Fake Products, Real Problems: Fighting Counterfeit Products at Scale
A snapshot of my presentation at the Financial Crime 360 Conference
Every time you search for a product on Google, YouTube, or even use Google Lens, there’s an invisible army working to make sure what you see is genuine. That’s our Commerce Safety team at Google Shopping, which reviews over 50 billion product offers across 30+ languages and 90+ countries.
Last week, I shared our story at the Financial Crime 360 conference. Today, I want to pull back the curtain on one of e-commerce’s most challenging problems: counterfeit detection at scale.
For most of us, commerce is the medium through which we engage with the broader world. As adults, we rely on a global supply chain we may never fully see, which includes everything from coffee beans grown in Brazil, semiconductors manufactured in Taiwan, to smartphones assembled in Vietnam.
When we hand our money to strangers (especially online strangers), we’re exercising profound trust. That trust is fragile, and counterfeits shatter it.
The Questions We Ask About Every Product
At Commerce Safety, we look at each product offer and ask four fundamental questions:
Is it safe? Will consumers or others be harmed by using this product?
Is it appropriate? Is it suitable for children, those with explicit content preferences, and people with different cultural values across regions?
Is it misleading? Does the description hide, manipulate, or obscure relevant information?
Is it supported? Is Google Shopping the right place to list this offer?
For most policies—prohibiting weapons, drugs, or explicit content—enforcement is relatively straightforward. But counterfeit detection? That’s where things get interesting.
The Context Problem
Here’s the definition we work with: Counterfeit products mimic trademarks, logos, or brand features of other goods to pass themselves off as genuine products.
Simple enough, right? Except context changes everything:
Is that price a great deal, or is it too good to be true?
Is the product used or new?
Are only authorized merchants allowed to sell this brand, or can any merchant resell it?
The Brake Pad Challenge
Let me show you why this matters. Forget luxury handbags for a moment—those are actually easier to detect. Instead, consider Bosch brake pads.
I showed five product listings at the conference and asked the audience whether they could tell which ones were counterfeit. The price signals that work for a $3,000 handbag don’t help when legitimate brake pads range from £17 to £26. But the stakes? They couldn’t be higher. Faulty brake pads don’t just disappoint customers, they put lives at risk. The same goes for counterfeit iPhone chargers that can cause electric shocks or house fires. These aren’t victimless crimes. They’re public safety threats hiding in plain sight.
Scale Meets Nuance
Let’s put this in perspective: 50+ billion total listings means that even 1% of our catalog represents 500 million items. Traditional content moderation approaches simply don’t scale to this level while maintaining the nuanced judgment counterfeit detection requires.
This is where it gets fascinating.
Our Four-Part Solution
We’ve developed a strategic approach that combines adaptive AI with human expertise:
1. Enhance Detection with AI
We use Google’s Gemini models to detect mimicking behaviors—subtle patterns that suggest a product is copying trademarks, logos, or brand features. The AI teases out patterns and highlights anomalies across billions of products that human reviewers alone could never catch.
2. Broaden Product Coverage
We work directly with brands to leverage their authoritative sources of trademarks, logos, and brand features. They know their products better than anyone, and their intelligence feeds our systems.
3. Integrate Expert Intelligence
Human expertise remains irreplaceable (As I wrote here, the data from 10 billion content moderation decisions highlights that human judgment is still needed to make the final call). Our specialists provide the contextual judgment that trains and refines our AI models, especially for edge cases where context determines everything.
4. Feed Insights Back into Models
Every decision, every anomaly detected, every false positive corrected—it all flows back into our models. This creates a learning loop that makes our detection smarter over time.
The Bigger Picture
We collaborate across legal, policy, product and trust & safety teams and with external partners to share intelligence and refine our defenses against fraud and abuse.
Advanced AI has become critical to this mission, significantly enhancing our ability to proactively prevent and combat bad actors across the industry. But make no mistake: this is an ongoing arms race. As our defenses improve, so do the tactics of those selling counterfeits.
We’re dedicated to future-proofing these defenses through continuous investment in new solutions. The threat landscape is dynamic, but our commitment to providing a safe and reliable shopping experience is unwavering.
What’s at Stake
The next time you search for products online, remember there’s an invisible battle happening behind the scenes. It’s a battle fought with cutting-edge AI, human expertise, and partnerships with brands—all to ensure that when you trust an online merchant with your money, you get what you paid for.
Have questions about counterfeit detection, content moderation at scale, or how AI is shaping e-commerce safety? Drop a comment below or reach out. I’d love to hear your thoughts.