How Many People Use Twitter's Own Apps?

23 July 2012

On the latest Talk Show, Gruber wondered what the spread of Twitter clients really is. It made me realise that no one has really explored how many people use Twitter’s own apps. I thought I’d (try to) find out.

Firstly, Twitter doesn’t report the information itself, so the only way to find out is by brute force. Due to the Twitter Streaming APIs, this is relatively straightforward. I ran a script which literally scanned tweets for their source link and counted each unique source.

Due to the time sensitivity of Twitter, finding one exact data set is impossible as client usage varies constantly. For instance, a big global event may encourage people to tweet on a day they otherwise wouldn’t. On a normal day, though, this method should be a good indicator of average user behaviour and app usage. For reference, the results that I collected are from a random sampling of tweets across a 9-hour period, approximately 9am to 5.30pm on the 18th July.

In total, I collected and analysed one million tweets. Whilst this seems like a lot, it still pales in comparison to daily tweet volume1. As always, I would have loved to collect more data, but there are, naturally, physical limitations to doing so. However, I still feel the study was significant enough to portray the trends to a good degree of accuracy and warrants publishing.

Below is a sorted list of the findings. Any app that created more than 0.02% of tweets2, in the period observed, is listed.

Source Count Client?
Web 243431 Yes
Twitter for iPhone 152116 Yes
Twitter for Android 116765 Yes
Twitter for Blackberry 115244 Yes
UberSocial for Blackberry 34502 Yes
Mobile Web 28563 Yes
TweetDeck 19627 Yes
Echofon 18718 Yes
Keitai Web 14381 Yes
twicca 12901 Yes
TweetCaster for Android 11746 Yes
Write Longer 11172 No
twitterfeed 11020 No
Facebook 10488 No
Twitter for iPad 9603 Yes
twittbot.net 8712 No
Tweet Button 8639 No
Janetter 7593 Yes
twipple 7515 Yes
Tweetbot for iOS 6845 Yes
Instagram 5917 No
Plume for Android 5621 Yes
SOICHA 5259 Yes
txt 5180 Yes
HootSuite 4560 Yes
UberSocial for Android 4543 Yes
Tumblr 3938 No
Tween 3909 Yes
dlvr.it 3327 No
ついっぷる for iPhone 3194 Yes
Google 3117 No
Twitter for Mac 2446 Yes
Twipple for Android 2438 Yes
jigtwi 2261 Yes
Twittascope 2201 Yes
Ask.fm 1940 No
m.tweete.net 1866 Yes
TweetList 1831 Yes
foursquare 1825 No
Saezuri 1743 Yes
Tweetlogix 1649 Yes
YoruFukurou 1410 Yes
movatwi.jp 1322 Yes
ツイタマ 1263 Yes
Samsung Mobile 1262 No
Seesmic 1134 Yes
TweetCaster for iOS 1113 Yes
ついっぷる for iPhone 1100 Yes
Tweet Old Post 1097 No
yubitter 1052 Yes
ShootingStar 983 Yes
UberSocial Mobile 948 Yes
Tuitwit 938 Yes
jigtwi for Android 882 Yes
OpenTween 815 Yes
A.plus for Blackberry 779 Yes
Twipple for Android 764 Yes
SocialScope 759 Yes
Twitter for Windows Phone 745 Yes
IFTTT 720 No
Twiterous 707 Yes
Twidroyd for Android 688 Yes
EasyBotter 650 No
Tweet ATOK 639 Yes
Social by Nokia 612 Yes
TwitBird 609 Yes
ついっぷる Pro for iPhone 553 Yes
Addictweet 511 Yes
Krile2 500 No
Photos on iOS 493 No
UberSocial for iPhone 489 Yes
twitcle 462 Yes
twtkr 459 Yes
CitCuit 442 Yes
UbеrSociаl for BlackBerry 439 Yes
MetroTwit 438 Yes
vk.com 435 No
Dabr 429 Yes
Ustream.TV 420 No
ニコニコ動画 413 No
Pinterest 404 No
TweetList Pro 398 Yes
BotMaker 395 No
Twil23 394 No
SocialOomph 388 No
TwitCasting 384 No
TwitPal 377 Yes
buysellyourtweets 377 No
UberSocial 368 Yes
PlayStation®Vita 357 Yes
TwitMania 357 Yes
Silver Bird 357 Yes
Osfoora for iPhone 354 Yes
Teewee 321 Yes
STATUSEM 321 No
Tower Heist Takeover, BlackBerry 317 Yes
Twittelator 316 Yes
mixi ボイス 309 Yes
Buffer 309 No
Twipple Pro for Android 302 Yes
ツイ助 299 No
Nimbuzz Mobile 298 No
Ultwimate Mobile 285 Yes
HTC Peep 274 Yes
モバツイ touch 273 Yes
LinkedIn 261 No
Twitscoop 253 Yes
vk.com pages 249 No
Camera on iOS 248 No
weheartit.com 242 No
Paper.li 237 No
Twitter for iAppli 232 Yes
Hotot for Chrome 232 Yes
TwitIQ 231 Yes
Azurea for Windows 212 Yes
モバツイsmart / www.movatwi.jp 210 Yes
Follow to Follow 206 No
Twittanic 203 Yes

