Ca ah, Nicholas e al.
A icle
Obse ing " uned" ad e ising on digi al pla o ms
In e ne Policy Re iew
P o ided in Coope a ion wi h:
Alexande on Humbold Ins i u e o In e ne and Socie y (HIIG), Be lin
Sugges ed Ci a ion: Ca ah, Nicholas e al. (2024) : Obse ing " uned" ad e ising on digi al pla o ms,
In e ne Policy Re iew, ISSN 2197-6775, Alexande on Humbold Ins i u e o In e ne and Socie y,
Be lin, Vol. 13, Iss. 2, pp. 1-26,
h ps://doi.o g/10.14763/2024.2.1779
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/300741
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 ps://c ea i ecommons.o g/licenses/by/3.0/de/legalcode
Volume 13 |
Obse ing “ uned” ad e ising on digi al
pla o ms
Nicholas Ca ah Uni e si y o Queensland
Lau en Hayden Uni e si y o Queensland
Ma ia-Gemma B own Uni e si y o Queensland
Daniel Angus Queensland Uni e si y o Technology
Aimee B ownbill Founda ion o Alcohol Resea ch and Educa ion
Kiah Hawke Uni e si y o Queensland
Xue Ying Tan Queensland Uni e si y o Technology
Amy Dobson Cu in Uni e si y
B ady Roba ds Monash Uni e si y
DOI: h ps://doi.o g/10.14763/2024.2.1779
Published: 26 June 2024
Recei ed: 19 Oc obe 2023 Accep ed: 21 Feb ua y 2024
Funding: This esea ch was unded by Aus alian Resea ch Council (ARC) Linkage
P ojec (LP190101051) and conduc ed in collabo a ion wi h he ARC Cen e o
Excellence in Au oma ed Decision-Making and Socie y (CE200100005).
Compe ing In e es s: The au ho has decla ed ha no compe ing in e es s exis ha
ha e in luenced he ex .
Licence: This is an open-access a icle dis ibu ed unde he e ms o he C ea i e
Commons A ibu ion 3.0 License (Ge many) which pe mi s un es ic ed use,
dis ibu ion, and ep oduc ion in any medium, p o ided he o iginal wo k is p ope ly
ci ed. h ps://c ea i ecommons.o g/licenses/by/3.0/de/deed.en
Copy igh emains wi h he au ho (s).
Ci a ion: Ca ah, N. & Hayden, L. & B own, M.-G. & Angus, D. & B ownbill, A. & Hawke ,
K. & Tan, X. Y. & Dobson, A. & Roba ds, B. (2024). Obse ing “ uned” ad e ising on
digi al pla o ms. In e ne Policy Re iew, 13(2). h ps://doi.o g/10.14763/2024.2.1779
Keywo ds: Digi al pla o ms, Ad e ising, Obse abili y, Algo i hms, Ha m ul indus ies
Issue 2
Abs ac : The hype - a ge ed ad e ising ha eme ged on digi al pla o ms o e he pas wo
decades is now mo e p oduc i ely unde s ood as uned ad e ising, a dynamic and un olding
p ocess whe e ads a e con inuously algo i hmically “op imised” o use s in eal ime. Following
Riede and Ho mann (2020), we aim o de elop a amewo k o he “condi ions o he p ac ice o
obse ing” algo i hmically- uned digi al ad e ising. We d aw on ou esea ch ac oss he Aus alian
Ad Obse a o y and a mul i-yea esea ch p ojec on digi al alcohol ad e ising. Ac oss hese
p ojec s we build cus omised ools o collec ads om pla o m ad lib a ies and h ough da a
dona ion om ci izen scien is s. We a gue ha he powe o digi al ad e ising is inc easingly
loca ed in i s capaci y o une. Pla o ms’ ad anspa ency ools d aw ou a en ion o ads, bu we
need o de elop he capaci y o obse e he dynamic socio- echnical p ocess o uning. We
concep ualise and p esen isualisa ions o “ uned sequences” o ads, as an al e na i e o “lib a ies”
o ads. We a gue ha de eloping he capaci y o obse e hese uned sequences be e a icula es
he mode o obse a ion equi ed o de elop he o ms o public unde s anding and accoun abili y
bo h ci il socie y o ganisa ions and esea che s a e looking o .
This pape is pa o Loca ing and heo ising pla o m powe , a special issue o In e ne
Policy Re iew gues -edi ed by Da id Niebo g, Thomas Poell, Robyn Caplan and José an
Dijck.
In oduc ion
The publicness o ad e ising in mass and b oadcas media was a de ini i e cha -
ac e is ic o i s o m and he basis o accoun abili y and egula o y amewo ks.
The pa adox o ad e ising on digi al pla o ms, howe e , is ha as i makes use s
mo e isible o i s algo i hmic ope a ions, i becomes less obse able o he pub-
lic. Digi al pla o ms like Me a and Alphabe de i e hei capaci y o enginee new
echnologies ha shape ma ke s and socie ies om ad e ising e enue. The pow-
e o hei ad e ising models es bo h in hei capaci y o ansla e social li e in o
da a ha ains algo i hmic models and he opaci y o hose models o public ob-
se abili y (C ain, 2021). The Aus alian Compe i ion and Consume Commission's
landma k Digi al Pla o m Se ices Inqui y (2020-2025) has highligh ed he ex a-
o dina y and la gely unchecked ma ke powe o Me a and Alphabe and hei s a-
us as an ad e ising duopoly (ACCC, 2020). Public conce n abou his ma ke pow-
e is compounded by ou limi ed abili y o bo h unde s and ou sha ed expe ience
o ad e ise - unded algo i hmic lows and o mi iga e he p oblems hey aise, in-
cluding a ge ing and disc imina ion, ci cula ing misleading messages, and shap-
ing social a i udes and iden i ies ( o ins ance, ela ing o body image, gende , and
ace) (Bu gess e al., 2022; Phan & Wa k, 2021; T o e al., 2021). In his a icle,
we ocus on he case o alcohol ad e ising on Me a pla o ms in Aus alia o
b oadly concep ualise he obse abili y o digi al ad e ising as a p econdi ion o
making i accoun able o use s, ci il socie y, and egula o s. Following Riede and
Ho mann (2020), we aim o de elop amewo ks ha c ea e he “condi ions o he
2 In e ne Policy Re iew 13(2) | 2024
p ac ice o obse ing” digi al ad e ising as a p ocess embedded in he comme -
cialisa ion and da a ica ion o ou cul u e.
