Shadowbanning

Risius, Ma en; Blasiak, Ke in Ma c
A icle — Published Ve sion
Shadowbanning
Business & In o ma ion Sys ems Enginee ing
P o ided in Coope a ion wi h:
Sp inge Na u e
Sugges ed Ci a ion: Risius, Ma en; Blasiak, Ke in Ma c (2024) : Shadowbanning, Business &
In o ma ion Sys ems Enginee ing, ISSN 1867-0202, Sp inge Fachmedien Wiesbaden GmbH,
Wiesbaden, Vol. 66, Iss. 6, pp. 817-829,
h ps://doi.o g/10.1007/s12599-024-00905-3
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/315771
S anda d-Nu zungsbedingungen:
Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen
Zwecken und zum P i a geb auch gespeiche und kopie we den.
Sie dü en die Dokumen e nich ü ö en liche ode komme zielle
Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich
machen, e eiben ode ande wei ig nu zen.
So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen
(insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en,
gel en abweichend on diesen Nu zungsbedingungen die in de do
genann en Lizenz gewäh en Nu zungs ech e.
Te ms o use:
Documen s in EconS o may be sa ed and copied o you pe sonal
and schola ly pu poses.
You a e no o copy documen s o public o comme cial pu poses, o
exhibi he documen s publicly, o make hem publicly a ailable on he
in e ne , o o dis ibu e o o he wise use he documen s in public.
I he documen s ha e been made a ailable unde an Open Con en
Licence (especially C ea i e Commons Licences), you may exe cise
u he usage igh s as speci ied in he indica ed licence.
h p://c ea i ecommons.o g/licenses/by/4.0/
CATCHWORD
Shadowbanning
An Opaque Fo m o Con en Mode a ion
Ma en Risius •Ke in Ma c Blasiak
Recei ed: 21 Feb ua y 2024 / Accep ed: 26 Sep embe 2024 / Published online: 28 Oc obe 2024
ÓC own 2024
Keywo ds Shadowbanning Visibili y educ ion T us
and Sa e y Con en mode a ion Algo i hmic con en
mode a ion Social media
1 In oduc ion
Social media pla o ms ace nume ous socie al, e hical, and
poli ical issues. Online ex emism (Spieke mann e al.
2022), disin o ma ion campaigns (S a bi d e al. 2019),
ha e speech (Oksanen e al. 2020), and cybe bullying
(Chan e al. 2019) a e jus a ew examples o he social
media ela ed p oblems. Social media pla o ms ha e
esponded by implemen ing a ious con en mode a ion
mechanisms o go e n communica ion on social media
pla o ms (G immelmann 2015). Con en mode a ion is a
apidly g owing US$ 9.8 Bn ma ke (Bloombe g 2022;
Wankhede 2022). Pa icula ly algo i hmic con en mode -
a ion is an inc easingly popula app oach (Ka zenbach
2021). Algo i hmic con en mode a ion encompasses pla -
o m design decisions ha dic a e how communi y mem-
be s in e ac wi h one ano he and de e mine who ge s o
see which con en (Du y and Meisne 2022; Zeng and
Kaye 2022). Algo i hmic con en mode a ion o e s scal-
able, au oma ed sys ems ha classi y use -gene a ed con-
en o in o m go e nance decisions (e.g., emo al,
geoblocking, accoun akedown) (Go wa e al. 2020;
G immelmann 2015).
The cu en discussion on con en mode a ion pays li le
a en ion o shadowbanning (Gillespie 2022a; Gillespie
e al. 2020). Shadowbanning sec e ly demo es o sup-
p esses isibili y o use s, con en , o g oups wi hou
ale ing he a ec ed en i y (Gillespie 2022a). A ecen
su ey o 1,000 U.S. social media use s ound ha abou
10% o esponden s – ypically non-cisgende ed, Hispanic,
o Republican use s epo being shadowbanned ac oss all
majo social media pla o ms like Facebook, Twi e ,
Ins ag am, Reddi , and TikTok (Nicholas 2022). Shadow-
banning is he concep ual coun e pa o pla o m’s
ampli ica ion o p oblema ic con en o he sake o
boos ing engagemen h ough ecommende algo i hms
(Gillespie 2022a). Social media pla o ms gene ally a oid
using he e m shadowbanning as pa o hei con en
mode a ion mechanisms. Ins ead, hey e e o i as isi-
bili y educ ion echniques (Gillespie 2022b). Social media
pla o ms ha e good easons o use shadowbanning as i
allows hem o con ain unwan ed con en wi hou eleasing
in o ma ion ha would help malicious ac o s adjus hei
ac ics and a oid de ec ion (e.g., spam bo s), o mi iga e
access o undesi able con en (e.g., suicide, p o-ea ing
diso de ) (Nicholas 2022), o o a oid pola iza ion and
public ou c y esul ing om ce ain con en mode a ion
decisions (Gillespie 2022a).
Accep ed a e wo e isions by Susanne S ah inge .
M. Risius (&)
In o ma ion Managemen , Uni e si y o Applied Sciences Neu-
Ulm, Wileys aße 1, 89231 Neu-Ulm, Ge many
e-mail: [email p o ec ed]
M. Risius
Adjunc Senio Fellow, School o Psychology, Uni e si y o
Queensland, S . Lucia, B isbane, QLD 4072, Aus alia
K. M. Blasiak
Cen e o Technology & Socie y, TU Wien, Gußhauss aße
27-29, 1040 Vienna, Aus ia
K. M. Blasiak
In o ma ics, TU Wien, Fa o i ens aße 9-11, 1040 Vienna,
Aus ia
123
Bus In Sys Eng 66(6):817–829 (2024)
h ps://doi.o g/10.1007/s12599-024-00905-3
Shadowbanning is also hea ily sc u inized, p edomi-
nan ly o i s opaci y. Shadowbanning p e en s use s om
co ec ing o dispu ing con en mode a ion decisions
(Nicholas 2022), lea ing use s le o specula e abou
whe he hey ha e been shadowbanned (Delmonaco e al.
