A new study out today says a surprising proportion of programmatic mobile ad impressions — 34 percent to be exact — is at risk of being fraudulent.

Programmatic mobile ads are those served from an exchange to a network of publisher sites based on various kinds of targeting data, usually in real time.

The study, conducted by mobile ad tech company AppLift and security company Forensiq, found that 22 percent of the total number of impressions studied were “suspect” for being fraudulent, while 12 percent of the total were at “high risk” of being fraudulent.

AppLift and Forensiq studied bid requests on ad impressions running through the AppLift’s DataLift ad buy platform from September 21 through October 19 to come up with the results.

“Through our technology, we listen to the impressions and analyze the bid requests from the exchanges,” wrote AppLift’s Thomas Sommer in an email to VentureBeat. “The bid requests comes with different parameters from users seeing the ad, which are dependent on things like: device, time of day, phone model, etc.”

The ad industry’s main trade group, the Interactive Advertising Bureau, recently released a study saying that ad fraud, in its various forms, is costing the industry $8.2 billion a year in the U.S. alone. About $1.3 billion of that comes from mobile ad fraud, it said.

eMarketer predicts that mobile programmatic media buying will reach $20.45 billion by 2017. With that much money being spent, the idea that so many impressions could be bogus is of grave concern.

The study found that traffic from campaigns run on a CPC (cost per click) basis is three times less likely to be fraudulent than for CPM (cost per thousand) campaigns. And traffic from campaigns run on a CPI basis is 10 times less likely to be fraudulent than for CPM campaigns.

The researchers also believe that fraud risk goes way up on impressions that happen at night.

Hat tip: AdWeek