Disaster, emergency, and conflict reporting: a practical guide for 2027

What to take away

Emergency coverage has two jobs at once. It must help people act on reliable instructions now, and it must create an accurate record of what happened. Speed matters, but an unsupported location, count, or route can expose people to harm.

Open with an action box

When verified protective instructions exist, place a short box above the narrative.

Sample line: Evacuate [street or district] via [route] before [time] local time. Order issued by [agency] at [time]. Next official update at [time] on [channel].

Copy the instruction accurately. Do not add advice from memory. If several agencies issue different directions, identify each jurisdiction and seek clarification. The same fidelity rule governs reading an official document: quote the operative text, not a memory of it.

Establish the event identity

Create a stable record for the incident: official name, local name, start time, time zone, location, event type, reporting authority, and document identifiers. Use geographic boundaries rather than a broad place name when only part of an area is affected.

For conflict coverage, record who made each claim, what location and time it covers, and whether the newsroom confirmed it independently. A party to a conflict can be a primary source for its own statement without being an independent source for the event.

Separate four information states

Do not promote a report to a verified fact because several accounts repeat the same origin. Trace the chain and seek an independent route to the underlying evidence. The four working states match the table in developing news stories.

Build count ledgers

Keep deaths, injuries, missing people, rescues, evacuations, shelter registrations, damaged buildings, destroyed buildings, customers without service, and assistance applications in separate columns. Each count needs a source, coverage area, cutoff time, definition, and version.

The two rows are illustrative. Replace them with counts your newsroom can name and time.

Illustrative run, not a real incident: a county flood. The action box tells residents to leave the low-lying area east of the levee before 20:00 ET and names the issuing agency. The identity record fixes the event name, start time, time zone, and boundaries.

One ledger row carries count, source, coverage area, cutoff, definition, and version. The update log records the 16:00 check and the revised count.

Ask whether a number counts people, incidents, households, structures, visits, or applications. Count hygiene of this kind is the whole lesson of the developing story desk case, where two counts described different assets.

Communicate uncertainty early

The CDC's Crisis and Emergency Risk Communication manual provides an evidence-based framework for organizations communicating during major emergencies and includes guidance on audiences, planning, spokespeople, and media work. Its scope is organizational emergency communication, so a newsroom should adapt its principles while retaining editorial independence.

In practice, say what is known, what is not known, who is checking, and when the next update is due. Put the main uncertainty beside the claim, not under a generic caveat at the end. The early health alert desk case runs that practice through a full day of updates.

Verify visual evidence

Preserve the original file and source post. Check upload time, claimed location, landmarks, weather, shadows, signs, terrain, prior appearances, edits, and audio. Contact the creator when safe. A matching place does not prove the stated time or cause.

Manipulated video adds a failure mode that location checks alone can miss. The Journalist's Resource roundup of research resources on deepfake video gathers five papers explaining how convincing fabrications are generated and how researchers detect them. It dates from 2019, so treat it as background rather than a current tool list.

Combine provenance and detection checks. For provenance, look for C2PA Content Credentials, the signed capture and edit history promoted by the Content Authenticity Initiative and written by some cameras and editing tools.

For detection, run the clip through the InVID-WeVerify verification plugin, Google Lens or Yandex reverse image search, and FotoForensics for compression and error-level analysis. Detection tools flag likelihood, not proof.

Do not publish images that expose a shelter, military position, rescue route, private medical information, or a person's identity without weighing the risk. Blur or withhold when disclosure could cause harm, and explain material edits.

The photograph documents coordination during Hurricane Gustav recovery. It does not depict a current response or prove the effectiveness of any agency.

Maintain a live update log

Use a visible time and time zone. For each update, record what changed: order expanded, road reopened, count revised, source corrected, or claim disproved. Replace obsolete safety instructions prominently while retaining an archive for accountability. The log doubles as the private ledger of a news timeline.

Copyable log line: [time and time zone] | [what changed] | [source] | [next scheduled check].

An update that says "no change" can still be useful when the newsroom promised a scheduled check.

Set the cadence and publish it: hourly while a warning or evacuation order is active, every two to three hours during recovery, and once a day after the response closes.

Common questions

Should an unconfirmed casualty figure be published?

Only with a clear source, status, cutoff time, and reason it matters. Do not place an unverified figure beside a verified total as if they are equivalent.

Can social media confirm an emergency location?

It can provide evidence, but a post needs origin, location, time, and context checks. Reposts are not independent confirmation.

When should a live article receive a correction note?

Use one when the change affects safety, location, identity, responsibility, scale, or the reader's understanding of events.

How should a reporter describe access limits?

State the blocked area, period, and verification work that could not be completed. Do not convert lack of access into proof that an event did or did not occur.