The Photo Is Real. The Disaster Is Not. — Why the Wrong Images Spread After Every Event

Direct answer: in the first hours after a disaster the demand for images is enormous and the supply is almost nothing, because the people who could take them are busy surviving. That gap gets filled — usually not by fabrication but by recycling, where a real photograph of a different event is shared with a new caption. It spreads faster than the agency reports because it arrives sooner, looks more compelling, and asks nothing of the reader. The most harmful items are not images at all, but the rumours that travel with them.

Four kinds of wrong, and only one of them is fake

• Recycled. A real photograph from another disaster, another country, or another decade. This is by far the most common, and it is the hardest to doubt because the image itself is authentic — flood water looks like flood water everywhere.
• Recontextualised. A real image from the right event with a wrong claim attached: a rescue drill described as a rescue, a routine controlled release described as a dam failure, an old photograph of a building presented as new damage.
• Rescaled. One street presented as a city. A single collapsed structure becomes “the town is gone”. Nothing in the frame is false; the implied extent is.
• Fabricated. Generated imagery. Less common than the other three, and increasingly the hardest to rule out.

Why disaster scenes are the easiest thing to fake

Most people can spot an implausible photograph of a place they know. Almost nobody has a reference for what a flooded street in a town they have never visited should look like, so the usual check — does this match what I know — is unavailable exactly when it is most needed.

It is worth being blunt about the current state: advice to look for visual giveaways in generated images is no longer reliable. Guidance about hands, text and lighting was useful for a while and is not a defence any more. Telling readers they can identify fakes by eye makes them more confident rather than more accurate, which is worse than saying nothing. Provenance is the check that still works; appearance is not.

The part that actually hurts people

Wrong images distort how severe an event looks, in both directions, and they misdirect donations. But the genuinely dangerous material is the text that travels alongside them.

• Predicted aftershocks. After every major earthquake a message circulates naming a magnitude and a time — often a specific hour that night. Earthquakes cannot be predicted, which is a settled matter rather than an opinion, and no agency issues such forecasts. These messages move frightened people outdoors at night, sometimes into cold, into traffic, or away from buildings that were never at risk. What is real is that aftershocks follow a decaying pattern in time — that is a statistical statement about many shocks, not a schedule for one.
• Dam and barrage rumours. Claims of an imminent failure spread very fast in flood-affected areas and can trigger movement in the wrong direction, at night, through water.
• Water and supply rumours. Claims that supplies are contaminated or exhausted cause hoarding and panic, and can keep people from using water that was safe.
• Fake appeals. Donation requests attached to recycled images divert money from responders who are actually present.

Why this happens in every event, regardless of platform

It is a timing problem before it is a truthfulness problem. Verified information is slow because verification is slow — the same reasons set out in what “no alerts” actually means. Attention peaks in the first hours; official confirmation arrives later. Anything that fills that window travels without competition, and by the time the accurate version is published the audience has moved on.

Corrections also travel poorly. The recycled image reaches far more people than the correction ever will, which is why the goal is not to correct faster but to forward less.

Checking something before you pass it on

1. Reverse image search it. One search catches most recycled images immediately, because the older copies are still indexed.
2. Look for the same picture with a different caption. If it has been attached to two events, it belongs to neither claim.
3. Ask whether any agency has said it. A dam failure, an evacuation order or a death toll would be published by the responsible authority. If the only source is a forwarded message, that is the finding.
4. Check the season and the surroundings. Vegetation, clothing, signage, language on shopfronts and vehicle registrations often place an image somewhere other than where it is claimed to be.
5. Distrust precision. Real early reporting is vague. A message giving an exact magnitude at an exact hour tomorrow is describing something nobody can know.
6. Wait. Almost nothing is lost by waiting for the agency, and forwarding is publishing — in a disaster your forward carries your credibility with the people who know you.

If you are near the event

Your local knowledge is genuinely valuable, and so is your restraint. Report what you have seen yourself and say plainly that it is what you saw; do not relay what someone else sent you as though you had. Follow the national agency for anything with a decision attached — every one of our country pages links that country's own authorities — and, where an official channel is asking for reports, use that rather than a public feed.

