Küble , Raoul; Adle , Susanne J.; Welke, Lina; Sa s ed , Ma ko; Pauwels, Koen
A icle
Mining consume mindse me ics wi h use -gene a ed
con en
Schmalenbach Jou nal o Business Resea ch (SBUR)
P o ided in Coope a ion wi h:
Schmalenbach-Gesellscha ü Be iebswi scha e.V.
Sugges ed Ci a ion: Küble , Raoul; Adle , Susanne J.; Welke, Lina; Sa s ed , Ma ko; Pauwels, Koen
(2025) : Mining consume mindse me ics wi h use -gene a ed con en , Schmalenbach Jou nal o
Business Resea ch (SBUR), ISSN 2366-6153, Sp inge , Heidelbe g, Vol. 77, Iss. 3, pp. 497-525,
h ps://doi.o g/10.1007/s41471-025-00219-4
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/331934
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/4.0/
ORIGINAL ARTICLE
h ps://doi.o g/10.1007/s41471-025-00219-4
Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525
Mining Consume Mindse Me ics Wi h
Use -Gene a ed Con en
Raoul V. Küble · Susanne J. Adle · Lina Welke · Ma ko Sa s ed ·
Koen Pauwels
Recei ed: 8 Oc obe 2024 / Accep ed: 16 July 2025 / Published online: 4 Sep embe 2025
© The Au ho (s) 2025
Abs ac In he wake o digi al ans o ma ion, ma ke e s gained access o la ge
amoun s o use -gene a ed con en and da a in which consume s speci ically men ion
and discuss b ands, p oduc s, and se ices. This da a o e s ich in o ma ion po en-
ial and may ul ima ely p o ide ma ke e s wi h he abili y o use his da a pool o
app oxima e su ey-based consume mindse me ics ha mi o consume a i udes
alongside he di e en le els o he decision-making p ocess. We a gue ha le e -
aging his po en ial may ul ima ely help ma ke e s o e come common limi a ions
o su ey-based me ics and enable companies o obse e and ack mindse me ics
ha ha e been so a inaccessible due o inancial and o he cons ain s. To his
end, we p opose a ou -s ep p ocess ha i s iden i ies he key aspec s o a mindse
me ic based on he exis ing body o de eloped cons uc s, hen pinpoin s po en ial
da a sou ces, and subsequen ly chooses an adequa e da a ans o ma ion ool.
Keywo ds Mindse me ics · Use -gene a ed con en · Cons uc alidi y ·
Machine lea ning · Cus ome decision jou ney
Raoul V. Küble · Lina Welke
ESSEC Business School, 3 A enue Hi sch, 95800 Ce gy, F ance
E-Mail: kuble @essec.edu
Susanne J. Adle · Ma ko Sa s ed
Ins i u e o Ma ke ing, LMU Munich School o Managemen , Ludwig Maximilian Uni e si y o
Munich, Ludwigs . 28RG, 80539 Munich, Ge many
Ma ko Sa s ed
Facul y o Economics and Business Adminis a ion, Babe,
s-Bolyai-Uni e si y, S ada Teodo
Mihali 58–60, 400591 Cluj-Napoca, Romania
Koen Pauwels
D’Amo e-McKim School o Business, No heas e n Uni e si y Bos on, 450 Dodge Hall,
360 Hun ing on A enue, Bos on, Massachuse s 02115, USA
K
498 Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525
1 In oduc ion
The digi al ans o ma ion o ou wo ld led o subs an ial changes in how much da a
is a ailable o decision-make s wi hin he las wo decades. Wi h millions o digi al
de ices obse ing and acking ou e e yday li es, and consume s ac i ely c ea ing
da a poin s whene e hey b owse, shop, o in e ac wi h each o he , we ha e en e ed
he da a- iches pe iod in human his o y. Today, humans and he de ices hey use
willingly and unwillingly c ea e mo e han 402.74 million e aby es o da a each
day (Dua e 2024), su passing he cumula ed amoun o da a humankind has c ea ed
since he dawn o ci iliza ion. A la ge p opo ion o his da a is ac i ely c ea ed
by consume s h ough use -gene a ed con en (UGC) such as (bu no exclusi ely)
social media pos s, e iews, ideos, messages, online sea ch pa e ns, online se ice