2024) and unclea abou he c i e ia ha igge shadow-
banning (Elmimouni e al. 2024). Shadowbanning is
accused o sys ema ic bias agains mino i ies (Du y and
Meisne 2022). In a ecen con en mode a ion su ey ha
o e sampled ma ginalized iden i ies (i.e., acial and e hnic
mino i ies, LGBTQ ?people, ans and/o nonbina y
people), 21.78% o esponden s epo ed expe iencing
shadowbans (Delmonaco e al. 2024). Subjec s o shad-
owbanning epo men al and emo ional ha m (Nicholas
2022) anging om eelings o us a ion, sadness (Del-
monaco e al. 2024), ma ginaliza ion, anxie y, and help-
lessness (Elmimouni e al. 2024), leading o sel -
censo ship, wi hd awal om social media, and inancial
damages (Delmonaco e al. 2024). The po en ial ami ica-
ions o shadowbanning a e expec ed o ex end a beyond
he silenced indi iduals o mino i y g oups di ec ly a ec-
ed. This mechanism is belie ed o e ode us and con i-
dence in social media pla o ms, os e ing an en i onmen
conduci e o conspi acy heo ies (Chen and Zaman 2024).
Fo ins ance, shadowbanning uels belie s ha pla o ms
hold biased agendas, such as aligning wi h speci ic go -
e nmen s (e.g., ‘‘pla o ms align wi h he s a e o Is ael’’).
Simila ly, shadowbanning can exace ba e socie al pola -
iza ion by il e ing ce ain opinions o indi iduals om
public discou se. This can bias he p ocess o o ming
public opinion, o example, when es ic ing p o-Pales-
inian oices on Facebook du ing he Is ael-Hamas wa
(Elmimouni e al. 2024) o TikTok’s supp ession o
#BlackLi esMa e and LGBTQ ?con en (Delmonaco
e al. 2024). Finally, shadowbanning can be weaponized by
malicious ac o s o silence dissen ing oices (Nicholas
2022), unde mining open dialogue and empowe ing hose
who seek o manipula e online discou se.
A balanced and in o med discussion o shadowbanning
is u gen ly needed. Rela ed esea ch is s ill nascen and – o
he bes o ou knowledge – absen in he ield o In o -
ma ion Sys ems (IS). The objec i e o his a icle is o
in oduce IS p ac i ione s and esea che s o shadowban-
ning. By in oducing shadowbanning, we aim o make
h ee key con ibu ions. Fi s , we con ibu e o he eme -
gen li e a u e ha aises awa eness o his opaque con en
mode a ion mechanism (Gillespie 2022a). Shadowbanning
complemen s exis ing IS esea ch on con en mode a ion
beyond he mo e commonly discussed o ms o anno a ing
(He e al. 2024; Kim and Dennis 2019; Kim e al. 2019),
banning (Russo e al. 2023), blocking (McDonald 2022),
and depla o ming (Kelle 2019). Bu also o he ela ed IS
esea ch on pla o m go e nance (Halckenhaeusse e al.
2020), algo i hmic audiencing (Rieme and Pe e 2021),
and algo i hmic con ol (Benlian e al. 2022) ough o
conside he ole o algo i hms in sec e ly demo ing con en
– in addi ion o hei augmen ing, ampli ying, and
se endipi ous e ec s (Milli e al. 2023). Second, we o e
concep ual cla i y in o wha cons i u es shadowbanning.
Building a common unde s anding o shadowbanning
ough o help b idge he gap be ween social media pla -
o ms who a oid he e m shadowbanning (Gillespie
2022b), use s who specula e whe he hey ha e been sub-
jec ed o his p ac ice (Elmimouni e al. 2024), esea che s
who y in es iga e his phenomenon (Jaidka e al. 2023),
and lawmake s who need o unde s and his mechanism o
de ise meaning ul egula ion (Nicholas 2023). Thi d, we
ou line ways in which in o ma ion sys ems esea ch wi h
i s ocus on socio echnical sys ems can help add ess and
in o m he ela ed con e sa ion a ound con en mode a ion,
censo ship, eedom o exp ession o con ibu e o socie y
and make he online en i onmen sa e o e e yone
(Sa ke e al. 2019; Spieke mann e al. 2022).