The raw data, by itself, is useless. In order to determine how many people would be affected by a ban on third-party clients, the first step is to identify anyone using native clients. There are a handful of them: Twitter.com, m.twitter.com, Twitter for iPhone, Android, Blackberry, Windows Phone, iPad, Mac as well as TweetDeck and Keitai Web (Twitter’s Japanese client) and SMS (identified as “txt”). Twitter recently announced a client for Nokia S40, but I don’t think it has caught on — zero of the million tweets originated from this client.

The total of these rows is 708,101 out of 1,000,000.

Therefore, this means that at least 70.8% of the total originated from first-party clients, and at most 29.2% of people use third-party Twitter clients.

Already, first-party apps clearly have a monopoly. The actual share of first-party usage is even higher, however.

This is because not all of the observed apps are actually “clients”. Many are simply apps which post to Twitter. Instagram is an example of this; it just posts links back to its own service. The “Tweet Button” is a first-party example: it allows the user to tweet, but it isn’t a client. Therefore, it is necessary to filter out these “non-clients” to show a true representation of first-party-client dominance.

In an ideal scenario, a human would individually assess the ‘clientness’ of each app that created a tweet in the period. However, this is too time-consuming to be feasible. Instead, I have hand-checked only the 118 apps listed above, and use that ratio to extrapolate across the 291,899 remaining tweets to gain a relatively-accurate estimate.

My assessment of each app is denoted in the third column of the table. I found that 36 of those apps were not clients (identified by the third column). Applying that ratio4 (30.5%) to the 291,899 tweets, it is estimated that 89,053 tweets were not from clients. By discounting these ‘invalid’ tweets, the overall bucket is reduced, thereby increasing the final proportion of first-party app usage to 708,101 out of 910,947 tweets, equivalent to a percentage share of over 77%.

For people that think Twitter will never ban third-party clients because there would be too much backlash, I think this 77% figure shows that Twitter could do it with ease. A large portion of the 23% would be happily herded to a first-party client, as they don’t really care what app they use — it just turned out that the client they first downloaded wasn’t a Twitter-owned app. The only people who would care would be the geeks, like me and anyone else who could be bothered to read this post, who actually care about the client they are using. And let’s face it, Twitter doesn’t care about geeks.

1 This is reported to be 340 million daily tweets, as of March 2012.

2 This accounts for the 46,628 tweets that are missing from that list; any client that produced less than 200 tweets has been omitted.

3 The full name of this app is “Twil2 (Tweet Anytime, Anywhere by Mail)”. I shortened it in the table as it was unnecessarily wide.

4 In fact, I think that that this ratio should be even higher, as I would expect a lot of the smaller sources are from obscure apps that simply post to Twitter. There is a bias for clients to produce proportionally more tweets (which are more likely to appear in the list I checked, thus skewing the ratio) than non-clients which is not factored in the calculations above.