I pla o ms’ ad models a e opaque, we mus de elop he combina ion o concep u-
al app oaches, echnical ools, and ins i u ional o ma ions equi ed o obse e
hem. We d aw on ou esea ch ac oss he Aus alian Ad Obse a o y (Bu gess e
al., 2022) and a mul i-yea esea ch p ojec on digi al alcohol ad e ising (Hawke
e al., 2022; Hayden e al., 2023). These p ojec s in ol e ex ensi e collabo a ion
among esea che s om compu e science, law, media, communica ion, and cul u -
al s udies disciplines o build cus omised ools o collec ads om pla o m ad li-
b a ies, unde ake da a dona ion wi h ci izen scien is s and engage in ex ended
collabo a ions wi h ci il socie y g oups. This ne wo k o collabo a ion is equi ed
o de elop bo h he echnical and concep ual in as uc u e o obse ing he au-
oma ed lows o digi al ad e ising. We begin by a guing ha digi al ad e ising
has shi ed owa d he logic o “ uning” o e “ a ge ing”, wi h signi ican implica-
ions o e o s o obse e digi al ad e ising. We hen ou line how ci il socie y
o ganisa ions in Aus alia ha e amed he p oblem o digi al ad e ising by ha m-
ul indus ies and p esen an audi o he ad anspa ency ools p o ided by digi al
pla o ms (Hawke e al., 2022). We ind ha none o he majo pla o ms we in-
es iga e — Me a (Facebook and Ins ag am), Alphabe (Google and YouTube), Tik-
Tok, Snapcha , o X ( o me ly known as Twi e ) — p oduce a du able public a chi e
o ads o con ex ual in o ma ion abou a ge ing, each, and spend. This unda-
men ally limi s he isibili y and anspa ency o ad e ising on hese pla o ms
and inhibi s public sc u iny, bu we also conside he implica ions o pla o m
anspa ency ools like ad lib a ies encou aging us o ocus ou analysis on he ol-
ume and con en o ads.
Ou analysis o ad anspa ency on digi al pla o ms and ou a emp s o build ap-
p oaches o obse ing ads lead us o conside whe he ad lib a ies add ess, exac-
e ba e, o ob usca e ou e o s o obse e and explain digi al ad e ising (Zal-
nie iu e, 2021). I we could see all he ads, wha would be e ealed? We ame ou
esponse by conside ing he p oblems o explainabili y, glu , and app oxima ion
gene a ed by associa i e algo i hmic ad e ising models. We a gue ha lib a ies
may se e pla o ms’ in e es s by leading bo h esea che s and ci il socie y owa d
obse ing he con en o ads, placing hem in he ole o moni o ing “bad” ads and
ad e ise s, a he han analysing he echnical and ins i u ional o ma ion o au-
oma ed ad e ising sys ems. Using ou collabo a ion be ween esea che s and
ci il socie y o ganisa ions conce ned abou “ha m ul” o ms o ad e ising, we a -
gue ha we need o de elop he concep ual and echnical amewo ks o obse -
3 Ca ah e al.
ing digi al ad e ising as a dynamic, e ol ing p ocess o “ uning” he ela ionships
be ween ad e ise s and consume s. We ad ance his a gumen by concep ualising
and p esen ing isualisa ions o “ uned sequences” o ads as an al e na i e o “li-
b a ies” o ads. Tuned sequences be e a icula e he p ocess we need o obse e
in o de o u he de elop he o ms o accoun abili y bo h ci il socie y o ganisa-
ions and esea che s a e looking o .
F om a ge ed o uned ad e ising
The hype - a ge ed ad e ising ha eme ged on digi al pla o ms o e he pas
wo decades can be p oduc i ely unde s ood as uned ad e ising. Ta ge ed ad e -
ising in ol es iden i ying and a ge ing a speci ic use on he basis o disc e e ca -
ego ies, such as gende , age, o e hnici y. Tuned ad e ising, on he o he hand, is a
dynamic and un olding p ocess whe e ads a e con inuously algo i hmically “ uned”
o “op imised” o use s in eal ime on he basis o an e e -changing and shi ing
se o a iables (B own e al., 2024). Va iables based on an opaque and p op ie a y
se o inpu s and algo i hmic weigh ings, om likes, commen s, and sha es o
dwelling on ce ain pos s, associa ions wi h o he con en , sea ch his o ies, ne -
wo k a ini ies, and no doub much mo e. I a ge ed ad e ising emphasises he
expe ience o ecei ing a pa icula ad a a pa icula momen , uned ad e ising
d aws a en ion o he media ed eeling o a con inuous empo al low o hy hm
o ads (B own e al., 2024; Ca mi, 2020). This low exceeds any speci ic ad, pos , o
piece o con en o p oduce a “mo e gene al mood” (Dean, 2010). In his espec ,
he concep o uning also helps o see b oade con inui ies be ween he algo i h-
mic low o digi al ad e ising and he low i s obse ed on ele ision by Ray-
mond Williams (B own e al., 2024; Ca mi, 2022; Lupinacci, 2020; Williams, 1974).
Digi al ad a ge ing and uning bo h ely on con inual op imisa ion o ad ex s in
he pu sui o eaching he desi ed audience. Howe e , uning accoun s o he
concu en e inemen o he ad sys em a la ge as well as he dynamic subjec e-
cei ing he ad. Op imisa ion, in he a ge ing sense, assumes he con inual im-
p o emen o ad ex s and da a ex ac ion o ma ch ideal ads o ideal audiences
(C ain & Nadle , 2019). In con as , uning aims o illus a e how he ela ions be-
ween ad ex s, he ad model, and use s shape each o he o e ime. While a ge -
ing assumes a s able subjec , uning implies he eme ging subjec which is ac i e
and lexible (B own e al., 2024). Tuning a ends o he p e-emp i e empo ali y o
digi al media, as no ye ully a icula ed, esponding in eal- ime, a ec i e, and
an icipa ing he u u e h ough he moni o ing, adjus ing, and modula ion o algo-
i hmic models (Amoo e, 2013; And eje ic, 2011; Buche , 2020; Coleman, 2018).
4 In e ne Policy Re iew 13(2) | 2024
“Wha is he nex ad in he sequence?” is he p oblem ha uned ad e ising
sol es o .
Piece by piece, o e he pas wo decades, uned ad e ising has eme ged as digi-
al pla o ms ha e au oma ed he collec ion o da a, a ge ing o ads, cons uc ion
o audiences, and c ea ion o con en . Ea ly digi al ad e ising models au oma ed
da a-collec ion and a ge ing based on p ecise c i e ia named by he ad e ise ,
such as gende , age, and geog aphy. F om 2013, digi al ad e ising models on
pla o ms like Facebook and Google became much mo e associa i e and p oba-
bilis ic, lea ning o op imise and une audiences o e ime. Fo example, Face-
book’s launch o “cus om” and "lookalike" audiences we e a key shi whe e ad e -
ise s uploaded da a abou exis ing audiences o gene a e new audiences. "Dy-
namic" ad-building ools also eme ged o e he pas decade ha au oma e he as-
sembly and a ge ing o ads. Ad e ise s upload "ad componen s" like images,
ideo, ex , and bu ons and he model lea ns wha combina ions o se e o pa -
icula consume s (Me a, n.d.). An ad e ise migh begin wi h in o ma ion abou
hei audience, basic c i e ia like age, gende and loca ion, o an exis ing cus ome
da abase. The ad model hen dynamically es s and e ines he ads, he audiences,
and he ela ionships be ween hem. Ra he han pay jus o imp essions, ad e -
ise s pay o digi al pla o ms o algo i hmically " une" an audience o ac in p e-
dic able and quan i iable ways: dwelling on con en , apping a bu on, making a
pu chase.
