Case study

Hinge

Moderating toxic ad comments decreases customer acquisition costs by 19%

Protected paid performance and improved conversion by cleaning ad comments at scale. Arwen reduced toxic comments on Hinge's ads by 98%, leading to a 19% decrease in customer acquisition costs on Instagram.

Products used:

Moderate

Hinge

98%

of toxic comments accurately moderated

19%

reduction in customer acquisition cost on Instagram

7%

reduction in customer acquisition cost on Facebook

Overview

Better paid ad KPIs by auto-moderating toxic comments

Hinge's zero-tolerance for toxicity is driving strong results - powered by Arwen.

Social media advertising has become complex, with various factors determining the success of a campaign. Marketers frequently look at multiple variables to find the most effective strategies to maximize impact and return on investment.

This comprehensive analysis delves into the complex relationship between toxic comment moderation on social media ads and key performance indicators. The research evaluated two campaigns to measure the effect of toxic comment moderation on customer acquisition costs.

The study's findings highlight a significant correlation between active comment moderation and improved social media campaign performance across various metrics. The study concluded that social media users are more likely to click on an ad if the ad doesn't have any toxic comments; this was proved by a higher click-through rate and lower customer acquisition costs on Instagram.

Challenge

Hinge, a global dating platform with 229k followers, runs 119 ads monthly and gets more than 6000 comments. They came to us with a big issue: nearly 60% of the comments on their ads were toxic.

They were concerned about the detrimental effect this had on their audience, the creators they featured, and their ad spending. Patterns of toxicity ranged from racism and homophobia to more subtle, targeted insults. They wanted to ensure we could catch all types.

An infographic showing the impact of toxic comments on Hinge's social media ads

Solution

  • Before implementing Arwen, 60% of comments were identified as toxic, establishing a baseline for comparison.
  • Arwen's sophisticated AI, which combines both closed and open models, was deployed to actively moderate toxic, homophobic, racist, and spam content comments in real-time on Hinge's ads.
  • After one month, we revisited Hinge's Facebook Ads Manager to collect updated data on comment volume and toxicity levels following Arwen's moderation.
  • We then compared Instagram and Facebook CAC before and after implementing Arwen.
A visual comparison of Hinge's social media ads, highlighting improvements after using Arwen AI moderation

Results

Using Arwen, Hinge significantly improved its advertising effectiveness and brand perception. By proactively filtering toxic and harmful comments, Hinge created a safer and more positive environment for engagement on their ads. This reduced the incidence of negative interactions and fostered a more trustworthy and supportive community around the brand.

After using Arwen, the ads performed much better, leading to:

  • 19% decrease in customer acquisition cost on Instagram
  • 7% decrease in customer acquisition cost on Facebook
  • 98% of toxic comments moderated accurately

The results of using Arwen were twofold: one measurable and the other deeply felt by the creators. All part of Arwen's mission to make social media more social.

Hinge ad comment section after Arwen moderation
“

The whole team is just so excited to have this protection in place. It makes us feel like we've raised the bar, made the creators feel safe, and moved from a reactive approach to a proactive one. Arwen solved a massive problem around our paid ads which we've had for years; I can't believe we've only just started working with you.

Marketing Manager North America, Hinge

Elke Luria

Ready to turn every comment into measurable value?

See how Arwen helps teams protect performance, spot opportunity, and act in real time, without adding complexity.