2 Backg ound
2.1 De ini ion and O igins o Shadowbanning
The e m shadowbanning i s appea ed in 2001, whe e i
e e ed o he mechanism o emo ing pos s o e e yone
else excep o he pos e in an online o um (Sa olainen
2022). I eached b oade public awa eness in 2018 when
US conse a i es began accusing Twi e (now X) o
shadow-banning ‘p ominen Republicans’ by no sugges -
ing hem on he pla o m’s au o ill d op-down sea ch ba
(Sa olainen 2022; S ack 2018). Shadowbanning ades
unde many di e en names such as s eal h banning, ghos
banning, hell banning, commen ghos ing, isibili y mod-
e a ion, ( isibili y) educ ion, supp ession, bo de line
con en policies, and unin o med o undisclosed con en
mode a ion (Gillespie 2022a,2022b; Jaidka e al. 2021;
Nicholas 2023). Shadowbanning supp esses he isibili y
o each o con en , use s, o g oups wi hou ale ing he
a ec ed pa y (Gillespie 2022a). Pla o ms limi he con-
di ions unde which he con en ci cula es, o example,
whe he i appea s as a ecommenda ion, a sea ch esul , in
a news eed and use s’ que ies, o in a s eam o commen s
(Gillespie 2022b). I is an opaque mechanism whe e con-
en emains heo e ically accessible and isible o (some)
use s (e.g., he pos e hemsel es) (Sa olainen 2022). I is
impo an o emphasize ha oday’s unde s anding o
shadowbanning by demo ing people, g oups, o con en
wi h algo i hmic suppo is much mo e nuanced han he
o iginal 2001 mechanism o simply wi hholding con en
om anyone (Gillespie 2022b).
123
818 M. Risius, K. M. Blasiak: Shadowbanning, Bus In Sys Eng 66(6):817–829 (2024)
Fou main ypes o shadowbans (see Table 1) can be
iden i ied (Jaidka e al. 2023): (1) Ghos bans a e he mos
es ic i e ype o shadowbans ha ob usca e con en by
allowing only he accoun holde s hemsel es o see hei
own con en (Jaidka e al. 2023). (2) Sea ch bans p ohibi
he in ended encoun e wi h use s o con en by emo ing
hem om he pla o ms’ in e nal sea ch index (e.g.,
qua an ined sub eddi s om Reddi ’s in e nal sea ch, alse
pos s om Ins ag am’s hash ag sea ch) (Goldman 2021).
(3) Sea ch sugges ion bans emo e con en om he pla -
o m’s in e nal sea ch engine’s ‘‘au o-sugges ’’ ea u e
(Goldman 2021) so ha he accoun does no appea when
o he s look i up in he sea ch in e ace (Jaidka e al. 2023).
(4) Down ie ing dep io i izes con en o limi he unin en-
ional exposu e and o make i unlikely o o he use s o
ind i by hiding eplies unde an in e s i ial and only
loading when p omp ed (Jaidka e al. 2021,2023; Nicholas
2022; Ryan e al. 2020), downg ading he in e nal sea ch
isibili y, educing in e nal p omo ion (e.g., ecommenda-
ions), and educing o emo ing na iga ion links (e.g.,
explo e pages) (Goldman 2021).
Shadowbanning di e s om he s anda d con en cu a-
ion mechanisms employed by social media pla o ms
h ough ecommende algo i hms. Con en ecommenda-
ion ollows a di e en se o conce ns and p io i ies
(Gillespie 2022a). Recommende sys ems and news-
eeds selec o wha is deemed mos appealing o op imize
o engagemen measu ed by he ime spen on he pla -
o m, he ypes o ac ions aken, and sa is ac ion p oxies
(Buche 2018; McKel ey and Hun 2019). They collec and
analyze use da a oge he wi h he co pus o all a ailable
con en o pe sonalize use eeds and maximize use
engagemen . Con en ha is mo e posi i e – e.g., in e ms
o ecency, se endipi y, close ie popula i y – has g ea e
likelihood o be selec ed by ecommende algo i hms.
Shadowbans a e no simply an ou come o byp oduc o
ecommende algo i hms because shadowbanned con en is
ac i ely supp essed o obscu ed, e en in sea ch esul s
(Jaidka e al. 2023). Shadowbanning aims o selec ou -
wha is deemed leas appealing based on nega i e signals
ha indica e an i em ough no o be ecommended
(Gillespie 2022a). The gold s anda d o T us & Sa e y
pe o mance is ‘p e alence’ and shadowbanning aims o
minimize he unc ion o how o en use s iew ce ain
con en (Fishman and Ha is 2023).
We b oadly unde s and shadowbanning as social media
pla o ms’ use o algo i hms o align use beha io ,
accoun s, and con en wi h o ganiza ional objec i es. This
concep ualiza ion closely esembles he p e alen de ini-
ion o algo i hmic con ol (Wiene e al. 2023), being
embedded in o a b oade o ganiza ional con ex (Alizadeh
e al. 2023) wi h human con olle s (C am and Wiene
2020) and o ganiza ional in en ions (Kellogg e al. 2020;
Sulli an e al. 2024), ul illing sanc ioning unc ions –
among o he s – (Hi sch e al. 2023), complemen ing and
elie ing bu no en i ely eplacing human ac o s (Wiene
e al. 2023). Cu en esea ch on algo i hmic con ol
unc ions (Alizadeh e al. 2023; Hi sch e al. 2023; Kellogg
e al. 2020; Sulli an e al. 2024) is ye o conside isibili y
educ ion echniques om pla o ms like Ube (Ube
2023). Shadowbanning’s implici o m o algo i hmic
con ol deli e y by a oiding o ale use s also appea s o
be an ex eme o m o he p e iously in es iga ed algo-
i hmic opaci y and limi ed anspa ency o algo i hms
(Mo
¨hlmannn e al. 2023). Gi en ha algo i hmic man-
agemen esea che s ecognize he ela ion o algo i hmic
con ol o b oade o ganiza ional con ex s beyond wo k
se ings (Came on e al. 2023), we d aw on he ela ed
algo i hmic managemen and con ol li e a u e (Benlian
e al. 2022) o guide ou concep ualiza ion o shadowban-
ning in he ollowing (Fig. 1).