I we’ e begun o ge accus omed o algo i hmic models ha une he placemen
o ads, we need o now also an icipa e he eme gence o algo i hmic models ha
une he ads hemsel es. Al eady we a e seeing ads made om images ha a e
syn he ic, whe e he images and ex a e " uned" by gene a i e models. Au oma ed
ad-building ools will be in eg a ed in o pla o m models o gene a e syn he ic
digi al con en (Meh a, 2023; Vincen , 2023). Fo ins ance, local ba s will c ea e
syn he ic con en ha posi ions d inks in di e en in e io s, o ea u es di e en
kinds o people a he ba . Google ecen ly announced ha ad e ise s will be able
o supply c ea i e con en and a gene a i e algo i hmic model will “ emix” his
ma e ial o gene a e ads based on he audience i aims o each, as well as o he
goals such as sales a ge s (C iddle & Mu phy, 2023). Me a oo is signalling he in-
eg a ion o gene a i e AI in o i s ad building ools, oge he wi h a nex gene a-
ion o models ha would u n uning in o a medium whe e mul iple senso y in-
pu s — ex , sound, ision, mo emen , empe a u e — can gene a e ou pu s in mul-
iple media (Mis a e al., 2023). This will scale up o majo b ands unning highly-
uned c ea i e campaigns. In Ma ch 2023, Le i’s announced a pa ne ship wi h he
5 Ca ah e al.
digi al agency Lalaland.ai o use gene a i e AI o c ea e "di e se" syn he ic ashion
models. The Ve ge epo ed his would allow cus ome s o see clo hing on "body-
inclusi e" models, "spanning a wide ange o body ypes, ages, sizes, and skin
ones" (Wea he bed, 2023). A p esen a dynamic ad is compiled om a da abase o
ex s p e-loaded by he ad e ise , bu hese will inc easingly be supplemen ed by
a que y ha gene a es an ad "on demand". Ad e ising p o essionals will p oduce
b and e i o ies, campaigns, and con en , bu he inal execu ion will be " uned"
by au oma ed models ha e ine choice o igu es, skin ones, clo hes, se ings,
colou ways, ex , bu ons. Ad lib a ies s and o become obsole e as ads a e no
longe ixed o s able ex s ha can be a chi ed.
The powe o digi al ad e ising is loca ed in he shi o con ol o e ad e ising
in o he in as uc u e o digi al pla o ms like Me a and Alphabe . The capaci y o
une consolida es his powe because ad e ise s become mo e ins i u ionally de-
penden on pla o ms no only as a channel ha con ols access o consume a -
en ion, bu as an in as uc u e ha builds and op imises ads, audiences, and he
ela ion be ween hem, o e ime. This powe is u he concen a ed by pla o ms
as hey disin e media e media buying, ma ke esea ch, and da a analy ics by ei-
he assuming hese unc ions o making hem elian on he pla o m ecosys em
( an de Vlis & Helmond, 2021). This ma ke powe educes compe i ion and ac-
coun abili y among i ms, and i also diminishes he possibili y o obse ing he
ope a ions o pla o ms in shaping consume s’ li es. The powe o he au oma ed
ad e ising model o digi al pla o ms is con es ed by public in e es g oups wi h
longs anding conce ns abou ha m ul indus ies and digi al igh s. And, i is also
con es ed by ad e ise s seeking mo e obse abili y, accoun abili y, and con ol in
he audience and ad e ising p oduc s hey pay o . In Aus alia, his has been ob-
se ed mos clea ly in submissions om bo h ci il socie y and ad e ise g oups o
he Aus alian Compe i ion and Consume Commission’s Digi al Pla o m Se ices
Inqui y (ACCC, 2019, 2020).1
Ha m ul indus ies ma ke ing on digi al pla o ms in
Aus alia: The case o alcohol ma ke ing, ci il socie y,
and Aus alian policy consul a ions
In Aus alia, he use o digi al pla o ms by ha m ul indus ies has been aised in a
ange o policy consul a ions and legisla i e e o ms including he Digi al Pla o m
1. The ACCC’s Digi al Pla o m Se ices Inqui y began when he commission was di ec ed o in es i-
ga e he impac o online sea ch engines, social media and digi al pla o ms on compe i ion in me-
dia and ad e ising ma ke s. A inal epo was published in 2019 and hen a i e yea inqui y es-
ablished ha uns om 2020-2025 (see ACCC, n.d.)
6 In e ne Policy Re iew 13(2) | 2024
Se ices Inqui y (2020-2025), P i acy Ac Re iew (2023), and Basic Online Sa e y
Expec a ions in he Online Sa e y Ac (2022). The Aus alian eSa e y Commissione
has ecen ly no ed ha ha m ul ad e ising, including ad e ising ela ing o alco-
hol and gambling is in need o egula o y a en ion (eSa e y Commissione , 2022).
The issue o ha m ul digi al ma ke ing has simila ly been iden i ied in he UK as a
egula o y gap be ween online sa e y and p i acy and compe i ion and consume
p o ec ion egula ions. In esponse, he UK Go e nmen is conside ing speci ic
measu es o egula ing online ad e ising h ough hei Online Ad e ising P o-
g amme consul a ion (Depa men o Cul u e, Media and Spo , 2022). The Online
Ad e ising P og amme iden i ies a ull s a u o y app oach ha add esses all ac-
o s wi hin he digi al ma ke ing ecosys em as he mos likely s a egy o be e ec-
i e a inc easing anspa ency and accoun abili y o digi al ma ke ing, including
illegal and legal online con en ha can c ea e ha m o consume s and business-
es. I speci ically iden i ies ad e ising o alcohol, gambling, and unheal hy oods
as ha m ul ad e ising con en . The Wo ld Heal h O ganiza ion (2021) has called
o he "digi al ma ke ing ecosys em and global pla o ms" o be "mapped and un-
de s ood by policy-make s a local, na ional and in e na ional le els" (p. xii). They
ha e called on go e nmen s o "es ablish and und esea ch o moni o de elop-
men s" and a gued o he possibili y o an in e na ional con en ion on alcohol
ma ke ing, simila o he obacco con en ion. They ha e also poin ed o he Eu o-
pean Union Digi al Se ices Ac as a model o build on because i names he need
o con ol he ela ionship be ween he "manipula i e echniques" o digi al ma -
ke ing and nega i e impac s on public heal h and wellbeing.
