In 2018, the BBC published an investigation that found it was easy to find deceptive advertising on Google, and Google promised to fix it with "machine learning." Whether this was just bunkum or attempted in good faith, the result two years later is that it's still easy to find deceptive advertising on Google.
Travel permit adverts are not the only ones slipping through Google's systems. Earlier in April, news website This Is Money set up a fake online investment company and was able to advertise it on Google. In May 2020, the BBC found a suspected scam store had topped Google's search and shopping results for weeks, encouraging customers to pay for tech via direct bank transfer. Last week, the BBC found another fake gadget store at the top of Google's search results when customers were hunting for gadgets such as iPhone 12, Samsung S21 and Sony headphones.
The striking thing is that the BBC investigation tracked a specific type of TOS-violating ad targeting ESTA, the United States' electronic system for travel authorization. This choice of subject was rooted in the initial scandal—advertisers charging exorbitant fees for free or cheap government services—but remains useful because it shows Google is not simply overwhelmed by vague fakery in the ocean of online commerce. It's matching specific searches to specific advertisements. Why can't Google flag such specific, unambiguous search terms?
The implication is that it can. A straightforward word filter can spot ESTA-related ads and searches. What it can't do, though, is tell if the ad served is a swindle, TOS-violating, or illegal. And it turns out that machine learning can't tell either, after a few years of trying. This means humans would have to check the flagged ads. But this will not happen, because it's expensive.
Google doesn't make significant money from visa fee swindle ads, but it makes a lot of money by avoiding human moderation.