In he con ex o shadowbanning, he human con olle s
a e T us & Sa e y eams who de ine and communica e
o ganiza ional in en ions in he o m o pla o m policies o
policy iola ions, help implemen policies in o con en
mode a ion algo i hms, manage o conduc con en
Table 1 Summa i e o e iew o di e en ypes o shadowbans
Shadowban
ype
Desc ip ion Ou come Examples om li e a u e
Ghos bans Con en is hidden om e e yone excep
he accoun holde
Comple e in isibili y o o he s Jaidka e al. (2023)
Sea ch bans Con en is emo ed om he pla o m’s
in e nal sea ch index
Reduced disco e abili y,
con ained isibili y
Goldman (2021)
Sea ch
sugges ion bans
Con en is emo ed om he pla o m’s
sea ch ‘‘au o-sugges ’’ ea u e
Dec eased isibili y, po en ial
loss o o ganic each
Goldman (2021)
Down ie ing Con en ’s isibili y is educed o limi ed Dec eased isibili y, educed
engagemen
Jaidka e al. (2021); Jaidka e al. (2023);
Nicholas (2022); Ryan e al. (2020)
123
M. Risius, K. M. Blasiak: Shadowbanning, Bus In Sys Eng 66(6):817–829 (2024) 819
mode a ion hemsel es, and deal wi h con en mode a ion
appeals (Fishman 2023). Shadowbans a e ypically au o-
ma ed (e.g., o educe isibili y o spam bo pos s) bu s ill
allow human mode a o s o en o ce shadowbans di ec ly
(e.g., o occul ce ain ypes o use s) (Biddle e al. 2020;
Du y and Meisne 2022). Algo i hmic con en mode a ion
au oma es he human T us & Sa e y eams mode a ion
asks (He e al. 2024) in o de o gain scalabili y and
consis ency in con en mode a ion (Jiang e al. 2023). Fo
shadowbanning his in ol es he algo i hmic de ec ion o
unwan ed con en o use s based on T us & Sa e y policies
and enac men o espec i e ghos -, sea ch-, sea ch sug-
ges ion ban, o down ie ing. Shadowbanning solu ions can
be de eloped by pla o ms hemsel es o sou ced om
ex e nal en i ies (e.g., Reddi ’s Au oMode a o ) (W igh
2022).
2.2 The Role o Shadowbanning as a Fo m o Con en
Mode a ion
Shadowbanning is pa o social media pla o ms b oade
con en mode a ion e o s o con ain he p e alen online
issues which means ‘go e nance mechanisms ha s uc-
u e pa icipa ion in a communi y o acili a e coope a ion
and p e en abuse’ (G immelmann 2015, p. 47). While
con en mode a ion mechanisms a e hypo he ically unlim-
i ed om a echnological s andpoin , esea ch has iden i ied
a ange o common mechanisms (Goldman 2021)
(Table 2). These con en mode a ion mechanisms a y (1)
in hei deg ee o se e i y and (2) ega ding hei unde -
lying philosophy.
The deg ee o se e i y is commonly b oken up in o
ei he ‘‘ha d’’ o ‘‘so ’’ ypes o con en mode a ion
(Zanne ou 2021). Ha d mode a ion means o suspend,
block o emo e con en o en i ies om social media
pla o ms (Go wa e al. 2020). So o ms o con en
mode a ion wa n abou con en o con ain i s impac
wi hou suspension o ake-downs (Jaidka e al. 2023).
Con en mode a ion mechanisms’ unde lying philosophies
di e in he deg ee o which hey ely on nu u ing o
punishing o c ea e a posi i e online en i onmen (Jiang
e al. 2023). Punishing is a eac i e app oach ha p ima ily
ocuses on applying consequences o ule- iola ing
beha io s. Nu u ing p edominan ly in ends o educa e,
imp o e o e o m online (mis)beha io . In he ollowing,
we apply hese dimensions o in eg a e he di e en o ms
o shadowbans in o he b oade con ex o o he con en
mode a ion mechanisms as iden i ied and desc ibed by
Goldman (2021) (Table 2).
F om a philosophical s andpoin , shadowbanning is
pa icula ly opaque compa ed o nu u ing mechanisms.
Use s a e usually no in o med ha (o o how long) hey
a e shadowbanned o need o e ie e he in o ma ion
hemsel es when possible (Sil a 2022). Indeed, a majo
eason o using shadowbans is o disallow malicious ac o s
(e.g., spam bo s) o lea n om and adjus o con en
mode a ion algo i hms (Biddle e al. 2020). This p e en s
use s om lea ning om he shadowbanning decisions and
is hus a mo e puni i e measu e a he han one ha nu -
u es use beha io in line wi h communi y s anda ds.