Ha m ul indus ies’ ma ke e s, including alcohol co po a ions, a e inno a i e ac o s
who p esen a use ul case o explo ing he powe dynamics in he digi al pla -
o m’s ad e ising models mo e gene ally. As one example, he Aus alian alcohol
and gambling co po a ion Endea ou G oup ha e de eloped and con inue o in es
in hei Endea ou X ini ia i e, which uses algo i hmic models o d i e inc eased
sales ac oss online and in-s o e pla o ms (C ozie , 2021; Webe , 2022). These
models op imise he composi ion and a ge ing o ads, he p o ision o home de-
li e y, and he layou o s o es. Endea ou G oup has collec ed da a on 6.2 million
Aus alians, abou 1 in 3 adul s, h ough hei cus ome loyal y p og amme alone
(E ans, 2022). This da a can be in eg a ed wi h digi al pla o ms’ ad e ising ools
which a e uned o ind people wi h simila cha ac e is ics o alcohol companies'
mos p o i able exis ing cus ome s, people who ei he equen ly buy alcohol o
buy alcohol in la ge amoun s. These ools he e o e disp opo iona ely a ge peo-
ple mos a isk o ha m om hese p oduc s by design. These ha ms a e po en ial-
ly exace ba ed by he in eg a ion o alcohol ad e isemen s on digi al pla o ms
7 Ca ah e al.
wi h he online sale and deli e y o alcohol (Hayden e al., 2023). Two- hi ds o al-
cohol e aile ad e isemen s on Facebook in Aus alia con ain a ‘shop now’ bu -
on, p omp ing immedia e pu chase, collapsing he dis ance be ween he momen s
o seeing an ad and making a pu chase. We ea ha m ul indus ies ad e ising on
digi al pla o ms as bo h an issue o public conce n, bu also a o m o ad e ising
ha helps o highligh la ge ques ions abou obse abili y and accoun abili y in
he ad model.
Pla o ms de i e much o hei ma ke powe and capaci y o shape public a en-
ion and opinion om ad e ising e enue. Ad e ise s and pla o ms bo h claim o
“empowe ” consume s by se ing hem mo e ele an messages, a he same ime
hey ag ee ha he powe o digi al ad e ising is i s capaci y o “in e ac wi h
consume s a a highly in ima e le el, and con ol he communica ion and con-
sump ion en i onmen ” (Da mody & Zwick, 2020). Rega dless o whe he we a e
conce ned abou pa icula ad e ise s o he ad e ising model in gene al, we
need o be able o obse e how i ope a es a he le el o con olling ou en i on-
men and ea ing us as a “ agmen a y, co ela ional, p obabilis ic, en i onmen al-
ly-con ingen ” pla o m use (Golden ein & McGuigan, 2023).
This iew o pla o m powe is e lec ed in he calls om ci il socie y o ganisa-
ions ac oss digi al igh s, consume igh s, and public heal h ha ha e ecom-
mended ha digi al pla o ms should be accoun able o he au oma ed p edic-
ions and decisions hei ad e ising models make abou consume s (ACCC, 2019,
2020). They a gue ha pla o ms should be equi ed o p o ide he means o ob-
se e no jus ad e isemen s bu how he ad e ising model wo ks, including in-
o ma ion abou how people a e a ge ed, who ad e isemen s each, and how
much ad e ise s spend (Hayden e al., 2023). A he same ime hey a e acu ely
awa e ha policy ecommenda ions a ound a ge ing, such as p ohibi ing pa icu-
la kinds o a ge ing c i e ia, ha e become ou da ed as ad models mo e owa d
he associa i e logic o uning. They need o espond o his challenge by de elop-
ing he capaci y o unde s and and obse e he eme ging o m o digi al ad e is-
ing. While he case o ha m ul indus ies ad e ising is a speci ic si e whe e powe
is being con es ed by pla o ms, ad e ise s, ci il socie y, esea che s, and policy-
make s, i also p o ides a se ing o hinking p oduc i ely abou he powe o dig-
i al ad e ising mo e gene ally.
T anspa ency and obse abili y o ad e ising on
digi al pla o ms
Riede and Ho mann (2020) “p opose he concep o obse abili y as a mo e p ag-
8 In e ne Policy Re iew 13(2) | 2024
On i s own, a close eading o a single sequence does li le o explain he associa-
i e algo i hmic logic o he ad e ising model. While an indi idual use migh be
able o see hemsel es e lec ed in hei sequence o ads, hei sequence only be-
comes meaning ul when hey see i associa ed o app oxima ed wi h o he se-
quences. Fo esea che s, ci il socie y, and egula o s, ques ions abou he powe
o he ad e ising model and he ha ms i may cause equi e an obse a ion o se-
quences a scale. To espond o his p oblem, Figu e 2 uses ime-se ies modelling
o ex end analysis o bo h indi idual and g oups o ad sequences. Figu e 2 illus-
a es how alcohol ads, as a subse o all ads seen by a pa icipan , a e deli e ed as
pa o a sequence o p o ide an image o he hy hm o uning. In he case o al-
cohol ads, i helps us unde s and how he ad model adap s o di e en consume s
in he pa e n and in ensi y o ads i se es. Fu he mo e, wi h associa ed in o ma-
ion abou he ads and consume s ecei ing hem, we could unde s and he ela-
ionship be ween di e en kinds o uned sequences and ha m. We could ask, o
ins ance, whe he d inke s who consumed alcohol a high- isk le els we e se ed
pa icula kinds o sequences, whe he younge consume s we e se ed di e en
kinds o sequences and so on. The app oach we ha e aken is o index all ads e-
cei ed by ime and hen cumula i ely sum he numbe o alcohol and non-alcohol
ads agains hese indices. This enables us o isualise alcohol and non-alcohol ads
ecei ed o e ime as a line cha , whe e he x-axis ep esen s he numbe o non-
alcohol ads and he y-axis ep esen s he numbe o alcohol ads. The sequence
hen e eals pe iods whe e a signi ican numbe o alcohol ads a e deli e ed in a
sho space o ime ( e ical lines), e sus pe iods whe e ela i ely ew alcohol ads
a e seen (ho izon al lines).