What this site does about it, and what that costs

• We publish no disaster photographs of our own. Nothing on this site is an image we sourced, chose or captioned. Since August 2026 a reader using our Android app may attach one photograph to a report about an event they are at; it appears in its own section, labelled unverified, and never becomes a data point, a map marker or an alert. Every image is re-encoded on arrival, which discards its metadata — a phone photograph otherwise carries the exact coordinates it was taken at, and publishing those would hand out the sender's home address.
• We accept reports, but not sentences. A reporter chooses from a fixed list of things they can see — water in the street, road blocked, power out — and there is no field anywhere in which free text can be written. That is the direct consequence of the argument above: the harm in disaster misinformation lives in the words attached to an image, so the vocabulary is the safeguard, and a rumour cannot be expressed in it. Anyone can flag a report; two flags hide it pending review. The rules are in our content policy.
• Every record names its agency and links the original, so anything here can be traced to the body that published it.
• The cost is that we are slower and less vivid than a social feed, and always will be. That is the trade, and it is deliberate: an aggregator that added dramatic imagery to compete on speed would be adding exactly the failure mode this article describes.

The honest summary

The problem is rarely that someone invented a picture. It is that a real picture of somewhere else arrived first, and arriving first is most of what determines what people believe. You cannot verify a disaster image by looking harder at it. You can ask where it came from, whether an agency has said the same thing, and whether anything at all is lost by waiting — and the answer to the last question is almost always no.

This article is about evaluating information during disasters in general. This site is not a verification service and not a warning authority; for what is happening and what to do, use your national agency.

Frequently asked questions

Why do old disaster photos spread after a new disaster?

Because demand for images peaks hours before real ones exist — the people who could take them are busy surviving. Something fills that gap, and the easiest thing to reach for is a genuine photograph of an earlier event. The image is authentic, which is exactly why it is hard to doubt: flood water looks like flood water everywhere, and most people have no reference for what the affected place should look like.

How can I tell whether a disaster image is real?

Not by looking at it. Reverse image search catches most recycled photographs immediately because older copies stay indexed. Then check whether the same picture appears with a different caption, whether any agency has said what the caption claims, and whether the vegetation, clothing, signage or vehicle plates match the place named. Provenance is the check that works; appearance is not.

Can I spot an AI-generated image by looking for visual mistakes?

Increasingly not, and relying on that makes you more confident rather than more accurate. Advice about hands, text and lighting was useful for a while and is not a defence now. The reliable question is where the image came from and whether a responsible authority has said the same thing — not whether it looks convincing.

Are messages predicting the next earthquake ever real?

No. Earthquakes cannot be predicted, and no agency issues messages naming a magnitude and an hour. Such messages circulate after every major earthquake and are actively dangerous, because they move frightened people outdoors at night, sometimes into worse conditions than the building they left. What is real is that aftershocks follow a decaying statistical pattern — a description of many shocks, not a schedule for one.

What is the most harmful kind of disaster misinformation?

Not the images, but the text that travels with them: predicted aftershocks, rumours of imminent dam failure, and claims that water or supplies are contaminated or gone. Each of these causes people to move, hoard or avoid safe resources at exactly the moment when a wrong decision is most costly.

Why do corrections not undo the damage?

Because they arrive after the audience has moved on. The recycled image reaches far more people than the correction ever will, so the useful goal is not correcting faster but forwarding less. In a disaster, a forward carries your credibility with the people who know you — which is precisely why passing on something unverified does real harm.

Does this site publish disaster photographs?

None of our own. Everything we publish comes from a monitored agency feed with the agency named and the original linked, and we never source or caption an image ourselves. Since August 2026 a reader using our Android app may attach one photograph to a report about an event they are at. Those sit in their own section, are labelled unverified, are re-encoded so no location metadata survives, and never become a data point, a map marker or an alert. Anyone can flag one, and two flags hide it pending review.

Sources

• UN OCHA — Information management and rumour tracking in emergencies
• IFRC — Community engagement, accountability and rumour management
• WHO — Infodemic management
• USGS — Earthquake prediction and why it is not possible
• UNDRR — Risk communication and public understanding
• WMO — Official warning authority and single authoritative voice
• ReliefWeb (UN OCHA) — Verified humanitarian situation reporting

Found an error? See our corrections policy.