in e ac ions, websi e isi s, o podcas s. Use s p o ide aluable and in o ma ion-
ich da a wi h a speci ic ocus on consump ion pa e ns, p oduc p e e ences, sa -
is ac ion in o ma ion, o company and p oduc e alua ions. In ligh o he g owing
a ailabili y o da a, new me hods ha e eme ged o ex ac in o ma ion and insigh s
om la ge da ase s a ailable o decision-make s (Küble e al. 2017). Wi hin he las
decade, subs an ial de elopmen s in machine lea ning echniques, coupled wi h he
a ailabili y o mo e powe ul ha dwa e, enabled decision-make s o accu a ely de-
sc ibe, p edic , explain, and e-gene a e human beha io . Deep lea ning algo i hms
ha e cumula ed in la ge language models, which, h ough hei cha in e aces o
simple applica ion desc ip ions, now allow access o powe ul algo i hms o gene -
a ing consume insigh s (e.g., h ough he gene a ion o syn he ic da ase s; Sa s ed
e al. 2024a) wi hou he need o sophis ica ed coding expe ise (e.g., Ha mann
e al. 2023). The combina ion o la ge amoun s o da a, machine lea ning me hods,
and powe ul ha dwa e has also enabled manage s wi h he abili y o accu a ely
measu ing, acking, o p edic ing consume beha io such as b and and p oduc
conside a ion and pe cep ion (Ringel and Skie a 2016), as well as ad clicks (Wang
e al. 2018), pu chase (Che alie and Mayzlin 2006), o chu n beha io (Khodaban-
dehlou and Rahman 2017).
While machine lea ning applica ions ha e been shown o be able o ul ima ely
p edic inal decision ou comes wi h he help o UGC, he p edic ion o app oxima-
ion o la en a iables such as consume mindse me ics (CMMs) wi h UGC, has
no ye been ully explo ed (Hai and Sa s ed 2021). Ins ead, manage s commonly
s ill ely on p ima y da a o measu e and ack p e-pu chase consume a i udes such
as b and awa eness, b and conside a ion, o pu chase in en ion, as well as pos -
pu chase a i udes such as b and sa is ac ion, o ecommenda ion beha io . This
is especially su p ising as p ima y da a collec ed h ough su eys wi h consume
panels has been shown o ace se e al es ic ions such as high cos s, educed ime-
liness, and p oneness o sampling, as well as esponse e o s (Hulland e al. 2018).
In addi ion, manage s o en ely on la en CMMs as key pe o mance indica o s in
ma ke ing, which a e o en included in co po a e dashboa ds and used o e alua e
he sho - and long- e m impac o ma ke ing ac ions.
Simila o o he asks in consume (Ringel and Skie a 2016) and ma ke obse a-
ions (Ma he e al. 2023), UGC may be an al e na i e da a sou ce o app oxima e,
measu e, and ack CMMs. Doing so would o e mul iple ad an ages: Fi s , sec-
K
Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525 499
onda y da a such as UGC is con inuously a ailable and can hus also be used on
a daily base o moni o b and heal h o measu e and ack he impac o ma ke ing
ac i i ies. Second, UGC is eely a ailable and is commonly no p one o adi ional
biases known om p ima y da a esea ch such as s a egic answe ing beha io , su -
ey a igue, and non- esponse bias. Thi d, UGC is commonly ied o ue cus ome
expe iences (i.e., p oduc e iews o cus ome complain s on X) and can hus be
unde s ood as a o m o implici , e ealed p e e ences, whe eas CMMs a e com-
monly unde s ood as s a ed p e e ences. Finally, being implici eedback, UGC can
be conside ed as a mo e na u al o m o eedback han su ey-based CMMs, which
a e—being an explici o m o eedback—known o be mo e p one o, o example,
social desi abili y bias. Despi e he po en ial o UGC o consume mindse acking,
so a li le insigh s exis in o he sui abili y and how o p ocess, use, and le e age
UGC o ack la en CMMs.
