one at a time.
An editor reads the story against that log, and a sentence the log does not support is sourced properly or taken out. Anything carrying a security claim, a casualty or displacement figure, or an allegation against a named person goes to a second editor, whose job is to argue with it. If the log is thin, the story waits or runs with the gaps written into the copy.
Primary sources, and what counts as one
We prefer the document to a description of the document. The signed agreement rather than a briefing about it. The ministry’s statement rather than a screenshot of it forwarded three times. The agency’s dataset rather than the round number a spokesman read out.
A person who was in the place is a primary source. An official in a capital describing a district he has not visited is not, and we attribute him as what he is. When a district changes hands, the government spokesman’s announcement and the announcement from the group that took it are both claims written for effect, so we call people who live there: traders, a hospital, drivers on the road, the Somali-language press.
Showing our work
Where a source is public we link to it in the sentence that uses it, so a reader can check us instead of trusting us. Statements, gazette notices, court filings, budget documents and agency situation reports all get linked. Where the original is a PDF or a Somali-language broadcast, we say what it is and where to find it.
Some sources cannot be linked: someone speaking on condition of anonymity, or a document we were shown but cannot publish without identifying who gave it to us. There the copy says as much about the source as we safely can, and why the rest is withheld. Who approves that protection is set out on our editorial standards page.
Photographs and video
Pictures recirculated from older incidents are the most common misinformation on this beat, and they travel fastest in the hours after an attack. No image reaches a story from a messaging group without provenance. We run reverse image searches, read whatever metadata survived the upload, and question the picture itself: the season, the vehicles, the uniforms, whether the shadows agree with the time we were given.
Where we cannot establish that an image is what it is claimed to be, we do not use it. Where a picture is genuine but old, the caption says when it was taken. Witness footage is credited to whoever filmed it when naming them is safe. We publish no photographs of the dead, and nothing synthetic runs without the caption saying so.
Numbers
Casualty and death tolls come from parties with reasons to round in opposite directions: hospitals that count arrivals rather than deaths, officials who gain from a low number, insurgent channels that gain from a high one. We publish a range with the sources attached rather than an average, date every figure to the hour, and say when a toll is expected to move.
Displacement and aid figures need the same handling and get less of it. A number of people displaced is an estimate produced by a method, and the method matters more than the total: who counted, over what period, and whether people displaced twice were counted twice. An appeal figure is what an agency says it requires, not money received. With studies we read the paper rather than the press release, and note the sample, the period and who paid.
Social posts, screenshots and rumour
Diaspora platforms carry the story before we do, and the rumour cycle has a familiar shape: a claim of responsibility, a resignation, a death, a coup that is not happening, all travelling as screenshots with the original stripped off. A screenshot is not a source. We find the account, confirm it is the one it appears to be, and where a claim is attributed to an armed group’s channel we establish it came from that channel.
Material from readers and witnesses can be strong evidence, and we use it when it holds up. It clears what everything else clears: an origin we can identify, a place we can locate, a time we can fix, and a second thread of evidence that agrees. Where a claim is circulating widely and we cannot verify it, we sometimes write about the circulation itself, labelled unverified.
The same test for everyone
A claim gets the same scrutiny whoever makes it. The federal government, a member-state administration, an opposition figure, an aid organisation, an armed group: the questions do not soften for the powerful or harden for the unpopular. We have no candidate, no party and no capital we speak for, and we take money from none of them, as our ownership and funding page sets out.
Neutrality is not splitting the difference. Where the evidence points one way we say so plainly rather than balancing a verified fact against a denial for the look of it. Where we cannot tell which account is true, we say that.
When verification fails after publication
Sometimes a claim survives all of this and turns out to be wrong anyway. The story is then corrected on the page with a dated note saying what changed, through the process on our corrections policy page rather than a private tidy-up.
We also go back to the verification log and work out how it got through, which is the only part of this that improves the next story. We work to the principles of the International Fact-Checking Network and claim no membership we do not hold. If you can show us something here is wrong, write to support@somaliupdate.com with the headline in the subject line.
