Media measurement company DoubleVerify has released new data on the threat posed by AI-generated app reviews. According to the latest DV report, “DV’s Fraud Lab has seen a significant increase in apps with AI-powered fake reviews in 2024, identifying over three times the number compared to the same period in 2023.”
The trend of Gen AI being used to scale positive reviews has been documented elsewhere. In April of last year, a CNBC report surfaced that Amazon reviews were being sloppily written using ChatGPT, even including telltale “as an AI language model” language. Amazon rolled out an AI tool to spot AI reviews a few days later. Similarly, platforms like TripAdvisor have also been challenged by fake AI reviews.
Why This Matters:
DV’s report highlights two problems with AI reviews. First, the consumer impact. According to DV, AI-faked reviews make an app seem more credible when, in reality, it might be malicious. “Some fraudulent apps can hijack devices, running ads incessantly, even when devices appear to be off,” reads the report. “This can lead to severe issues like battery drain, device overheating and excessive data usage, incurring significant costs for the user.”
Fake AI app reviews can also wreak havoc on the advertiser side. Lots of positive reviews get consumers to download and advertisers to want to target their apps. In reality, the reviews are the first step of fraud, with ad fraud being the second step once the app is downloaded. DV found that some of the apps with AI reviews were tied to CycloneBot, a fraud scheme it had previously identified.
DV did not explicitly name an app engaging in AI reviews in the report but did link to one – an app called Wotcho TV available on Amazon. It is a streaming app, and DV goes on to say that 50% of its reviews were fake. Copyleaks, a platform that can analyze and identify AI-generated content, confirms this, sharing that 6 of the reviews for Wotcho that they analyzed were AI-generated (they also note that the Wotcho website is AI-generated).
Streaming fraud is particularly problematic given the price, “where the cost per thousand impressions often ranges from $35 to $65,” according to DV. In streaming, full transparency is already challenging on where ads run and the platforms they run on, so this creates even more complexity and concern for CTV advertisers that are running in an open environment.
Experts React:
A few months ago, DV CEO Mark Zagorski, on CNBC, said this, fortuitously:
“There are some great things about AI… it’s helping create more creative ads… it’s helping buy media more efficiently… it’s also creating challenges. AI helps perpetrate fraud. It’s both a great tool… but it’s also being used to spread some pretty nasty stuff.”
Our Take:
This is the tip of the iceberg. DV identified both obvious examples of AI-generated reviews in their study and more discrete ones that require deeper analysis to discern linguistic and copy patterns, syntax patterns, and posting patterns by the accounts publishing the reviews. All of that is incredibly challenging to audit given the scale at which AI is growing both in terms of its technological evolution and adoption by “bad guys.” These technologies will only get better and make it more difficult to identify their nefarious use and the harm it causes. App stores, along with vendors like DV, will have to work closely to ensure consumers and advertisers aren’t hit hard.