Add essing his gap in esea ch, we se ou o in es iga e and discuss he po en ial
p os and cons o such an app oach, as well as he necessa y s eps o acili a e he use
o UGC in he con ex o CMMs. Ou goal is o iden i y, sys ema ize, and map he
necessa y asks, de elop guidelines o he usage o UGC o measu e CMMs, and
iden i y u u e esea ch needs o ul ima ely de elop a s uc u ed esea ch agenda in
his domain.
The emainde o his pape is s uc u ed as ollows: Fi s , we highligh and dis-
cuss he cu en applica ions o CMMs in ma ke ing esea ch and ma ke ing p ac ice
o de e mine he s eng hs and weaknesses o he cu en app oaches as well as o
de e mine he needs and equi emen s ha need o be me by a UGC-based measu e-
men app oach. Subsequen ly, we highligh cu en applica ions o UGC in mode n
ma ke ing esea ch o de e mine he po en ial o UGC o CMM measu emen and
po en ial bounda y condi ions. Linking insigh s om bo h sec ions we hen de elop
a amewo k o combining bo h wo lds in which we highligh necessa y s eps ha
need o be add essed o le e age he po en ial o UGC o CMM measu emen .
Finally, we de elop a esea ch agenda o s imula e u he wo k in he ield ha is
necessa y o ul ima ely each he goal o making UGC a ailable o manage s o
measu ing and acking CMMs.
2 Consume Mindse Me ics
2.1 Concep s o Cus ome Mindse Me ics
Bo h psychology and ma ke ing li e a u e iden i y consume a i udes as cons uc s
ha indica e how consume s hink abou (cogni ion), eel abou (a ec ), and ac
owa d (cona ion) a b and (Vak a sas and Amble 1999) and o he objec s o in e -
es . S a ing in he ea ly 1960s (Colley 1961; La idge and S eine 1961), ma ke ing
de eloped measu es o consume a i udes o e alua e he impac o ma ke ing cam-
paigns and o p edic hei sales e ec . In hei heo y o buying beha io , Howa d
and She h (1969, p. 14) no ed, “A i ude is an inpu in o execu i e decisions because
many ma ke ing decisions, including ad e ising, can be mo e adequa ely e alua ed
o measu ed in e ms o a i ude han o pu chase beha io .” Co esponding a i ude-
K
500 Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525
ela ed measu es a e also e e ed o as CMMs in ha hey seek o cap u e he why-
ques ion necessa y o in e p e obse ed consume beha io . Fo ins ance, consume s
may isi a b and’s websi e because hey a e conside ing buying i s p oduc s, a e
b owsing o un, ha e ques ions abou p oduc s al eady bough , o wan o a ional-
ize hei choice o a compe ing b and (Do son e al. 2017). Pauwels and an Ewijk
(2020) show ha a i ude- ela ed me ics o e a supe io p edic ion o b and sales
o e se e al mon hs, as compa ed wi h agg ega e online beha io al me ics such as
he amoun o weekly clicks and isi s.
The li e a u e has b ough o wa d a mul i ude o CMMs. The mos p ominen
CMMs a e a guably b and awa eness, b and conside a ion, pu chase in en ion, b and
sa is ac ion, b and ecommenda ion, and b and equi y. B and awa eness cap u es
whe he a b and is op-o -mind and can be measu ed as a consume ’s abili y o ecall
a b and’s name (unaided) o as a consume ’s abili y o ecognize a b and among o he
b ands (aided; see, e.g., Mec edy e al. 2022; Pauwels and an Ewijk 2020). B and
conside a ion desc ibes which b ands consume s would ega d as sui able op ions
o usage o pu chase while pu chase in en ion cap u es which b ands consume s
would ac ually be willing o buy (e.g., Küble e al. 2020; Mec edy e al. 2022).