The di e en o ms o shadowbans a y in hei deg ee
o se e i y (Jaidka e al. 2023). Shadowbans can include
bo h ha d and so ypes o con en mode a ion depending
Con en Mode a ion, Rules En o cemen , and Appeals
De ining Policies
Communi y Managemen
Con igu a ion
(P oduc Suppo ,
Enginee ing)
Con en Mode a ion Algo i hm
(De ec ion, Enac men )
Con olle
(Social Media Pla o ms,
T us & Sa e y Teams)
Use Communi y
In o ma e Au oma e
igu e inspi ed by Benlian e al. (2022)
Shadowban
(Ghos , Sea ch, Sea ch
Sugges ion, Down ie ing)
Fig. 1 Concep ualizing shadowbanning as a o m o algo i hmic con ol
123
820 M. Risius, K. M. Blasiak: Shadowbanning, Bus In Sys Eng 66(6):817–829 (2024)

on he deg ee o which hey en o ce isibili y es ic ions
(i.e., do no ecommend a all/ as much/ o some) (Gillespie
2022a). A ghos ban ma ches an accoun suspension
wi hou no i ying he use (Goldman 2021). Simila ly,
shadowbans ha se e ely es ic isibili y in si ua ions
wi h a sho ‘‘li espan’’ o engagemen equa e o a sus-
pension (Jaidka e al. 2023) (e.g., emo al om he in e nal
sea ch index). Ins ead o making con en inaccessible,
shadowbans can also g ea ly conceal i ems o make i less
likely ha use s may ind hem (Jaidka e al. 2021;
Nicholas 2022; Ryan e al. 2020) (e.g., downg ade in e nal
sea ch isibili y, no au o-sugges in he sea ch unc ion, no
o educed in e nal p omo ion, no o educed exposu e in
na iga ion links such as ‘‘mos popula ’’ o ‘‘newly a ail-
able’’). O e all, we conside shadowbanning o be a p e-
dominan ly puni i e mechanism wi h i s speci ic o ms
di e ing in hei deg ee o se e i y be ween ha d and so
o ms o con en mode a ion.
2.3 Unpacking he Con o e sy Su ounding
Shadowbanning
The deba e a ound he p e alence and impac o shadow-
banning is highly poli icized. On one side, social media
pla o ms a e eluc an o admi o shadowbanning (Gille-
spie 2022a), o en aming i as a speci ic kind o beha io
(e.g., ‘‘i doesn’ hide people’s con en o pos ing oo many
hash ags’’) (Co e 2021, p. 1234) o e e encing i s
a o emen ioned o iginal meaning in in e ne o ums
(Gillespie 2022b; Sa olainen 2022). The absence o a
uni ied, indus y-accep ed de ini ion o shadowbanning
(Gillespie 2022a) complica es discussions and uels
misunde s andings (Elmimouni e al. 2024). Pla o ms a e
unde s andably wa y o being sc u inized o hei policies
(ei he o being in e en ionis and biased, o opaque and
unaccoun able) and aim o a oid he poli iciza ion o hei
con en mode a ion p ac ices (Gillespie 2022a).
Table 2 Concep ualiza ion o shadowbanning among o he con en mode a ion mechanisms
Philosophy
Nu u ing Punishing
Deg ee o Se e i y Ha d Assign s ikes/wa nings Fine au ho /impose liquida ed damages Ou ing/unmasking
Communi y se ice Pu use /con en on indus y-wide blocklis
Educa e use s Remo e con en
Redi ec me hod

Repo o law en o cemen
Res o a i e jus ice/apology Suspend accoun
Suspend con en
Te mina e accoun
Ghos ban

Remo e om in e nal sea ch index*
So Age-ga e Disable commen s
Coun e speech Edi / edac con en
Display con en only o logged-in eade s Fo ei acc ued ea nings
In e s i ial wa ning No ollow au ho s’ links
Shaming Ou ing/unmasking
Suspend u u e ea nings Reduce se ice le els
Te mina e u u e ea nings Reduced i ali y
Wa ning legend Reloca e con en
Remo e c edibili y badges
Remo e om ex e nal sea ch index
Suspend pos ing igh s
Downg ade in e nal sea ch isibili y***
No au o-sugges **
No/ educed in e nal p omo ion***
No/ educed na iga ion links***
Types o shadowbans:
*
sea ch ban,
**
sea ch sugges ion ban,
***
down ie ing; adap a ions om Goldman (2021):

o iginally e e ed o as
‘‘shadowban’’,

adop ed om Sc i ens and Gaude e (2024); in non- emedial con ex s (e.g., emo e om ex e nal sea ch index, no ollow
au ho s’ links) (Goldman 2021) he classi ica ion o he punishing philosophy occu s based on he mechanisms’ eac i e na u e
123
M. Risius, K. M. Blasiak: Shadowbanning, Bus In Sys Eng 66(6):817–829 (2024) 821
F om he pla o ms’ pe spec i e, shadowbanning se es
as an e ec i e ool o manage p oblema ic con en . Fi s , i
helps mi iga e ‘‘law ul bu aw ul’’ con en . This con en
ha ades unde di e en names ac oss pla o ms (e.g.,
bo de line, sensi i e, ha m ul, undesi able, o objec ionable
con en ) (Co e 2021) is conside ed de imen al o he use
expe ience, h ea ens he heal h o he communi y, is
misleading o salacious. I ails o iola e pla o m policies
(Gillespie 2022b) and he e o e e ades o he o ms o
con en emo al. Examples include p o-ea ing diso de ),
sexually sugges i e, i ea m, and implici ly d ug- ela ed
con en (Gillespie 2022a) o misin o ma ion (Gillespie
2022a). Second, shadowbanning allows pla o ms o hide
mode a ion ac ics om malicious ac o s, such as spam-
me s (e.g., clickbai , links o malicious o decep i e si es)
(Co e 2021), bo s (e.g., as o u ing, sockpuppe ing), o
o ches a ed disin o ma ion campaigns, p e en ing hem
om adjus ing hei s a egies (Llewellyn e al. 2019;
Nicholas 2022). Thi d, i p o ides a means o a oid public
ou c ies o e ‘‘censo ship’’ while main aining lexible
mode a ion, pa icula ly in esponse o e ol ing h ea s like
inc easing adicaliza ion (Gillespie 2022a).
