In the summer of 2013, Dick Costolo, the CEO of Twitter, reflected on his vision of the company as a “global town square.” The social network is “all public, real-time conversational, and widely distributed, and public is the first word in there,” he told an audience at the Brookings Institution.

Thousands of years ago, he added, the Greek Agora was “where you went to find out what was going on and talk about it, right? You came and talked about what was going on in your part of the village, and I came and talked about what was going on in mine, and the politician was there, and we listened to the issues of the day, and a musician was there and a preacher was there, etcetera, and it was multidirectional and it was unfiltered, and it was inside out, meaning the news was coming from the people it was happening to, not some observer. And, you know, along comes the printing press, and then radio, and then television, etcetera, etcetera, and all of these advances in technology are in service to removing the friction of distance and time in distributing the information. So we get to the point, ultimately, with CNN World Wide News, that you’ve completely eliminated the friction of time and distance, and then along comes a service like Twitter that has the elimination of time and distance built into it, but also brings back all those capabilities of the Agora. It’s inside out again, it’s coming from the participants.”

It was a bold vision of Twitter, which at the time was just seven years old, as the preeminent resource for accessing opinion across countries and cultures. And yet two years later, Costolo has resigned as CEO and Twitter is sputtering. In an earnings call late last month, interim CEO Jack Dorsey acknowledged that the growth of Twitter’s active-user base had slowed to an “unacceptable” rate.

This stagnation, and its implications, are often obscured by Twitter’s very real achievements. For instance, when Barack Obama traveled to East Africa in July, news outlets lavished attention on the Twitter hashtag #SomeoneTellCNN, a vehicle for ridiculing CNN’s assertion that the U.S. president was headed to a “hotbed of terror” in Kenya, as emblematic of Kenyan public opinion on the matter. And perhaps it was. (Kenyan Twitter is a vibrant community.) But Twitter’s insight into how people in the region were assessing Obama’s visit was far more limited than coverage of the various hashtags associated with the trip suggested. As the human-rights advocate Jeffrey Smith tweeted at the time, “People noticed that ‘Ethiopians aren’t tweeting about #ObamaInEthiopia.’ Well, internet penetration is 2% & there’s mass state surveillance.” To cite just one other example: A Pentagon official recently marveled that U.S. intelligence first learned of a Scud-missile launch in Yemen via Twitter. Wow! But the official didn’t mention that the vast, vast majority of Yemenis don’t use Twitter. We are still only hearing from a tiny sliver of the battlefield in Yemen.

Few hard numbers are available regarding the level of engagement of Twitter users and the service’s geographic reach. But a close look at the data that the company has made public suggests that if Twitter is indeed a global town square, it’s one that most of the town hasn’t entered yet—and one where the townsfolk who have entered seem to be doing more listening than talking these days. This reality has broader consequences for the promise of social media as a platform for hearing from and engaging with the world in unprecedented ways.

Using Twitter’s Streaming API, a live feed of approximately 1 percent of all public messages sent on the social network, I was able to chart the volume of tweets posted from January 1, 2012 through December 31, 2014. Between January 2012 and March 2013, the volume of monthly tweets in the Streaming feed doubled. However, from April 2013 through December 2014, monthly tweets remained roughly constant, suggesting that Twitter’s growth, as measured by total tweets, has stalled at an estimated 440 to 480 million tweets per day. (While it is unclear whether the Streaming API offers a representative sample of the entirety of Twitter, the timeline below does appear to align with several known data points about the growth in total tweets per day over time.)

Tweets per Month: January 2012 – December 2014

The data represents 1 percent of all tweets via Twitter’s Streaming API, and is incomplete for October 2012 and January and February 2014. (Kalev Leetaru)

Recently, Twitter has emphasized the size of its user community over daily tweet volume, announcing in April that it had reached 300 million “monthly active users,” up from 204 million in the first quarter of 2013. However, Twitter’s definition of a monthly active user includes automated spam accounts and users who simply read content from the service through customized filters rather than publish to it. In 2013, Twitter itself noted that “you don’t need to tweet to be on Twitter. … 40% of our users worldwide simply use Twitter as a curated news feed of updates that reflect their passions.” In other words, the conversation in Costolo’s proverbial town square appears to not be as “multidirectional” as he suggested. Many of those assembled are just eavesdropping.

The timeline below, derived from the Streaming API, shows the total number of Twitter accounts per month that posted at least one tweet. As with the total volume of tweets, the number of unique users sending those tweets appears to have leveled off in April 2013, remaining stagnant through the end of 2014. This suggests that the 100 million additional users that Twitter has added in the past two years are largely listeners instead of contributors.

