High school students pose with a cardboard Instagram frame encouraging youths to vote.Toru Hanai / Reuters

Here is a question that is probably best posed to a teenager: How many likes does an Instagram photo need to get before it’s safe from deletion?

There are plenty of adults who use the social photo-sharing platform, but deleting photos that aren’t well-liked enough is a distinct behavior among teens who use the site, according to researchers at Penn State University.

“Teens want to be very popular so they’re very conscious of the likes they’re getting,” says Dongwon Lee, an associate professor in the school’s College of Information Sciences and Technology. He and his colleagues have published several papers about how social media behaviors vary by age, and his research focuses more broadly on like-mediated interactions in social environments.

On Instagram, for instance, teens interact more with photos than adults do—they comment more frequently, and like more photos—but, unexpectedly, they seem to publish less frequently themselves. (There may be a simple explanation, though: It’s possible teens only appear to publish less frequently than adults because teens are more likely to delete photos they perceive as under-liked, and to return to their own feeds and prune them over time.)

Adults also post more diverse topics, which may be a reflection of disposable income that allows for vacation or travel, Lee suggests. Teens mostly post photos that reflect their “mood or personal well being,” he says. Instagram’s core users, overall, skew relatively young. Some 90 percent of them are under age 35, according to Business Insider. I should note that the grand total of two teens I asked, via colleagues who are their siblings, about the teen rules of Instagram had little to say on the question of deletions. Maybe because, as one 14-year-old explained, Instagram is basically over and everyone is Snapchatting instead.

Part of why the researchers at Penn State have focused on Instagram is because its API—the software interface that allows third parties to gather information about activity on the site—makes data relatively easy for them to work with. Age differences notwithstanding, the taxonomy of a like on Instagram is less complex than comparable reaction buttons on social platforms like Facebook or Twitter.

Facebook recently introduced a series of new emoji-like buttons—in addition to the classic blue thumbs-up “like”—to allow for more precise reactions to status updates. Now, it’s possible to react with the classic like, a heart for love, or faces meant to represent haha, wow, sad, or angry. And while that change has allowed for a degree more nuance, it doesn’t alleviate confusion entirely. Recently, for instance, I saw a friend clarify that her “heart” reaction to a friend’s post about police violence wasn’t meant to convey that she loved police violence, but that she was sending love to her friend. On Twitter, the meaning of a “fave” or “like”—it was once a star-shaped icon, now it’s a heart—is arguably even more layered. That may be because of there’s just one button for reaction a complex array of social interactions.

A basic taxonomy of Twitter faves would have to include the following uses, at least:

• When you genuinely like the tweet
• When you genuinely don’t like the tweet; also known as a “hate fave”
• When you want to mark a tweet to revisit it later
• When you want to acknowledge that you’ve seen a tweet, but without text; also known as the farewell fave

My colleague Megan Garber has delved into these shades of like, and the farewell fave in particular, in more detail. “With a single click, I can do something that is actually extremely difficult to do in the digital context: end a conversation,” she wrote in 2014. Many have pointed out how useful a like button would be in the realm of email. (Microsoft actually introduced such a feature.) On any social platform, Lee told me, “the scope of the taxonomy is quite blurry. People use the button for different purposes, even the same person will use the like for different meanings. There’s no clear meaning.”

“Even if we develop an algorithm that classifies a like,” he added, “no one can tell if our algorithm’s result is correct or not.”

Lee and his colleagues are now working on building a classification scheme that might help them better understand how Facebook’s new reaction buttons are being used. But, he concedes, data-based findings on usage only represent one semantic layer. It’s easy to track how a person reacts, particularly when the reaction is logged with the press of a button, but it’s not always clear what it really means. People are still getting accustomed to this ambiguity in the context of communicating on social platforms—especially when conversations are marked (or ended by) emoji and other imagery. But the slipperiness of the meaning of a “like” is no different than the subtext that comes with any other form of human language. The way we speak to one another evolves, we try to keep up, and sometimes we actually understand each other.

We want to hear what you think about this article. Submit a letter to the editor or write to letters@theatlantic.com.