When a disaster is unfolding, your feed fills with images faster than anyone can check them β and that gap is exactly what bad actors exploit. It's tempting to assume the fakes are all AI now. They aren't. The oldest and still most effective trick is far cheaper: a genuine photograph from a different flood, a different country, or a different year, re-posted with a caption tying it to today. The AI-generated ones are a fast-growing second category layered on top.
That order matters, because it tells you what to check first. After recent hurricanes, fact-checkers and outlets like Yale Climate Connections documented both patterns spreading side by side β recycled real photos and slick AI fabrications β often while people were still deciding where to send help. The workflow below is built to catch the common case before you spend any effort on the exotic one.
The verification workflow
Reverse image search it before you do anything else
This one step catches most fakes, because the most common disaster hoax isn't AI at all β it's a real photograph from another flood, another country, or another decade, stapled to today's headline. Drop the image (or a screenshot of it) into Google Lens and TinEye. On TinEye, sort by 'oldest first'; if the same shark-on-a-highway or submerged-airport photo was already circulating in 2011, you're done. The Google News Initiative teaches this as the first move for exactly this reason: search tells you where a picture has been before it ever reached you.
Check it against official sources and the wire services
During any major hurricane, flood, or wildfire, the response agencies run live rumor pages β FEMA's Hurricane Rumor Response is the US example, and most national civil-protection agencies publish something similar. If a dramatic image shows damage or a rescue that no official briefing, local newsroom, or wire service (AP, Reuters, AFP) is reporting, treat it as unconfirmed, not as a scoop. Real catastrophes are covered by many outlets within hours; a world-ending photo that exists only on one anonymous account is a warning sign in itself.
Read the frame for the AI tells
If reverse search comes up empty, look closely at the pixels. AI disaster images tend to over-dramatize: cinematic, golden-hour lighting on a scene that should be chaotic, waxy over-smooth surfaces, and backgrounds that dissolve into smeared texture. Check hands, limbs and faces on any people β especially the inserted 'victim' β for extra fingers or features that don't resolve. Zoom into any signage: street signs and storefronts usually render as confident gibberish. Our guide on whether an image is AI-generated walks through these tells in detail.
Look for provenance signals β but don't trust their absence
Some generators mark their output. Google's tools embed an invisible SynthID watermark, and images from Adobe, OpenAI and others may carry signed C2PA Content Credentials you can read at verify.contentauthenticity.org. A positive hit tells you the image was AI-made or edited. The reverse is not true: a screenshot, a re-save, or a crop strips those marks, so 'no watermark found' never means 'this is a real photo.' Use provenance as a tiebreaker, not a verdict.
Verify before you share β and especially before you donate
Fake disaster images aren't just noise; they're bait for scam fundraisers and ad-farm traffic. Before you reshare or give money, run the claim through a fact-checker β FAXTR searches 100+ IFCN-signatory organizations at once, and viral disaster images are often debunked within hours. If an image is being used to solicit donations, check the fundraiser itself the same way you'd vet any other.
What AI disaster images tend to get wrong
If reverse search doesn't surface an older copy, the image may be synthetic. These are the details generators still fumble β none is proof on its own, but two or three together are a strong signal.
Hands, limbs and faces
Extra fingers, fused limbs, or melted features β most visible on an emotionally charged figure like a stranded child or a rescuer. Generators still struggle most with the thing your eye is drawn to.
Signage and text
Street signs, shopfronts, number plates and news chyrons come out as plausible-looking gibberish. Zoom in on any writing in the frame; it's one of the quickest tells.
Physics that doesn't add up
Floodwater meeting a wall at the wrong angle, flames with no matching smoke, debris that floats where it should sink. Disasters are messy; AI tends to render a tidy, impossible version of the mess.
Too perfect a story
A single image that delivers the whole emotional arc β the soaked dog on a rooftop, the flag in the rubble, the lone firefighter at sunset β is worth a second look. Real disaster photography rarely composes itself that neatly.
Tools worth having
Google Lens / Google Images
Broadest match database and the fastest way to find where a photo first appeared. Use the camera icon at images.google.com, or long-press an image on your phone.
TinEye
Reverse search with an 'oldest first' sort β the single quickest way to expose a recycled photo that predates the disaster it's being tied to.
verify.contentauthenticity.org
Reads C2PA Content Credentials embedded by some cameras and AI tools. Needs the original file; a screenshot won't carry the data.
FEMA / national rumor-response pages
Official ground truth during a major event. If a claim isn't reflected here or in wire-service coverage, it isn't confirmed.
Google Fact Check Explorer + FAXTR
Searches published debunks. Recycled and AI disaster images are frequent enough that a verdict often already exists.
A caution on false matches
Every step here produces a lead, not a ruling. Two different floods can look nearly identical in a compressed, low-resolution image, so a reverse-search hit is a hypothesis you still have to confirm with a second signal β a matching landmark, a corroborating report, a fact-check that reached the same origin. The same goes for AI detectors: a confident "98% AI" score on a photo you've already traced to a real 2018 news wire is the detector being wrong, not the archive. When signals conflict, provenance and the original source win.
If it turns out to be fake
Don't re-upload the image to debunk it β that just hands it more reach. Reply with a link to the original context or the fact-check, report the post through the platform's misinformation flow, and if money is involved, verify the fundraiser before anyone gives. See our guides on checking whether a fundraiser is legitimate, reverse image search, and telling if an image is AI-generated for the deeper version of each step.
Check a viral claim in seconds
FAXTR searches 100+ fact-checking organizations in one query β free, no login required.