Who Really Writes Twitter’s ‘Trending’ Summaries

‘Twitter description guy’ isn’t a guy. It’s Twitter’s curation team, and I talked to the woman who runs it.

Twitter’s director of curation, Joanna Geary. Photo courtesy of Joanna Geary

“Twitter description guy,” in users’ collective imagination, is a beleaguered soul, constantly scrambling to comprehend the bizarre subcultural memes that go viral on the site so that he can write sober-minded summaries of them for Twitter’s trending section. In December, Twitter’s description of a Minecraft-related trending topic led Twitch streamers and gamers to imagine a beleaguered “Twitter description guy”. They worked to make #TwitterGuyIsOverParty a trending hashtag in hopes that said Twitter guy would be forced to write a description of his own cancellation.

There is, of course, no single “Twitter description guy.” The descriptions are written by Twitter’s curation team, which is run by Joanna Geary, Twitter’s senior director for curation. To better understand how the curation team views its job, I spoke with her in February.

“Our goal is to give you the gist so you’re not spending five minutes looking at the trend trying to figure out what it even means,” she told me.

The trending module, which highlights words or phrases that are suddenly appearing in far more tweets than usual, has long been one of Twitter’s defining features. It has gone through several names and iterations over the years, and today it’s most visible in the Twitter app’s “Explore” tab or in a sidebar called “What’s happening” on the Twitter website.

By any name, trending topics are an open window into the messy, beating heart of a service that’s built on public conversations about current issues. Sometimes the picture they show is funny or uplifting; sometimes it’s ugly; and sometimes it’s just baffling. Other times they’re a vector for misinformation, spreading false or misleading memes to a much wider audience than would have otherwise encountered them. Paradoxically, attempts by Twitter users to debunk a false meme can result in it trending more widely because the algorithm takes the debunkings as a further signal of the meme’s popularity as a discussion topic. I wrote last year about how this dynamic led the phrase “Covid-19 is a lie” to trend in California.

“We’re attempting to remove the ‘WTF?’ from the trending tab.”

For years, the trending topics list was almost entirely automated save for Twitter’s occasional manual removal of a particularly toxic, hateful, or obscene topic. The company has developed ways to adjust it by geography and personalize it based on users’ interests. Last fall, Twitter began adding context to some trending topics, either by pinning a tweet that helped to explain them or by adding a title and description of the topic written by Twitter employees. In October 2020, ahead of the U.S. presidential election, Twitter went a step further, pledging to only surface trends in the “For you” and “What’s happening” sections after it had added context or a description. That meant showing the same trends to everyone rather than personalizing them. (Automated trends remained available under the “Trending” section of the “Explore” tab and elsewhere.)

This seemed to make a significant difference in the quality of trending topics, while it lasted. The screenshots below show Twitter’s desktop “What’s happening” module before (right) and after (left) the new policy. In the screenshot at right, from June 2020, three of the five top trends were driven by misinformation. In the one at left, from November 2020, Twitter’s policy of contextualizing trends ensured that all of them pointed toward real news from reliable sources.

After the U.S. election, however, Twitter stopped limiting the “For you” section to curated trends, potentially allowing misinformation to appear there once again. A Twitter spokesperson told me that the company is actually still curating just as many trends as before; the main difference is that it has reopened the tab to once again include automated trends personalized to each user. Those personalized trends may include some that are less high-profile on the site overall and are thus not among those selected for added context.

Twitter’s curation team was originally formed in 2015 as its Moments team, under then-product manager Madhu Muthukumar. (He’s now head of product at Robinhood.) That team is still responsible for curating Twitter Moments — strings of tweets that together tell a story. But the Moments product didn’t take off in the way Twitter envisioned, and in 2017, it replaced the dedicated Moments tab on mobile with a tab called “Explore” that also included search and trending topics. So when Twitter began curating trends, that became a significant part of the curation team’s mandate. The team follows a set of published guidelines and principles and maintains its own style guide.

Geary clarified during our interview that she’s not the decision-maker as to whether the “For you” and “What’s happening” tabs include automated trending topics, and her team doesn’t control those. Rather, the team is tasked with deciding which trends to contextualize and how to do so. Below is a transcript of our February 10 conversation, lightly edited for length and clarity.

OneZero: How did you become the person in charge of the team that writes Twitter’s “trending” descriptions?

Joanna Geary: I had been a journalist for The Guardian, the Times of London. Back in 2013, about a week after the IPO, I joined to do news partnerships management at Twitter U.K. In 2015, I moved on to this project called Project Lightning. I wanted to go back to something a bit more focused on the news or conversations that are happening. It started as curating Twitter Moments but has grown a lot in those five years.

We’re now a team that basically makes interventions to help people discover and get context on some of the bigger conversations that are happening on the platform. Basically, if it’s something that can’t necessarily be done algorithmically, we’re going to be there to assess and work alongside the algorithm. And that’s in a lot of different realms, from things that are not necessarily so visible, like topics, to stuff that’s very visible and very recent, like the trend descriptions. When there’s a gap between what customers want and what we can do algorithmically, then we’ll be bridging that gap.

So when you’re deciding which trends to write descriptions for and how to describe them, what do you see as your prime directive, your guiding light?

Our really specific role is to make sure people understand what it’s about when they see it. We’re attempting to — I’m trying to find kind of a polite way of putting it — we’re attempting to remove the “WTF?” from the trending tab. We want you to know that Betty White is just having her birthday. We don’t want you to think, “Oh no, she has died!” Our goal is to give you the gist so you’re not spending five minutes looking at the trend trying to figure out what it even means.

