Other Twitter users weighed in, noting that Sierra Burgess is a Loser and The Good Cop, both shows with majority-white casts, had black people featured prominently in the homepage artwork shown to them by Netflix.

For comparison, Kelly Quantrill (@codetrill) shared that she, a white person, saw mostly white cast members in the ads for those same productions.

It’s true that everyone’s Netflix homepage looks a little bit different, thanks to the company’s algorithms. But Netflix denies that it considers demographics when choosing the artwork shown to users.

“Reports that we look at demographics when personalizing artwork are untrue,” a Netflix spokesperson told Quartz. “We don’t ask members for their race, gender, or ethnicity so we cannot use this information to personalize their individual Netflix experience. The only information we use is a member’s viewing history.” The spokesperson confirmed that the thumbnails served up with the titles “do differ and regularly change. This is to ensure that the images we show people are useful in deciding which shows to watch.”

Netflix explained in a blog post last year that it uses A/B testing to personalize recommendations for its 137 million subscribers. Normally, this kind of optimization is a useful way for Netflix users to find content they’d find interesting based on their watching habits and preferences.

Last year Netflix rolled out an algorithm called “Artwork Personalization.” With this, multiple images are created for each title on the platform, and are rotated based on the viewer and the customer’s viewing history. The company used the film Good Will Hunting as an example, explaining that for romcom fans, the artwork would show Matt Damon canoodling with Minnie Driver, while comedy fans would see Robin Williams.

Brown and others who weighed in online noted that the kinds of personalization they were seeing seemed more misleading than helpful.

📬 Sign up for the Daily Brief

Our free, fast, and fun briefing on the global economy, delivered every weekday morning.