Analysis
While i is beyond scope o p o ide a ull analysis, in Figu e 2 we isualise se-
quences o wo membe s (Pa icipan A and B) o he Aus alian Ad Obse a o y
coho o illus a e he alue in his app oach o compa ing sequences. In his
g aph we can i s ly no e how A is exposed o mo e ads o e ime. A ends o e-
cei e alcohol ads in e spe sed amongs hei s eam o non-alcohol ads wi h a
s eady, cumula i e egula i y. By compa ison, we no e wo s eep e ical segmen s
o B ea ly in hei ad dona ion jou ney, which co espond o pe iods o in ense and
epea ed exposu e o alcohol ad e ising. These di e en pa e ns be ween pa ic-
ipan s A and B e eal why i is impo an o de elop such a isual ime-se ies
analysis. The isualisa ion o nume ical dis ibu ions ha sha e nea -iden ical s a-
is ical p ope ies (such as mean and a iance), can be ma kedly di e en in e ms
o hei unde lying dis ibu ions. The impo ance o examining ealis ic da ase s
g aphically, no jus nume ically, has been an impo an ouchs one in he ield o
15 Ca ah e al.
s a is ics and da a science (Anscombe, 1973). Simple pe cen ages o coun s o al-
cohol ads hide he eali y o he sequencing o alcohol ad e ising which can be
highly egula ised (like ha o A), o punc ua ed by pe iods o in ense exposu e
(like ha o B). This plo ale s us o he di e en “ hy hms" o use s' sequences,
and we can specula e abou he di e en ways hese sequences migh be unde -
s ood as ha m ul o no . Fo esea che s, ci il socie y, and egula o s his would
lead us owa d modelling ypologies ha sequences c ea e, how hey a e associa -
ed wi h di e en kinds o use s, and wha ha ms hey may cause. Fo indi iduals, i
hey could see hei sequence ela i e o o he use s’ sequences, hen hey could
be e app ecia e how he model unes o hem and makes hei expe ience di -
e en o o he s. I migh also hen c ea e he condi ions o pla o ms o c ea e
con ols ha enable use s o shape hei expe ience less a he le el o an indi id-
ual pos and mo e a he le el o hei o e all sequence.
FIGURE 2: Compa ing wo use s’ sequences o alcohol and non-alcohol ads o e ime.
The modelling o uned sequences can also de elop in wo u he di ec ions:
• P obabilis ic p ojec ion o sequences: we can s a is ically model a
sequence o p edic u u e pa e ns. This could be done by using da a
abou all he sequences in an ad obse a o y o unde s and he olume
and pa e n o pa icula ads use s will see in o he u u e based on wha
hey ha e seen so a . This migh also enable he expe imen al modelling
o changes o algo i hmic models o c ea e di e en kinds o sequences. In
16 In e ne Policy Re iew 13(2) | 2024
he case o alcohol ad e ising, his would enable us o model he
di e ences be ween who “accumula es” alcohol ads and who ge s “bu s s”
o ads o e ime. Pla o ms and ad e ise s could be equi ed o p o ide
da a ha hen enables esea che s, ci il socie y, o egula o s o
unde s and he ela ionships be ween hese di e en kinds o sequences
and pa e ns o consump ion.
• Compa ison be ween sequence ypologies: we can model “ ypologies" o
sequences o examine s a is ically he o m hey ake and how hose o ms
a e associa ed wi h pa icula kinds o use s and use expe iences. Imagine
ha bo h sequence A and B a e each a composi e o many di e en use s.
We could hen s a o unde s and uning as no jus selec ion o ads bu
he c ea ion o di e en kinds o ad hy hms and eloci ies o di e en
kinds o use s. The c ea ion o ypologies could illumina e he unc ion o
p oxies by conside ing he a eas whe e g oups con e ge and di e ge.
F om ad lib a ies o ad obse a o ies
Ad lib a ies ca y well-es ablished isks wi h anspa ency: oo much in o ma ion
c ea es noise, he w ong kind o in o ma ion is a dis ac ion, he choice o wha in-
o ma ion is made public de ines he objec and p oblem in ad ance (Ananny &
C aw o d, 2018; Lee ssen e al., 2021; Riede & Ho mann, 2020; Zalnie iu e, 2021).
This is pa icula ly he case whe e “ he de ails o a sys em will no only be p o-
ec ed by co po a e sec ecy o indeciphe able o hose wi hou echnical skills, bu
insc u able e en o i s c ea o s because o he scale and speed o i s design” (An-
nany & C aw o d, 2018). Ad lib a ies a e “ anspa ency-washing” ini ia i es ha
ca e ully ob usca e and edi ec a en ion away om he p ocesses and echniques
ha make pla o ms powe ul (Zalnie iu e, 2021). Ad lib a ies ocus ou a en ion
on he ads, when we eally need o obse e he algo i hmically- uned sequences
ha make digi al ad e ising powe ul. Re lec ing on his challenge, we ame
h ee main concep ual p oblems wi h ad lib a ies, ega dless o how anspa en ,
comp ehensible, and accessible hey a e:
1. The p oblem o glu : lib a ies p esen in o ma ion ha is di icul o sea ch
and analyse and mo e in o ma ion would no enable be e o ms o
unde s anding.
2. The p oblem o explana ion: lib a ies p esen ads ha do no help o
obse e and explain he algo i hmically- uned ad e ising model, bu
mo e echnical explana ions o algo i hms would no help ei he .
3. The p oblem o app oxima ion: he associa i e and gene a i e algo i hmic
models ha d i e digi al ad e ising do no map on o s aigh - o wa d
symbolic and causal explana ions as o why a pe son saw a pa icula ad a
a pa icula momen .
17 Ca ah e al.
To espond o hese p oblems we p opose ha we need o de elop he capaci y o
obse e he dynamic social and echnical p ocess o uning h ough ad sequences.
The p oblem o glu
Ad lib a ies ep oduce he “pa adox o an e a o in o ma ion glu " whe e we hink
he mo e in o ma ion we ha e he mo e likely we a e o be ully in o med, ha
wi h enough da a eali y will speak o i sel , bu he da a glu becomes noise ha
ob usca es unde s anding (And eje ic, 2013). I we asked digi al pla o ms o
a chi e e e y ad e isemen hey e e published, we would end up wi h lib a ies
o ex ao dina y scale. They would be oo big o be made sense o a a human-
scale and would equi e signi ican esou ces o de elop and main ain au oma ed
echniques o o ganising, classi ying, and sense-making. Ou only ecou se would
be o au oma ed o ms o sense-making, likely de eloped and ope a ed by digi al
pla o ms hemsel es and o ganised a ound hei in e es s (And eje ic, 2013).
Powe o unde s and ad e ising would emain wi h ac o s ha ha e he capaci y
o p ocess, analyse, and isualise he da a. The exis ing ad lib a ies al eady hwa
obse abili y by c ea ing he condi ions o in o-glu : la ge collec ions o ads ha
a e no sea chable o in e p e able in ways ha enable use s o unde s and how
digi al ad e ising wo ks. Whe e anspa ency was once a esponse o a lack o in-
o ma ion, i now only bene i s hose wi h he capaci y o u ilise in o ma ion a
scale. Unde hese da a-sa u a ed condi ions, cu a ion eme ges as a c i ical issue
(Da is, 2020). We need o de elop o ms o sense-making ha enable us o unde -
s and how he ad e ising model’s “cu a o ial algo i hms” wo k o "encode use s’
implici desi es and p opensi ies based on indi idual and collec i e beha iou al
pa e ns” (Da is, 2020, p. 51). We need o ask algo i hms o gi e an accoun o
hei cu a o ial p ac ices as hey une lows o con en , cu a ing wha o show
when and whe e (Da is, 2020).