Can you give an example of a time when you faced a dilemma as to whether to write a description for a given trend or how to describe it?

There are so many. And I think this is the reason why it’s a uniquely human task at the moment: All trends are different. We deliberately try and bring people in with as many different backgrounds and skillsets as possible because on any given day we don’t necessarily know where these trends are going to be coming from.

So there’s this incredibly popular Japanese manga series, Attack on Titan. And the name of one of the characters started trending, and it was because they died. There was massive conversation around that when it happened. We wanted to contextualize that by explaining that people were coming to mourn the death of this beloved character. But in doing so, we ended up spoiling that plot twist. And, of course, fans are not afraid to give you their feelings when they feel you’ve ruined it for them. People weren’t coming to Twitter to be told Sasha was dead. They were coming because they wanted to discuss it, and if not, they didn’t want to know. So after that, we updated our style guide to be conscious of spoilers. And we changed the description to something like, “Fans weigh in on latest episode of Titan.”

But there are so many different reasons why trends can be complex. One might be — I’m sure you’ve gone into a trend and found, “Wait, this is three different conversations.” It’s difficult to contextualize a trend like that, so we might not. We are a human team, so we’ve got to choose the ones that have the biggest impact. We have an algorithm obviously that monitors the biggest trends within each country. And we’re trying to tackle the highest-volume trends, and we will do the ones where context is most needed. So nobody, I think, is going to criticize us if we skip #MotivationMonday, for example.

So you’re looking for the trends that are the most newsworthy?

I think newsworthiness is maybe slightly the wrong proxy. It’s more about the likelihood of something to trend. I don’t think Attack on Titan is a front-page headline, but it might be one of our top trends.

“We are a human team, so we don’t scale like algorithms do.”

Once you’ve chosen a trend, what’s the next step in figuring out how to contextualize it?

Well, there are a number of different ways we can do it. We can choose a representative tweet [to pull out and highlight]. We can choose to write a short description and title. We can do a title with a longer description. Or we can attach an Event or Moment to the trend. Each has different benefits depending on the context that’s needed.

You’ve talked about this team of people you have from different backgrounds working on your curation team. How big is this team? How many people are in fact “Twitter description guy?”

I don’t think that’s something we disclose publicly, but there’s more than one, I can tell you. (Laughs.) And they’re in more than one country. And I can tell you they’re not all guys — especially the ones who have helped us to monitor the Minecraft streaming communities, which are one of the places that “Twitter description guy” meme came from.

Some of the topics you deal with, such as Covid-19, are obviously complex and would seem to require some subject-matter expertise. Like, I can write about a lot of things for OneZero, but I probably wouldn’t cover Covid-19 because I’m not a science writer and I don’t have that background. What’s your process for those kinds of topics? Do you have domain experts on the team, or consult outside experts?

We do hire people from as wide a variety of backgrounds as our team size will allow. And we have briefings [on a given topic that’s in the news]. So there was a recent briefing on GameStop and the various trading rules that were coming into play. We’ve had briefings on Minecraft, on K-pop. Different people have different expertise they can bring, and they don’t hoard that expertise; they share it.

In terms of the actual descriptions themselves, we’re very, very tight on sourcing. Everything that’s placed within a Twitter description has to be sourced, and at least two credible organizations or credible individuals have to have provided that info before we can use it. So that’s another reason you might not see a description on a given trend. We don’t create content; we’re referencing and curating from individuals and organizations that have already put that out there.

Since Twitter went back to showing automated trends without context, there are times when it can seem like the trending box is undermining Twitter’s own efforts on something like misinformation. For example, on December 4, #Suitcasegate was trending on Twitter even as Twitter was labeling some of those tweets as disputed and potentially misleading — and limiting their spread — under its policy on election-related misinformation. Or on January 4, after the Washington Post surfaced audio of Trump pressuring Georgia secretary of state to “find” votes, Twitter’s trending section highlighted the Post’s story. But then right below it, an automated trending topic was “President acted properly,” which was a misleading Newsmax headline. Does that bother you when it happens? Does it sometimes feel like you’re fighting against the algorithm?

I don’t feel like it’s a fight, but I think — so, just a bit of context. As I mentioned, we are a human team, so we don’t scale like algorithms do. So we will have to pick what we contextualize within a set of rules, like, “Is it really high-profile? Does it need context?” And what you’re seeing on the “For you” page [which includes personalized trends based on users’ interests] is going to be so, so much more. In terms of topics people are engaging in, maybe they’re still high-profile trends but not as high-profile as the ones we’re focused on at a given moment.

I want people to be able to see the trends that are important for them. I don’t want for us to be saying, as a team, you can only see the trends we’ve chosen to contextualize. We want to be able to give people insight into the Twitter that is their Twitter. So an interesting challenge for my team in the next year or so is, “How do we take the thoughtfulness and nuance we can provide and scale that to some degree?”

Senior Writer, OneZero, at Medium

Sign up for Pattern Matching

By OneZero

A newsletter that puts the week's most compelling tech stories in context, by OneZero senior writer Will Oremus. Take a look.

By signing up, you will create a Medium account if you don’t already have one. Review our Privacy Policy for more information about our privacy practices.

Check your inbox
Medium sent you an email at to complete your subscription.

The undercurrents of the future. A publication from Medium about technology and people.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store