On he o he side, he lack o anspa ency a ound
shadowbanning has ueled olk heo ies and specula ion
among use s, policymake s, and wa chdogs (Jaidka e al.
2023) (e.g., Elon Musk uses shadowbanning on X o sup-
p ess Tesla’s employee union accoun (Masnick 2023)).
Repo s, especially om ma ginalized communi ies, sug-
ges shadowbanning is used o supp ess ce ain oices
(Elmimouni e al. 2024). Use su eys and in e iews on
shadowbanning a e o en disca ded h ough ‘‘black box
gasligh ing’’ by asse ing ha use s do no ha e su icien
unde s anding o con en cu a ion algo i hms o de e mine
whe he hey we e shadowbanned (Co e 2021). Pla o ms
explain media epo s on shadowbanning as echnical
gli ches, he use s’ ailu e o c ea e engaging con en , o as
a ma e o chance h ough he pla o m’s black-box
algo i hms (Co e 2021). This back-and- o h e odes
public us in social media pla o ms and con ibu es o
socie al pola iza ion (Chen and Zaman 2024; Jaidka e al.
2023).
The sc u iny su ounding shadowbanning also s ems
om i s po en ial o misuse. Cases ha e eme ged whe e
shadowbanning was allegedly used o mee go e nmen
demands (e.g., silence dissen in China, con ain Co id-19
messages in he USA) (AP 2024; He n 2019), ma ginalize
mino i y oices (e.g., handicapped, black, o LGBTQ ?
use s) (Delmonaco e al. 2024; Du y and Meisne 2022),
o manipula e public opinion (e.g., mu e p o-Pales inian
pos s) (Chen and Zaman 2024; Elmimouni e al. 2024; Luu
2023). This o m o exclusion can esul in emo ional
dis ess (Lu z and Schneide 2021), inancial ha m o
con en c ea o s (Du y and Meisne 2022), and inc eased
ulne abili y o online a acks ( o an o e iew, see Del-
monaco e al. 2024; Nicholas 2022). Fo ma ginalized
g oups, shadowbanning can lead o hei exclusion om
public discou se o bans by associa ion (Delmonaco e al.
2024; Elmimouni e al. 2024). On a b oade socie al le el,
he sec ecy and lack o ecou se os e dis us and con-
spi acy heo ies, while en enching socie al di ides (Chen
and Zaman 2024; Nicholas 2022).
2.4 Es ablishing he P e alence o Shadowbanning
The opaque na u e o shadowbanning has inspi ed con-
side able wo k on es ablishing he p e alence o shadow-
banning. A majo sou ce o e idence a e social media
pla o ms hemsel es when epo ing con en mode a ion
mechanisms ha impose a o m o isibili y educ ion
while a oiding he e m ‘‘shadowbanning’’ (Me e e al.
2021). Reddi is he only pla o m ha openly con i ms he
adi ional o m o shadowbans (Nicholas 2022). Elon
Musk p omised o p o ide mo e anspa ency on shadow-
banning a e aking o e Twi e (now X), which is ye o
be deli e ed (Pe ez 2023) and ins ead aces sc u iny him-
sel o e shadowbans on pos s wi h links o o he social
media pla o ms (New on 2023). Me a’s Facebook and
Ins ag am we e he i s majo social media pla o ms ha
decla ed pu suing ways o algo i hmically educe use
engagemen wi h bo de line con en in May 2018 (Gille-
spie 2022a; Zucke be g 2021). YouTube, X, LinkedIn, and
TikTok ha e since disclosed applying simila s a egies o
dealing wi h sensi i e con en (Du y and Meisne 2022)
and Ube epo s isibili y educ ions o es au an s in
esponse o un a o able cus ome e iews (Ube 2023) ( o
mo e de ails on pla o m s a emen s on shadowbanning, see
Delmonaco e al. 2024; Gillespie 2022a).
Beyond o icial social media pla o m s a emen s con-
ce ning isibili y educ ion mechanisms, addi ional e i-
dence sugges s he exis ence and p e alence o
shadowbanning. Anecdo al e idence comes om ace
e hnog aphy, su eys o in e iews (Nicholas 2022; W igh
2022), analyses o pla o ms’ con en mode a ion pa en s
(Nicholas 2023), and use s who closely moni o hei
engagemen s a is ics (e.g., con en c ea o s) (Du y and
Meisne 2022; Zeng and Kaye 2022). O he e idence
comes om sou ces ha a e di icul o e i y such as
in e nal whis le-blowe s (Chen 2019), in o ma ion leaks
(Gillespie 2022a), o in es iga i e jou nalism (Col e 2018;
Me lan 2020). To de e mine he ex en o shadowbanning,
a ecen su ey o 1,006 social media use s ound ha 9.2%
epo ha ing been shadowbanned. O hese 8.1% we e on
Facebook, 4.1% on Twi e (now X), 3.8% on Ins ag am,
3.2% on TikTok, 1.3% on Disco d, 1% on Tumbl , and less
han 1% on YouTube, Twi ch, Reddi , Nex Doo , Pin e es ,
Snapcha and LinkedIn (Nicholas 2022). A con en
123
822 M. Risius, K. M. Blasiak: Shadowbanning, Bus In Sys Eng 66(6):817–829 (2024)
mode a ion su ey ha o e sampled ma ginalized iden i-
ies (i.e., acial and e hnic mino i ies, LGBTQ ?people,
ans and/o nonbina y people) e en ound ha 21.78% o
esponden s epo ed shadowbanning (Delmonaco e al.
2024).
