Written by Tom Phillips

When Hurricane Sandy hit New York and New Jersey, I was running a blog called “Is Twitter Wrong?” an experiment at fact-checking viral images.

When a major natural disaster hits an area densely populated with heavy social media users - and media companies - one result is a huge number of images to sift through. Telling the good from the bad suddenly shot up the editorial agenda.

One particularly viral pair of images showed a shark supposedly swimming up a flooded New Jersey street. I teamed up with Alexis Madrigal from The Atlantic to try to verify these images.

One aspect of the images, shown below, is that they were strange enough to make you suspicious, yet they weren’t implausible enough to dismiss out of hand. In the end, and they proved very hard to definitively debunk.

Pre-existing images that have been misattributed (perhaps the most common form of “fake”) can often be debunked in a few seconds through a reverse image search. And pictures of major events can often be at least partly verified by finding mutually confirmatory images from multiple sources.

But neither of those work for a one-off chance sighting that’s either an original picture or an original hoax. (My experience is that verification of images that can’t be debunked/verified within a few minutes tends to take a lot longer.)

In the end, sometimes there’s no substitute for the time-consuming brute force approach of image verification: tracing an image’s spread back through social media to uncover the original; walking the streets of Google Street View to pinpoint a rough location; and/or scrolling through pages of Google Image results for a particular keyword, looking for possible source images.

In this case, the Google Image search approach paid off - we were able to find the exact image of a shark’s fin that had been Photoshopped into one of the pictures.

But even then, we were unable to say that the other image was definitively fake. It used a different shark.

Our attempts to find the origin of both shark images kept hitting the barrier of people saying, vaguely, that it was “from Facebook.” We eventually found the originating Facebook poster via a tweet directing us to a news site that credited the source. (Both the news report and Facebook posts have since vanished from the Web.) But even that didn’t entirely help, as the page owner’s other photos showed genuine flooding in the same Brigantine, New Jersey, location. He also insisted in replies to friends that the shark pictures were real. (In retrospect, he seemed to be intent mostly on pranking his social circle, rather than hoaxing the entire Internet.)

The fact that he was claiming one undoubted fake as real was enough for us to move the other shark image into the “almost certainly fake” category. But we still didn’t know for sure. It wasn’t until the next day, when the fact-checking site Snopes managed to identify the source image, that we were able to make that call with 100 percent certainty. This was the shark image that was used to create the fake:

That may be the main lesson from Sandy: Especially in rapidly developing situations, verifica- tion is often less about absolute certainty, and more about judging the level of acceptable plausibility. Be open about your uncertainties, show your work, and make it clear to the reader your estimate of error when you make a call on an image.