CMMs ha can only be o med a e ha ing pu chased and used a p oduc o se ice
include b and sa is ac ion and b and ecommenda ion. B and sa is ac ion cap u es
whe he a b and has me a consume ’s expec a ion while b and ecommenda ion
measu es a consume ’s willingness o sugges he b and o o he s, espec i ely o
ell o he s o a oid a b and (e.g., Anselmsson and Bondesson 2015; Küble e al.
2020). Las ly, b and equi y cap u es a consume ’s p e e ence o one b and o e
o he b ands (Washbu n and Plank 2002).
2.2 The Psychome ics o Cus ome Mindse Me ics
As he concep s unde lying CMMs a e inhe en ly unobse able, hei measu emen
ypically elies on consume s’ answe s o se s o su ey i ems. These i ems a e
supposed o cap u e he esponden s’ assessmen o speci ic ai s ha unde lie he
concep unde conside a ion (e.g., Anselmsson and Bondesson 2015; Pauwels and
an Ewijk 2020). Fo example, o measu e cus ome sa is ac ion, esea che s ou-
inely ely on h ee su ey i ems ha gauge he esponden s’ o e all sa is ac ion,
expec ancy con i ma ion, and pe cei ed pe o mance e sus he cus ome ’s ideal
p oduc o se ice in he ca ego y (Fo nell e al. 1996). F om a measu emen - he-
o e ic pe spec i e, hese i ems a e assumed o be e lec ions o consequences o
cus ome sa is ac ion (hence, e lec i e measu emen ), he eby ac ing as empi ical
ealiza ions o he unobse ed concep ; hei co ela ion is assumed o be “caused”
by he unde lying concep (Sa s ed e al. 2016a). A o ma i e measu emen o
cus ome sa is ac ion, on he o he hand, would cap u e con ibu ing ai s such as
sa is ac ion wi h he se ice, p ice, o p oduc ; assuming ha a composi e o such
indi idual ai s cap u es he a ge cons uc .
The logic unde lying o ma i e measu emen acknowledges he app oxima i e
na u e o measu emen whe eas e lec i e measu es implici ly assume ha any mea-
su emen can, in p inciple, cap u e a concep in ull (Rigdon and Sa s ed 2022).
The mani old unce ain ies ha come wi h any measu emen , o example, wi h e-
K
Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525 501
ga d o he concep ualiza ion, i em wo dings, measu emen scales, da a collec ion,
and alida ion, howe e , call he assump ion o pe ec measu emen in o ques ion
(Rigdon e al. 2019,2020; Sa s ed e al. 2024b).
Conside ing he p ac ical challenges ha come wi h collec ing da a on mul i-
i em measu es (e.g., wi h ega d o ca eless esponse beha io o non- esponse),
esea che s and pa icula ly p ac i ione s equen ly ely on single-i em measu es o
a concep . These single i ems can be seen as a global assessmen o he concep ,
such as when asking a esponden : “O e all, how do you a e you sa is ac ion wi h
he company’s se ice?” The p ac ical ad an ages, howe e , come a he expense
o educed eliabili y and lowe p edic i e alidi y, which may comp omise hei
use ulness in esea ch se ings, po en ially igge ing lawed manage ial ecommen-
da ions (e.g., Diaman opoulos e al. 2012; Sa s ed e al. 2016b, c).