Recen ly, esea che s ha e begun collec ing s a is ical
e idence o he p esence and ex en o di e en o ms o
shadowbanning. Analyses o 41 k o o e 2.5 million
Twi e (now X) accoun s ound ha be ween 3–6.2% o
accoun s had been shadowbanned a leas once (Jaidka
e al. 2021; Me e e al. 2021). These s udies iden i y
cha ac e is ics ha inc ease he likelihood o shadowbans.
Some o hese p edic o s a e pa icula ly new accoun s
(less han wo weeks old) wi h low ollowe numbe s
(below 200), using inci ili y (nega i e o o ensi e e ms)
o pos ing pic u es wi hou ex messages, and displaying
bo -like beha io (high bo ome e sco e) (Jaidka e al.
2023; Me e e al. 2021). A e i ied accoun (e.g., he blue
checkma k on X) helps o d as ically educe he chance o a
shadowban (Jaidka e al. 2023; Jo genson 2022). Based on
hese indings, compu e scien is s we e hen able o
de elop ools o wo ka ounds ha iden i y whe he
accoun s a e shadowbanned o di e en pla o ms like X
(hisubway,
1
yuzu isa
2
) o Reddi ( /Com-
men Remo alChecke , /ShadowBan) (Nicholas 2022).
3 Challenges and Oppo uni ies o IS Resea ch
Resea ch on shadowbanning mechanisms and hei impli-
ca ions emains in i s ea ly s ages. While legal schola s,
communica ion esea che s, and compu e scien is s ha e
con ibu ed o his nascen ield, exis ing s udies p ima ily
ely on quali a i e me hods. These me hods o en in ol e
in e iews wi h a ec ed con en c ea o s o u ilize anony-
mous sou ces wi hin social media companies mechanism
(Co e 2021; Sa olainen 2022; Zeng and Kaye 2022)o
he analysis o con en mode a ion pa en s (Nicholas 2023).
In o ma ion sys ems schola s can add ess shadowbanning
o help esol e hese appa en issues and b idge he con-
e sa ion be ween social media pla o ms, use s, and eg-
ula o s. Applying a socio echnical pe spec i e, in o ma ion
sys ems esea ch can help mi iga e he human and socie al
implica ions o his o m o algo i hmic con en mode a ion
and acili a e p oduc i e discou se on his challenge
be ween he s akeholde s (i.e., businesses, go e nmen s,
NGO’s) (Go wa 2022). Building upon he concep ualiza-
ion o shadowbanning as an algo i hmic con ol p ocess
in ol ing a ious en i ies (use communi y, social media
pla o ms’ T us & Sa e y Teams, con en mode a ion
algo i hms; Fig. 1), he ollowing sec ions will explo e a
non-exhaus i e lis o illus a i e esea ch ques ions
(Table 3).
3.1 Use Communi y
While he e is cu en ly li le anyone can do i hei accoun
o con en is shadowbanned, esea che s iden i ied com-
mon s a egies ha c ea o s use o p e en and cope wi h
shadowbans: Supp ession, expe imen a ion, ci cum en ion,
and esigna ion (Du y and Meisne 2022). Simila ly, c e-
a o s’ expe iences helped o de elop echniques o indi-
idual use s o de ec whe he hey we e shadowbanned
(Columb es 2023). Howe e , use epo s o shadowban-
ning equen ly encoun e ‘‘black box gasligh ing,’’ a
phenomenon whe e pla o ms dismiss hese conce ns by
claiming use s misunde s and he complexi ies o con en
mode a ion algo i hms (Co e 2021). IS esea ch can
explo e me hods o use s o sa ely communica e po en ial
shadowbanning expe iences (1.1)? Simila ly, shadowban-
ning has signi ican implica ions o he a ge s (Mye s
Wes 2018; Nicholas 2022). Co e Sel -E alua ion heo y
(Bono and Judge 2003; Judge e al. 2003), o example, can
help us unde s and (and mi iga e) he shadowbanning
e ec s on use s’ sel -es eem, sel -e icacy, locus o con ol,
and emo ional s abili y (1.2). Las ly, shadowbanning has
conside able mone iza ion ami ica ions o c ea o s (Du y
and Meisne 2022; Mye s Wes 2018). Sel -de e mina ion
heo y (Ryan and Deci 2000) allows o examine he com-
plexi ies o he algo i hmically d i en wo kplace a he
indi idual le el (Benlian e al. 2022). He e i can help us
unde s and how shadowbanning a ec s con en c ea o s o
egula use s’ (e.g., job candida e) pe cei ed compe ence,
au onomy, and ela edness a e being shadowbanned (1.3).
Shadowbanning also poses a majo challenge o
ma ginalized g oups ha appea o be disp opo ionally
a ge ed (e.g., ‘‘ugly, poo , o disabled’’ use s, ‘‘Black
Li es Ma e ’’ il e , LGBTQ ?keywo ds) (Du y and
Meisne 2022; Ryan e al. 2020; Walsh 2022). The DIME
model (Louis e al. 2020) allows IS schola s o explo e he
esponses by hese mo emen s o shadowbanning (1.4).
3.2 Social Media Pla o ms (T us &Sa e y Teams)
Shadowbanning helps pla o ms ul il impo an socie al
asks by mode a ing a ious o ms o ha m ul con en .
While i s sec ecy is s ongly con es ed (e.g., p omo e
conspi acies, educe us ) (Nicholas 2022), shadowban-
ning’s opaque na u e helps o a oid poli icizing con en
mode a ion (Gillespie 2022a). The b oade implica ions o
con en mode a ion mechanisms a e con o e sially dis-
cussed among law and policy schola s (e.g., Douek (2022;
Goldman (2021); Go wa e al. (2020)), in communica ions
1
h ps://hisubway.online/shadowban/.