The p oblem o explana ion
On he ace o i , ad lib a ies con ibu e o he p oblem o explana ion because
hey p o ide pa ial in o ma ion: only wo pla o ms p o ide lib a ies, no lib a y
p o ides a comple e o accessible a chi e o ads, and no pla o m p o ides use ul
in o ma ion on a ge ing, each, o spend. A mo e undamen al p oblem is ha he
ad lib a ies shape he esea ch agenda on digi al ad e ising. The p oblem wi h in-
dus y-c ea ed anspa ency ools like lib a ies is ha he pla o m plays he de e -
mina i e ole in choosing wha is made anspa en , esea che s a e unable o e -
i y he eliabili y o he da a, and he capaci y o analyse he da a is complica ed
by he con igu a ion o pla o m dashboa ds and APIs (B uns, 2019; Meh a & E ick-
18 In e ne Policy Re iew 13(2) | 2024
son, 2022). Resea che s become me hodologically and concep ually o ien ed o-
wa d analysing he objec in he lib a y: he con en and numbe o ad e ise-
men s. In he pu sui o mo e comp ehensi e lib a ies, esea che s and ci il socie y
g oups lose oppo uni ies o obse e how ads a e in eg a ed in o use s’ eeds in an
unin e up ed way, how ads appea in sequences o e ime, associa e wi h o he
ads and con en , and espond o igge s om use s’ ac ions (Meh a & E ickson,
2022).
A he same ime, explana ions o he echnical ope a ion o “ he algo i hm” would
no explain he powe o he ad e ising model o une and op imise. Amoo e
(2020) a gues ha we need o mo e away om o ms o explana ion o ganised
a ound echnical documen a ion and owa d app oaches ha enable an unde -
s anding o he associa i e and gene a i e ope a ions o algo i hmic sys ems in
he social wo ld. One way o see his in p ac ice is o look a he analy ics dash-
boa ds and ad builde ools ha pla o ms p o ide o ad e ise s, which di e om
he a chi ec u e o he lib a y (Meh a & E ickson, 2022). Ra he han echnical ex-
plana ion o algo i hms, hese ools and dashboa ds p o ide dynamic isualisa-
ions ha enable ad e ise s o unde s and how hei ads and audiences a e being
uned in ways ha espond o hei s a egic choices, he ac ions o hei con-
sume s, and he pla o m ools hey a e using. Resea che s and ci il socie y also
need he ools o obse e he dynamic ela ions be ween ad e ise s, pla o ms,
and use s. This mode o obse a ion is pa icula ly impo an as au oma ed mod-
els mo e beyond s uc u ed decision-making models and owa d open-ended and
associa i e ones ha op imise and une sequences o con en . We a e ul ima ely
ying o unde s and no he echnical ope a ions o an algo i hmic model bu how
i ope a es in he wo ld.
The p oblem o app oxima ion
Ad lib a ies p o ide li le in o ma ion abou how ads a e a ge ed. The wo excep-
ions a e poli ical ads on Me a and Alphabe and, since 2023, ads on Me a pub-
lished in he EU. Bu e en hen, in o ma ion abou a ge ing shows only e y basic
demog aphic c i e ia like age ange, gene al loca ion, and gende ha ob usca es
how he ad model ope a es. Ta ge ing implies a di ec symbolic link be ween he
choice o “a a ge ” by an ad e ise and he deli e y o an ad. In a uned ad model,
ads a en’ a ge ed only, o e en a all, by he selec ion o pa icula c i e ia ha
co espond wi h aspec s o ou iden i y, bu ins ead by i e a i ely and p obabilis i-
cally associa ing us wi h o he use s based on ou “likeness”, “app oxima ion” o
sha ed pa e ns o exp ession, mo emen , swiping, apping, and so on (B own e
al., 2024). E en i an ad model es ic s ad e ise s om using ca ego ies ela ed
19 Ca ah e al.
o ace, gende , o age, i lea ns o a ge no based on who we a e bu who we a e
p oxima e o (Phan & Wa k, 2021). Phan and Wa k (2021), o ins ance, a gue ha
Facebook’s emo al o “e hnic ca ego ies'' does no elimina e acial classi ica ion
and disc imina ion. While hese ca ego ies no longe exis as a means o ad e is-
e s o a ge people, hey a e einsc ibed h ough ou p oximi y o each o he . To
unde s and how pla o ms’ ad models posi ion us in p oximi y o each o he , we
don’ need o “p y open” he algo i hmic black box, bu ins ead de elop he social
and echnical capaci y o obse e he end p oduc s and he abs ac ions ha algo-
i hmic sys ems p oduce (and ep oduce) in he wo ld (Phan & Wa k, 2021).
These h ee challenges illus a e he need o de elop app oaches ha begin wi h
asking “wha do we need o obse e?” a he han jumping o echniques o ob-
se ing “ he ads”, “ he a ge ing”, o “ he algo i hm”. By ocussing on obse abili y
we aise “ he complica ed ques ion o how da a and analy ical capaci ies should
be made a ailable, o whom and o wha pu pose” (Riede & Ho mann, 2020).
Th ee issues a e impo an : a i ing a a sha ed unde s anding o wha we need o
obse e, ha ing he echnical capaci y o obse e i , and ha ing he social and in-
s i u ional ela ionships o make p ac ices o obse a ion meaning ul. While we
should con inue o ad oca e o ad lib a ies ha mee minimum s anda ds o
a chi ing, access, sea chabili y, and analysis, we also need o ensu e ha we a e
no con ined o he way pla o ms de ine ad anspa ency. To eckon wi h he pow-
e o digi al ad e ising we need o imagine o ms o obse abili y ha begin no
wi h he ads bu wi h he socio- echnical p ocess o ad e ising. In his a icle we
p oposed he “ uned sequence” as an obse able objec o enable be e o ms o
public unde s anding and accoun abili y o uned ad e ising on digi al pla o ms.
We a gue his helps esol e he p oblems o glu , explana ion, and app oxima ion
by ocussing on he eme gen p ocess o op imising he ela ionships be ween ads,
use s, and he eeds in which hey a e assembled.
Ou app oach aims o econ igu e a glu o ine , indi idual ads in o a sequence
ha enables us o obse e he eme gen hy hm and pa e n o ads. Tuned se-
quences a e a cu a o ial echnique ha helps o na iga e and make sense o la ge
amoun s o ad con en in ways ha enable us o app ehend he algo i hmic logic
o he ad e ising model. In he case o he Ad Obse a o y, as one example, an in-
e collec ion o 700,000 ads becomes 1,800 sequences. Those sequences can be
u he modelled o loca e meaning ul ypologies and pa e ns. Impo an ly, his
means we can e isi he con en o ads as no jus symbolically powe ul bu also
echnically powe ul because o he way hey unc ion as da a poin s in a p ocess
o algo i hmic op imisa ion. The con en o ads bo h symbolically shapes ou iden-
20 In e ne Policy Re iew 13(2) | 2024
i ies, eelings, and beha iou s, and helps o ain models ha op imise audiences
and hei consump ion p ac ices.