Accounts Sending at Least One Tweet per Month: January 2012 – December 2014

The data represents 1 percent of all tweets via Twitter’s Streaming API, and is incomplete for October 2012 and January and February 2014. (Kalev Leetaru)

Twitter has reported that as of June 2015, 77 percent of its active users were located outside the United States. But the company has not publicized further details of its geographic reach, or disclosed what percentage of non-U.S. users post tweets versus only read those of others. One way to approximate such information is by assessing tweets that are geotagged, or endowed with the precise GPS coordinates where the tweet was sent, accurate to the resolution of a street address. As of November 2012, just 2 to 3 percent of all tweets had such coordinates—a percentage that, based on my analysis of the Streaming API, has stayed roughly constant since that time despite enormous growth in GPS-enabled smartphones.

What do geotagged tweets tell us about Twitter’s global presence? The lessons are limited, since Twitter has not divulged whether geotagged tweets are representative of the company’s geographic footprint. Still, mapping them is instructive. The animation below shows every location with two or more tweets per month in the Streaming API, and how that’s changed on a monthly basis between January 2012 and December 2014.

The visualization suggests that rather than growing outwardly and spreading to new regions, Twitter is largely growing inwardly and intensifying its coverage of locations where it was already popular, including the United States, Indonesia, and Japan. Twitter does seem to have expanded considerably in Latin America between 2012 to 2014, though the majority of the region outside major population centers remains unrepresented in geotagged tweets. Most of Africa and Central Asia is similarly blank, as is India, though the service appears to have gained some traction in northwestern India and northern Pakistan. Central and Eastern Europe have far fewer geotagged tweets than Western Europe does. And China—because of its ban on Twitter and domestic competitor, Weibo—is a void save for a few scattered clusters. In the Middle East, Twitter is making the most inroads in Turkey and Saudi Arabia.

Click here to expand the map. (Kalev Leetaru)

Even for those locations that are strongly represented through geotagged tweets, the global town square is less a single square than a collection of local squares. Language is one of the greatest barriers between these squares, with English forming a central linguistic hub on Twitter and Arabic, Chinese, and Hindi occupying a more peripheral position, according to a study of Twitter data from December 2011 to February 2012. Twitter is trying to surmount this language barrier through initiatives like the machine translation of tweets, which it launched during the 2013 unrest in Egypt but which has clear limitations as a catalyst for cross-cultural dialogue.

Yet linguistic differences can’t fully explain Twitter’s stalled growth abroad. After all, Facebook, which also plays host to multiple languages, has five times as many monthly users as Twitter does, and far greater penetration in countries like India and regions like Latin America and Central Europe. On the earnings call in July, Twitter’s leadership blamed the service’s user interface, noting that “we have not … made it easy for [people] to understand how to use Twitter” and that new initiatives to make the interface more user-friendly have “not yet had meaningful impact on growing our audience or participation.”

There is another potential explanation for Twitter’s stagnation, and it has to do with the way information is shared on the platform. Twitter has been unique among major social networks in that the majority of tweets are publicly accessible. In contrast, Facebook, and other surging players in this space like Snapchat, utilize a model more akin to email in which much of the content posted to the platform is private, available only to a given user’s network of friends. People currently seem to be gravitating toward social networks that emphasize control over message distribution, with a bias toward circumscribed communication rather than broadcasting to the entire world. Gary Vaynerchuk, an early Twitter investor, described this challenge in particularly blunt terms in March, suggesting that Twitter’s town-square model could be its undoing. Twitter has “become a massive firehose,” he said, and if it doesn’t “stop showing you everything, Twitter will die.”

This shift to more closed networks makes it more difficult for big-data researchers like myself, and the public at large, to realize Twitter’s lofty aspiration of tapping into what the whole world is thinking about any topic at any time. Further complicating matters, the majority of social-media posts in the past were textual, meaning that computer algorithms could rapidly assess their content and emotional attributes. Today, however, photo and video content form the fastest-growing segment of social sharing. As Google and Yahoo’s image-recognition algorithms have demonstrated, the technology to automatically recognize even the basic subject matter of an image remains in its infancy (Google’s have identified African Americans as “gorillas,” while Yahoo’s have identified Nazi concentration camps as “jungle gyms”). Measuring the tone of textual tweets about Russian President Vladimir Putin could easily be automated, but asking for the percent of photos shared on Instagram that portray him in an unflattering light is beyond the capability of current technology.

The future of Twitter is really the future of the global town square—an answer to the question of whether social media can really offer a frictionless, unfiltered forum for real-time conversation across countries and cultures, or whether those conversations will increasingly occur less visibly and more narrowly. To borrow the terminology of Stanford Law’s Jennifer Granick, the fading of the concept may represent the end of a certain kind of Internet dream.