About Trustpilot

Trustpilot is the world's largest open customer platform where people's experiences are made visible through reviews, with a mission that goes well beyond aggregating star ratings. The company exists to be a universal symbol of trust, where customer experiences speak louder than marketing claims.

That mission depends on the integrity of its reviews. As fraud tactics have grown more complex, protecting the platform has become a sophisticated fraud challenge.

The problem: Professional fraud rings targeting the platform

In its early days, Trustpilot's fraud challenges centered around small-scale abuse: They faced mostly isolated incidents of businesses gaming their own ratings, and one-off attempts at manipulation.

That landscape has changed dramatically. Today, the reviews industry faces a professional ecosystem of bad actors, including "review sellers" specifically built to exploit review platforms.

"These are large-scale operations," explains Sona Pakhchanyan, Senior Applied AI Manager at Trustpilot. "We are seeing organized businesses with revenue models, technical teams, and strong financial incentives attempting to defeat detection systems. They have high incentives to bypass defenses, and they invest heavily in the technology to do so."

The rise of modern AI has made this conundrum even more complex to navigate. Generating large volumes of reviews that sound relatively natural and don't immediately look machine-made, is relatively simple.

As a result, some of the signals used to help identify fake reviews, like repetitive language, low variation, and obvious patterns, were becoming less reliable. Trustpilot's continuously evolving detection approach had to outpace these threats.

The solution: Device intelligence as a cornerstone of fraud detection

Trustpilot needed stronger signals at the device level to cut through the noise and reliably link activity that might otherwise look unrelated. Trustpilot sought to partner with a global specialist to achieve the most accurate, persistent device identification possible.

That's where Fingerprint came in. As an added layer of intelligence for its existing in-house fraud detection systems, Fingerprint provides a stable device identifier that allows teams to recognize the device across sessions, even if things like IP address, cookies, or browser settings change. That makes it much harder for bad actors to operate at scale from a single underlying device.

That signal feeds into Trustpilot's broader fraud detection system to strengthen real-time rules, improve machine learning models, and help uncover networks of coordinated activity.

And importantly, it's flexible. Trustpilot can adjust how aggressively they want to connect activity at the device level — zooming in on small clusters or expanding outward to identify larger fraud networks.

"If we set a high confidence threshold, we might tie five reviews together. If we lower the threshold, we might connect 50," Pakhchanyan says. "That control is extremely valuable."

The impact: Millions of fake reviews blocked at 99%+ accuracy

With Fingerprint as a core part of their fraud stack, the team is able to detect and remove fake reviews at a massive scale, while maintaining a high level of accuracy:

• 99%+ accuracy in moderation decisions
• 4.5 million fake reviews removed each year (about 7% of all incoming reviews)

Just as important as the results is the consistency behind them. As fraud tactics have evolved from manual abuse to AI-driven attacks, Fingerprint has remained a reliable foundation the team can count on.

Why it matters: When trust is your product, your fraud stack matters

As AI makes fraud easier to scale and harder to detect, maintaining that trust only gets more challenging. That's why having the right foundation matters.

"Device fingerprinting is a crucial element of our multi-layered fraud detection systems — which we continuously evolve to outpace new threats from bad actors," Pakhchanyan says.

Want to see how device intelligence can protect your platform's integrity? Talk to our team.