Thread by @Netzdenunziant

I have been asked for an English thread about Twitter's shadow banning practice. So here it is. An open source test that recognizes various types of shadow bans is available at shadowban.eu.
Why the uproar now? Because new ban methods appear to be politically biased.

The most famous ban is the conventional shadow ban. It causes your tweets to only be visible to your followers. In this case, retweets will not have any effect and tweets are hidden to non-followers within threads. Non-followers can only reach your tweets by visiting your /+

profile or directly through URLs. The screenshot below illustrates this ban method. Because @_JoeyMcGuire is affected by this type of ban, his tweets are hidden if you did not browse to their respective URL specifically.

Archive:
archive.fo/ITAva
archive.fo/BielX

If you have a conventional shadow ban, your tweets are not found by the Twitter search at all. This is true even for search queries with the "from:" operator. We test for this ban method by running a search with "from:" and look whether we find detached tweets on your timeline.

There is also the less restrictive search ban. It only hides your tweets from the search function in the same way the conventional ban does. However, this type of ban does not cause tweets to be detached and tweets continue to be visible to non-followers. At this time, I do /+

not know whether this type of ban impacts your reach negatively in any additional way.
I have seen people getting rid of conventional and search bans already. Thus, some of these bans are temporary. Nevertheless, I have seen inactive accounts that are still affected, too.

Now it gets a more interesting: I was never able to tell whether these bans were politically biased. (Conventional shadow bans are known since 2016. There are people who are convinced that there have been conventional shadow bans for political reasons: breitbart.com/tech/2016/02/1…)

However, this is completely different for the new method which we call QFD, which is an acronym for quality filter discrimination. QFD causes your tweets to not appear within the latest search results if the quality filter is turned on. Again, this is true for hash tags as well.

Your tweets will still appear within the top section. But it appears that comparable tweets, regarding impact, of affected users are ranked way below those of users that are not affected by QFD. People have tested this and more recent tweets of affected users with more likes /+

and retweets were ranked below others.
This has been tested, among others, by @aufklaerer2017 who published the following popular tweet which does not appear at any relevant position despite having most activity:

https://twitter.com/aufklaerer2017/status/1009393154951434240

Archived search:
archive.fo/DMEen

But what exactly is the search filter? The search filter is a well-hidden search feature that needs to be turned off again for every search. This is illustrated below. An exception is the "from:" operator causing the filter to be ignored. Within the first days after /+

introducing the QFD method, this was different. Therefore, the affected accounts were discovered by the old shadowban test hosted at shadowban.de/tester which simply detects whether an account is search banned by searching for its username using the "from:" operator.

When Twitter introduced its "healthy conversation" algorithm that is officially supposed to ban trolls, many users suddenly noticed that the test reported that they were shadow banned. In fact, this includes the users that are now still affected by QFD.

blog.twitter.com/official/en_us…

Thus, the @AfD, a German party, contacted Twitter and was told that this was a "bug". Obviously, this is a very unspecific response and since the QFD ban (still) exists, I assume that the bug was that the shadow ban test found the accounts affected by QFD.

https://twitter.com/AfD/status/996808592572469248

At this time, I tested all my then 3919 Twitter contacts using a "from:" query and noticed that I did not find any tweet for about 44 % of users I follow or that follow me. This is especially noteworthy since I also follow mainstream accounts. Trump supportes also noticed that /+

many of them were shadow banned. This includes @USWorkers4Trump who initiated #ShadowBanTwitter. I repeated the test procedure for accounts that are rather left-leaning and obtained a share of only up to 6 % affected users. (This includes search bans I believe to be unpolitical.)

It appears that the machine learning algorithm that is used by Twitter has a very strong political bias. There are strong hints that mainly the right-wing opposition is affected by QFD in Germany and I believe this is true for other countries as well, including the United States.

In Germany, many affected users have started to use the ❌ symbol in their name in order to protest censorship through the QFD measure and to raise awareness that algorithms are used in order to censor political opinions that are unconvenient to the left-leaning Silicon Valley.

I have conducted a small QFD analysis. Shares of affected users:
My followers: 50 %
Followings: 49 %
I also tested contacts of two left wing extremists (@DietherAst and @FrauLavendel) and the number of affected accounts was below 1 %. Not a single affected left-leaning user.

I have tested known Twitter accounts of German Members of parliament. (twitter.com/wahl_beobachte…)
14 out of 509 accounts were affected by #QFD. All affected MPs belong to the AfD, which is a German right-wing party.

Raw data of my analysis is available at pastebin.com/mPLbW727.

According to a list by @CSPAN (twitter.com/cspan/lists/me…), the only account belonging to a US Member of Congress that is affected by #QFD is @RepMattGaetz, a conservative Republican.
Archived search:
archive.is/ZXH0F
archive.is/DsmDo

Raw data:
pastebin.com/RLqMiPMJ

There is a mainly Arabic Twitter botnet generating new followers as soon as you follow a few bots. At least twelve users have managed to escape the #QFD by generating a large number of new followers through this botnet causing their share of #QFD followers to drop (below ~ 20 %).

The claims that Twitter does not shadow ban are false. "Spammy" users are indeed shadow banned using ghost commenting techniques.
archive.fo/https://twitte…
archive.fo/bzY6a
archive.fo/vFRss
archive.fo/AKlar
archive.fo/1JpcH
vimeo.com/282066400