Study: questionable Google ads had over 100 million views in Europe

A team of researchers from New York University and the Netherlands' Radboud University has developed a tool called AdLens, which uses artificial intelligence to detect ads that mislead users. As reported by the Polish site Spider's Web, the researchers checked nearly 189,000 versions of ads for software, mobile apps and electronics from Google's public ad transparency centre.

Three groups of questionable ads

The researchers identified three categories of problematic material:

• 238 ads that provoked fear of a supposed threat,
• 3,346 ads making false or unrealistic promises,
• 258 ads hiding the advertiser's identity or the nature of the service.

The fear-based ads had more than 16 million views, and those with unrealistic promises had 89.9 million. In total this exceeded 100 million views across European Economic Area markets. The authors stress that this does not mean 100 million people were deceived: a view is not a click, an installation or a financial loss.

Examples: viruses and phone tracking

A common pattern was scareware, an ad claiming that a phone has been hacked or a computer infected. The user is offered an app to download that supposedly fixes the problem. The ads included a Polish-language one claiming that an Android phone had picked up 178 viruses after visiting adult websites. It had about 1.1 million views.

Another ad promised to check the location of any person from their phone number. According to the researchers, it was active for over 2 years and had 14.5 million views. Another urged users to recover 284 photos deleted 7 years earlier, and was seen 2.8 million times.

Google's response to reports

The researchers reported some of the ads to Google. According to the description of the study, most of the reported ads were not considered by the platform to breach its rules. In the case of a domain linked to the TamperedChef campaign, which spread malware disguised as a PDF editor, 42 ads leading to the same site were found. One was reported. Google removed it, but the other 41 were not withdrawn as part of that action.

Of 147 apps promoted by the questionable ads, 34 were no longer available in app stores, although the ads were still running.

What this means for you

The study covers EEA markets, so it does not directly include the UK. However, the schemes described are widespread online. People living in the UK should treat with caution ads that threaten viruses or account lockouts and demand an immediate app download. The same goes for ads promising to track someone's phone. Apps are best installed only from official stores, and security warnings should be checked with the device maker or your bank, not an ad.

Limits of the tool

AdLens first compares the content of ads with known scam patterns, and then selected ads are checked by more advanced language models that analyse text and graphics. In tests the system achieved an F1 score of 0.922. The authors admit the tool has limitations and cannot fully replace human review. They have made it publicly available so that independent organisations and researchers can use it.