The Persistence of Deceptive Ads on Google

By Chris Greening

---

The Incident

A few months back, while browsing the YouTube app, I encountered a particularly "dodgy" advertisement. I'll admit, I fell for it—a momentary lapse in focus combined with the genuine anxiety of my iPhone running low on storage led me to click it.

Knowing that these types of deceptive ads are harmful, I did my part as a digital citizen and reported the content.

The Frustrating Feedback Loop

Unfortunately, the response from Google was underwhelming. I reported the ad again, assuming the first instance was just a fluke. The result? The same canned response:

"We found that the ad doesn’t go against Google’s policies, which prohibit certain content and practices that we believe to be harmful to users and the overall online ecosystem."

It turns out I wasn't alone; several other users reported the same ad and received the exact same dismissal.

Why is this happening?

There are two primary ways to interpret this failure:

1. The Charitable View: Human reviewers are overwhelmed and simply miss things. $\text{Error Rate} > 0$.
2. The Cynical View: Google benefits financially from these ads. If an ad generates a high click-through rate (CTR), it generates more revenue.

$\text{Profit} = \text{Clicks} \times \text{Cost Per Click (CPC)}$

Maybe it's just a mistake... or maybe the $\text{money} \approx \text{motivation}$.

The AI Experiment

If human "meatbags" are failing, why not use the very technology Google is currently championing? I fed the ad to Gemini, Google's own AI model, to see if it could spot the deception.

The Result: STATUS: DISAPPROVED

The AI didn't just reject the ad; it provided a detailed forensic breakdown of the violations:

{
  "classification": "DISAPPROVED",
  "violations": [
    {
      "category": "Misrepresentation",
      "type": "Misleading Ad Design",
      "detail": "Mimicking System Alerts"
    },
    {
      "category": "Misrepresentation",
      "type": "Unreliable/Deceptive Claims",
      "detail": "Fear-based tactics"
    }
  ]
}

Detailed Findings from Gemini:

• System Mimicry: The ad uses a banner that looks exactly like an iOS system modal ("iPhone Storage is Full"), utilizing native typography and styling to trick the user.
• Fake Interactivity: The "Yes" and "No" buttons are static images. They aren't actual controls; they are just click-bait to trigger a redirect.
• Psychological Manipulation: By claiming "some features may not work properly," the ad creates a false sense of urgency and panic to force an app installation.

Comparison: Human vs. AI Review

The Workflow of a Report

The following diagram illustrates the current (and broken) cycle of reporting deceptive ads:

Final Thoughts

It is staggering that Google's own AI can identify these violations in seconds, yet the actual review process approved the ad twice.

Google's immediate To-Do list:

• Integrate Gemini into the ad approval pipeline.
• Stop relying solely on overwhelmed human reviewers.
• Prioritize user safety over high-CTR deceptive revenue.

---

Support the Author: If you find this work valuable, consider supporting it via:

• Patreon (for the flush)
• Ko-fi (for a one-off donation)

Stay updated via the newsletter for more on ESP32, AI, and eclectic tech projects!