This would enable us o be e explain he empo al and associa i e ope a ions o
he ad model. The abs ac ions and pa e ns ha algo i hms p oduce e eal he
logic o au oma ed sys ems wi hou ha ing o “open he hood” o he algo i hms
hemsel es (Amoo e, 2020; Phan & Wa k, 2021). The uned sequence is he mode
h ough which he algo i hmic ad model gi es an accoun o i sel . While uned se-
quences a e he igh place o begin ou obse a ion, hey also se he g ounds o
de eloping meaning ul accoun s o how hey a e p oduced — h ough wha echni-
cal models, wi h whose da a, wi hin wha kind o pla o m and comme cial ela-
ionships. They also p omp ques ions abou how he sequence ope a es in he
li es o use s o digi al media pla o ms: how is he sequence expe ienced? No
jus by looking bu also by apping and swiping? How does i shape ou desi es
and subjec i i y bu also o ganise ou p ac ices?
The uned sequence enables us o model how use s a e app oxima ed o associa ed
wi h each o he h ough a ange o eme gen ca ego ies, wi hou speci ying any
a ge ing c i e ia. Fo example, whe e do he “ ypologies" o sequences we p o-
posed con e ge and di e ge? A e hose di e ences no jus abou he kinds o ads,
bu he pa e ns o ads, hei in ensi y, eloci y, and hy hms as he model looks
no o s able a ge s bu o eme gen possibili ies? By isualising and analysing
uned sequences o ads, we ge close o app ehending he ou pu s o he algo i h-
mic model ha makes digi al ad e ising powe ul, a ma e o public conce n, and
po en ially ha m ul. In his a icle, we’ e demons a ed ou a emp o build a so-
cio- echnical in as uc u e o obse ing digi al ad e ising ha includes e-
sea che s om compu a ional, legal, ad e ising, and media disciplines wo king
oge he wi h ci il socie y g oups and public in e es jou nalis s. This in as uc-
u e o obse abili y is needed o loca e he powe o digi al pla o ms and build
mul i- ace ed o ms o public accoun abili y, and complemen s e o s o build in-
as uc u e o mapping ela ionships be ween pla o m and ad e ising indus y
ac o s ( an de Vlis & Helmond, 2021).
Conclusion
The ela ionship be ween digi al pla o ms and ad e ising is a key ec o o pow-
e in ou public cul u e. The ma ke powe o many digi al pla o ms is g ounded
in a dominance o ad e ising ma ke s ha enables in es men , no in o he p o-
duc ion o con en like mass media indus ies, bu in o he enginee ing o pla o m
in e aces, ha dwa e, and au oma ed models. This ela ionship is complica ed. In
21 Ca ah e al.
one sense ad e ise s a e beholden o pla o m owne ship and con ol o he au o-
ma ed models and uned audiences hey use, bu a he same ime pla o ms ha e
helped ad e ise s o c ea e mo e op imised and esponsi e ma ke o ma ions.
Pla o ms and ad e ise s ha e collabo a ed and compe ed in he c ea ion o a
o m o ad e ising ha is powe ul because o i s capaci y o une sequences ha
channel, modula e, and p e-emp in e es s, p e e ences, and choices wi hin a la g-
e sys em o consump ion ha shapes ou li es, wellbeing, heal h, and en i on-
men .
In ou collabo a ion be ween esea che s and ci il socie y we aim o build obse -
a o ies a he han lib a ies. Lib a ies se e pla o ms’ in e es s because hey ob-
usca e he algo i hmic- uning ha makes digi al ad e ising powe ul, de e mine
how ci il socie y and esea ch g oups app oach ad e ising, and hus in luence
policy p ocesses such as public inqui ies and he de elopmen o egula o y ap-
p oaches. Obse a o ies a e socio- echnical a angemen s h ough which we can
de elop he capaci y o moni o how lows o ads mo e and in e ela e, a he han
s o ing and a chi ing collec ions o ads o e ime. As pla o ms’ algo i hmic mod-
els become mo e eme gen in hei o de ing o ads and in hei capaci y o c ea e
ad con en , we need modes o obse a ion ha a e ocussed less on ca aloguing
he ads and mo e on unde s anding he ope a ions o algo i hmic models in ou
e e yday li es. The algo i hmic sequences o ads a e en angled wi h he indi idu-
als who sc oll h ough hem. These sequences demons a e a key cha ac e is ic o
eme gen and gene a i e algo i hmic models — hey become powe ul by ope a -
ing and lea ning om he social and cul u al se ings hey a e deployed in o.
Obse a o ies equi e la ge mul i-disciplina y eams ha include esea che s
ac oss media and communica ion, public heal h, public policy, law, and compu e
science, and collabo a ions wi h ci il socie y and egula o y s akeholde s. This e-
qui es in es men no jus in esea ch p ojec s bu also esea ch cen es and in a-
s uc u e o en in pa ne ship wi h ci il socie y, indus y, and go e nmen . While
obse a o ies help de elop be e o ms o unde s anding and con ol o indi id-
ual use s, hey mos impo an ly enable he de elopmen o public cul u es, ech-
niques, and amewo ks o obse ing digi al pla o ms. The goal isn’ o es ablish
comp ehensi e a chi es o lib a ies o con en , bu o be engaged in he ongoing
social and echnical wo k o making he op imisa ion and au oma ion o public
cul u e obse able, unde s andable, and accoun able o sha ed in e es s. The shi
o obse a o ies ocuses ou a en ion on he dynamic powe ela ionship be ween
pla o ms and ad e ise s, aiming o c ea e o ms o obse a ion ha enable
meaning ul unde s anding and accoun abili y ocussed no on “bad” ads wi h “bad”
22 In e ne Policy Re iew 13(2) | 2024
messages, o “bad” o ms o a ge ing, bu on he eme gence o a new cul u al
o m whe e powe is loca ed in he capaci y o une and shape ou pa e ns o con-
sump ion and ways o li e.
Re e ences
Amoo e, L. (2013). The poli ics o possibili y: Risk and secu i y beyond p obabili y. Duke Uni e si y
P ess. h ps://doi.o g/10.2307/j.c 11sms8s
Amoo e, L. (2020). Cloud e hics: Algo i hms and he a ibu es o ou sel es and o he s. Duke Uni e si y
P ess. h ps://doi.o g/10.1215/9781478009276
Ananny, M., & C aw o d, K. (2018). Seeing wi hou knowing: Limi a ions o he anspa ency ideal
and i s applica ion o algo i hmic accoun abili y. New Media & Socie y, 20(3), 973–989. h ps://doi.o
g/10.1177/1461444816676645
And eje ic, M. (2011). The wo k ha a ec i e economics does. Cul u al S udies, 25(4–5), 604–620. h
ps://doi.o g/10.1080/09502386.2011.600551
And eje ic, M. (2013). In oglu : How oo much in o ma ion is changing he way we hink and know.
