UPDATE 1/30 @ 4:39: I created a Twitter account that will send out links to all non-DMCA takedown notices as they come in. Just follow @TakedownTweets and please RT to spread the word!

It’s probably a little too late, but before we all freak out and boycott Twitter further, let’s look at who has been requesting takedowns, what media were involved, and where they are from. Twitter posted 4411 cease & desist notices on Friday through Wendy Seltzer’s Chilling Effects website. The Twitter release of takedown notices was strangely timed, immediately following the height of SOPA protests and alongside their (more criticized) public statement that they will work with local authorities to filter tweets accessed from specific countries.

CLICK ON THE MANY EYES LINKS FOR INTERACTIVE VERSIONS. They are really quite a lot more usable.

Takedown requests by media
[ link to many eyes ]

What kind of stuff is being taken down? First, they are all DMCA (Digital Millenium Copyright Act) requests. Every one. So Twitter isn’t censoring protesters yet – if they do you can bet you will hear about it – and there’s good reason to think they are trying to err on the side of not censoring tweets. For one, there is a pretty easy work-around, at least of today, 1/28/12. Second, Twitter is probably holding its nose and retaining a good amount of leverage with what is a “reasonable” request. DMCA requests are also totally run of the mill for any service provider and you kind of must comply to stay in business. As Nancy Messieh puts it, “Twitter isn’t censoring you. Your government is.”

Not surprisingly, the requests are mostly related to links to media like music (1225), DVDs [I remember those things?] (142), and movies (141). The company Web Sheriff seems to be particularly vigilant in making requests, and for some reason have their own category of request. So does cricket (hey, it’s popular worldwide). Only 27 out of the 4411 were for adult material. Twitter isn’t a huge porn trading base, with the exception of some pesky bots. A small number (13) were for avatars. Yes, those little images that you mostly see as 48 x 48 squares. Another kind of amusing side to this is that the results on ChillingEffects for images mostly seem to include the links to the image source. (Should we make a takedown request for… a takedown request?)

Takedown requests by requester
[ link to many eyes ]

Who is making the requests? the big media conglomerates are there, but really not in the numbers that you might expect compared with the numbers that they are probably sending out to, say, cable modem subscribers that download shaky hand-cam torrents of pirated movies pre-opening day. Magnolia Pictures, a “theatrical and home entertainment distribution arm of the Wagner/Cuban Companies, a vertically-integrated group of media properties co-owned by Todd Wagner and Mark Cuban,” took the lions share.

Takedown requests by Country
[ link to many eyes ]

Where are they? The UK submitted more takedown requests (2194) than the USA (1815), mainly on the back of Web Sheriff, which is based out of England. Next in line was Belgium with 164. China and Saudi Arabia racked up 1 request each. Hardly a case of, as Forbes put it when talking about withholding Tweets based on local government, “accepting defeat by fascists.” Or, if they are fascists, right now they are American and British.

Takedown requests by month
[ link to many eyes ]

When were they made? I don’t put much stock in this data because it seems like companies accumulate lists of problematic tweets and then push out a bunch of requests in a row. So take the above graph with a grain of salt.

What does it mean? The takedown notice data reinforce several points. It really throws into stark relief how ineffective cease & desist orders are in curtailing the flow of content. Requests are mostly about tweets that have links to copyrighted material, such as through a torrent or download site. TechCrunch made the observation that some of the notices referenced bots, or automated programs that automatically collect and relay messages. Cease & desist orders to Twitter do nothing to curtail the spread of media. Remember, Twitter doesn’t host these media and the number of C&Ds is miniscule compared to the billions (trillions!) of tweets out there. It’s also mostly a conversation between and western media conglomerates and tech companies. Given that the requests so far rely on DMCA, this is also not surprising.

Personally I feel that Twitter is doing pretty much the minimum they can to stay in business. Like complying with laws of the United States in a paranoid climate from the Megaupload affair, leading to services such as Filesonic turning off the ability to share files with others and effectively removing any utility the site had. Twitter has done a valuable service by making takedown notices public, which they aren’t required to do. What would be more useful is if they were in a form that could be more easily accessed and reusable than a table-based HTML page. Google’s transparency report is quite comprehensive and provides a raw data set, PDF, and data visualizations. There are a lot of data not available in the visualizations above because they are difficult to scrape from the site.

If anyone wants the complete tab-delimited text file for the scraped data, it has some funky data that needs to be cleaned. You can see this in the set below, which has multiple requesters that should be consolidated. But I’m happy to post it if there’s any interest, just msg me on Twitter @aschrock.


  • Data were collected using a “web scraper,” which traverses a website much as you do with a web browser, and automatically collects textual data. Because not all of the DCMA notices followed the same format, not all data could be automatically collected in all cases. They probably could be.
  • Each DMCA notice may be linked to several (likely related) incidents. The above data only looks at the first one. So the data set is of takedown requests, not of the total amount of offending media/links posted on Twitter.
  • The data set used to generate the graphs above is public on IBM’s Many Eyes site. It contains the same information as on the website, just in a more easily-accessible format.