2
h ps://shadowban.yuzu isa.com.
123
M. Risius, K. M. Blasiak: Shadowbanning, Bus In Sys Eng 66(6):817–829 (2024) 823
esea ch (e.g., Suzo e al. (2019)), and compu e scien is s
[e.g., See ing (2020), Jha e e al. (2019a), Jha e e al.
(2019b)]. IS esea ch should join he deba e and explo e
shadowbanning’s abili y o mi iga e socie al o g oup
pola iza ion (Suns ein 2018) o p o ec ee speech (Rieme
and Pe e 2021) compa ed o o he o ms o con en
mode a ion (2.1, 2.2). Al e na i ely, esea ch demons a es
ha a e i ied accoun and blue checkma ks a e p o ec i e
ac o s agains shadowbanning (Jaidka e al. 2023; Jo -
genson 2022), which a e now pa ly o e ed o sale (Kla
2023). IS esea che s could add ess his issue, o example,
by applying he c i ical heo y o powe and e hics by
Foucaul (2007) o explo e whe he shadowbanning is an
e hical, poli ical, o p ima ily a echnical p oblem (Siape a
and Viejo-O e o 2021). Resea che s could also aim o
delinea e he in ended e sus unin ended consequences o
sys ema ic biases wi hin shadowbanning (2.3).
Mode a ing p oblema ic social media con en is a majo
conce n o ad e ise s ha wan o a oid ha ing hei ads
displayed nex o ha e ul con en (Cooban 2023). While
shadowbanning con ines he isibili y o p oblema ic con-
en , i emains po en ially accessible online. IS schola s
can apply, o example, b and sa e y (Bishop 2021)o
si ua ional c isis communica ion heo y concep s (Coombs
2007) o assess ad e ise s’ conce ns ega ding pla o ms
shadowbanning (2.4). We highligh ed he hea ed poli i-
cized deba e a ound he e m shadowbanning (Co e 2021;
Gillespie 2022a). The good easons o pe o ming shad-
owbans a e o en me wi h ha sh c i icism o censo ship
and ma ginaliza ion. Resea che s could ollow o he
examples o IS esea ch on con es ed online beha io s
(e.g., doxing) (F anz and Tha che 2023) o de e mine
bounda y condi ions o e hically accep able shadowban-
ning (2.5) and whe he some pla o ms’ e o s ha allow
use s o ace hei accoun s a us means meaning ul
imp o emen s (2.6) (Sil a 2022; Zakha chenko 2024). In
his ega d, while we ha e summa ized he e idence on he
p e alence o shadowbanning on social media pla o ms,
o he pla o ms (e.g., Gig-economy) also epo o ms o
isibili y educ ion (Ube 2023). Shadowbanning o e s
esea ch he oppo uni y o expand hei concep ualiza ion
o algo i hmic con ol unc ions (Alizadeh e al. 2023;
Hi sch e al. 2023; Kellogg e al. 2020; Sulli an e al.
2024), e ine algo i hmic managemen o bene i socie y a
la ge (Mo
¨hlmannn e al. 2023), and explo e he p e alence
Table 3 Sugges ed a eas o esea ch o ad ance knowledge on shadowbanning
Topic P oposed esea ch ques ions
Use communi y 1.1 How can people sa ely communica e ha hey we e shadowbanned?
1.2 How does shadowbanning a ec use s’ sel -es eem, sel -e icacy, locus o con ol and emo ional
s abili y?
1.3 Wha e ec s has shadowbanning on use s’ pe cei ed compe ence, au onomy, and ela edness?
1.4 How do ma ginalized g oups espond o being shadowbanned?
1.5 How does shadowbanning in luence use s’ online beha io , such as sel -censo ship o a oidance o
ce ain pla o ms?
Social media pla o ms (T us &
Sa e y Teams)
2.1 How do shadowbanning’s ade-o s compa e o o he o ms o con en mode a ion?
2.2 How e ec i e is shadowbanning in educing pola iza ion and p o ec ing ee speech compa ed o
o he o ms o con en mode a ion?
2.3 Wha ac o s d i e pla o m’s exe cise o shadowbanning?
2.4 Wha a e he b and sa e y implica ions o shadowbanning o ad e ise s?
2.5 Wha cha ac e izes e hically accep able shadowbanning cases?
2.6 How o design shadowbanning epo ing s uc u es ha o e easonable amoun s o anspa ency?
2.7 Wha o he ypes o pla o ms use shadowbanning o mode a e con en ?
Con en mode a ion algo i hm 3.1 How can shadowbanning be de ec ed ac oss pla o ms?
3.2 How can ad ancemen s in a i icial in elligence and machine lea ning imp o e he accu acy and
ai ness o shadowbanning algo i hms?
3.3 How can we implemen algo i hmic sensemaking in o shadowbanning decisions o educe algo i hm
a e sion?
3.4 Wha a e he implica ions o shadowbanning’s educ ionis e ec s o ee speech and algo i hmic
audiencing?
3.5 Wha si ua ions p omp manual s au oma ed shadowbanning?
3.6 Wha a e he ade-o s ega ding e icacy and ami ica ions be ween di e en ypes o shadowbans?
123
824 M. Risius, K. M. Blasiak: Shadowbanning, Bus In Sys Eng 66(6):817–829 (2024)