Rou ledge. h ps://doi.o g/10.4324/9780203075319
Angus, D. (2024). Alcohol ad e ising sequences [da ase ]. Tableau Public. h ps://public. ableau.com/
sha ed/PJ4MWJ425
Angus, D., Obeid, A. K., Bu gess, J., Pa ke , C., And eje ic, M., Bagna a, J., Ca ah, N., Fo dyce, R.,
Hayden, L., Lewis, K., O’Neil, C., Alba án-To es, C., & Cella d, L. (2024). The Aus alian Ad
Obse a o y: Technical and da a epo (P ojec Repo 009; ADM+S Wo king Pape Se ies). ARC
Cen e o Excellence o Au oma ed Decision-Making and Socie y. h ps://doi.o g/10.60836/bz z- k0
8
Anscombe, F. J. (1973). G aphs in s a is ical analysis. The Ame ican S a is ician, 27(1), 17–21. h p
s://doi.o g/10.1080/00031305.1973.10478966
A o ney-Gene al’s Depa men . (2023). P i acy Ac Re iew (Repo 2022). Aus alian Go e nmen . h
ps://www.ag.go .au/ igh s-and-p o ec ions/publica ions/p i acy-ac - e iew- epo
Aus alian Compe i ion and Consume Commission. (2019). Digi al pla o ms inqui y [Final epo ]. h
ps://www.accc.go .au/abou -us/publica ions/digi al-pla o ms-inqui y- inal- epo
Aus alian Compe i ion and Consume Commission. (2020). Digi al pla o m se ices inqui y [In e im
epo ]. h ps://www.accc.go .au/abou -us/publica ions/se ial-publica ions/digi al-pla o m-se ice
s-inqui y-2020-25- epo s/digi al-pla o m-se ices-inqui y-sep embe -2020-in e im- epo
Aus alian Compe i ion and Consume Commission. (n.d.). Digi al pla o m se ices inqui y 2020-25. h
ps://www.accc.go .au/inqui ies-and-consul a ions/digi al-pla o m-se ices-inqui y-2020-25
B own, M.-G., Ca ah, N., Roba ds, B., Dobson, A., Rangiah, L., & De Lazza i, C. (2024). No a ge s, jus
ibes: Tuned ad e ising and he algo i hmic low o social media. Social Media + Socie y, 10(1). h p
s://doi.o g/10.1177/20563051241234691
B uns, A. (2019). A e he ‘APIcalypse’: Social media pla o ms and hei igh agains c i ical
23 Ca ah e al.
schola ly esea ch. In o ma ion, Communica ion & Socie y, 22(11), 1544–1566. h ps://doi.o g/10.108
0/1369118X.2019.1637447
Buche , T. (2020). The igh - ime web: Theo izing he kai ologic o algo i hmic media. New Media &
Socie y, 22(9), 1699–1714. h ps://doi.o g/10.1177/1461444820913560
Bu gess, J., And eje ic, M., Angus, D., & Obeid, A. K. (2022). Aus alian Ad Obse a o y: Backg ound
pape (P ojec Repo 004; ADM+S Wo king Pape Se ies). ARC Cen e o Excellence o Au oma ed
Decision-Making and Socie y. h ps://doi.o g/10.25916/7BGE-BP35
Ca mi, E. (2020). Rhy hmedia: A s udy o Facebook immune sys em. Theo y, Cul u e & Socie y, 37(5),
119–138. h ps://doi.o g/10.1177/0263276420917466
Coleman, R. (2018). Theo izing he p esen : Digi al media, p e-eme gence and in a-s uc u es o
eeling. Cul u al S udies, 32(4), 600–622. h ps://doi.o g/10.1080/09502386.2017.1413121
C ain, M. (2018). The limi s o anspa ency: Da a b oke s and commodi ica ion. New Media &
Socie y, 20(1), 88–104. h ps://doi.o g/10.1177/1461444816657096
C ain, M. (2021). P o i o e p i acy: How su eillance ad e ising conque ed he in e ne . Uni e si y o
Minneso a P ess. h ps://doi.o g/10.5749/9781452971667
C ain, M., & Nadle , A. (2019). Poli ical manipula ion and in e ne ad e ising in as uc u e. Jou nal
o In o ma ion Policy, 9, 370–410. h ps://doi.o g/10.5325/jin opoli.9.2019.0370
C iddle, C., & Mu phy, H. (2023, Ap il 30). Google o deploy gene a i e AI o c ea e sophis ica ed ad
campaigns. Financial Times. h ps://www. .com/con en /36d09d32-8735-466a-97a6-868d a34bdd5
C ozie , R. (2021, May 27). Endea ou G oup builds an AI-powe ed pe sonalisa ion engine. iTnews. h
ps://www.i news.com.au/news/endea ou -g oup-builds-an-ai-powe ed-pe sonalisa ion-engine-56
5124
Da mody, A., & Zwick, D. (2020). Manipula e o empowe : Hype - ele ance and he con adic ions o
ma ke ing in he age o su eillance capi alism. Big Da a & Socie y, 7(1). h ps://doi.o g/10.1177/20
53951720904112
Da is, J. L. (2020). Theo izing cu a ion. In D. A. Rohlinge & S. Sobie aj (Eds.), The Ox o d handbook
o digi al media sociology (pp. 40–60). Ox o d Uni e si y P ess. h ps://doi.o g/10.1093/ox o dhb/97
80197510636.013.5
Dean, J. (2010). A ec i e Ne wo ks. MediaT opes, 2(2), 19–44. h ps://media opes.com/index.php/M
edia opes/a icle/ iew/11932
Depa men o Cul u e, Media and Spo . (2022). Online Ad e ising P og amme consul a ion
[Consul a ion ou come]. UK Go e nmen . h ps://www.go .uk/go e nmen /consul a ions/online-ad
e ising-p og amme-consul a ion/online-ad e ising-p og amme-consul a ion
eSa e y Commissione . (2022). Recommende sys ems and algo i hms [Posi ion s a emen ]. Aus alian
Go e nmen . h ps://www.esa e y.go .au/indus y/ ech- ends-and-challenges/ ecommende -sys e
ms-and-algo i hms
E ans, S. (2022, Ap il 3). Dan Mu phy’s loyalis s d i e Endea ou G oup. Aus alian Financial Re iew.
h ps://www.a .com/companies/in as uc u e/dan-mu phy-s-loyalis s-d i e-endea ou -g oup-202
20331-p5a9l
Golden ein, J., & McGuigan, L. (2023). Managed so e eigns: How inconsis en accoun s o he
24 In e ne Policy Re iew 13(2) | 2024