In hei s udies, manage s ypically cus omize he exac measu emen s o hei
b and o make changes in esponse o obse ed ela ionships, o example, p un-
ing me ics ha we e oo highly co ela ed amongs each o he o did no p edic
beha io o e ime (e.g., Pauwels and Joshi 2016). This measu emen lexibili y
also applies o measu es o CMMs ha a e no ully s anda dized. Resea ch has
wi nessed a p oli e a ion o all so s o me ics ha claim o measu e essen ially he
same concep , al hough o en wi h li le chance o con e one ins umen ’s mea-
su es in o any o he ins umen ’s measu es (Salzbe ge e al. 2016). Fo example,
esea ch and p ac ice ha e p oposed a mul i ude o measu emen ins umen s o
co po a e epu a ion, which es on he same de ini ion o he concep bu di e un-
damen ally in e ms o hei unde lying concep ualiza ions and measu emen i ems
(Sa s ed e al. 2013). Simila ly, Be gk is and Langne (2017,2019) ind conside -
able he e ogenei y in he ope a ionaliza ions o common ad e ising cons uc s, such
as a i ude owa d he ad, a i ude owa d he b and, ad c edibili y, ad i i a ion, and
b and pu chase in en ion. In addi ion, cons uc concep ualiza ions and ope a ional-
iza ions change o e ime (Be gk is and Eisend 2021), while he heo e ical en i y
o in e es (i.e., he concep ual a iable) gene ally emains he same. These indings
sugges he e is no se way o pe ec ly measu e a concep , e en hough some s an-
da dized measu emen app oaches can be obse ed. Fo example, b and awa eness
is measu ed as he unaided ecall and aided ecogni ion o a b and (e.g., Mec edy
e al. 2022; Pauwels and an Ewijk 2020). O he CMM measu es exhibi a lowe
consensus. To measu e b and conside a ion, pu chase in en ion, b and sa is ac ion,
o ecommenda ion in en ions, consume s may selec ele an b ands om a lis o
mul iple b ands (e.g., Colice and de B uyn 2023; Küble e al. 2020; Mec edy e al.
2022) o indica e hei assessmen on a Like -scale (e.g., Anselmsson and Bondes-
son 2015;Cain2022; Pauwels and an Ewijk 2020). Resea che s ha e also sugges ed
he use o a Ne P omo e Sco e o which esponden s indica e hei willingness o
ecommend a b and on a scale om 0 o 10 and a e so ed in o endo semen ca e-
go ies because o hei a ing (Baeh e e al. 2022). Measu es o sa is ac ion ange
om asking pa icipan s o selec b ands hey a e sa is ied, espec i ely no sa is ied,
wi h (Küble e al. 2020) o using mul i-i em scales (Anselmsson and Bondesson
2015; Pe e sen e al. 2018). Mo eo e , he ela ionship be ween su ey-measu ed
CMMs and consume beha io is no s ong poin ing owa d po en ially limi ed
p edic i e alidi y (Pauwels and an Ewijk 2020).
K
502 Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525
Table 1 Popula UGC Channel Types and Examples
Channel Type Examples Exempla y s udies
Social media
pla o ms
Facebook, Ins ag am, X, TikTok,
LinkedIn, Sina Weibo, Snapcha
Liu e al. (2016): Twi e (now X) Twee s o
p edic TV show demand
Online e iew
pla o ms
Yelp, T ipAd iso , T us pilo Rocklage e al. (2021): Emo ionali y o Yelp
e iews o p edic es au an able ese a ions
E-comme ce
pla o ms
Amazon, eBay, E sy Che alie and Mayzlin (2006)use e iews o
p edic sales
Sea ch engine
logs
Google, Bing, Yahoo Ringel and Skie a (2016) use Google sea ch
e m olume o map consume conside a ions
Blogs and
mic oblogs
Medium, Subs ack, Blogge ,
Tumbl
Onishi and Manchanda (2012) examine impac
o blog pos s on b and awa eness and sales
Q&A o ums Quo a, S ack O e low, Reddi Bu ch e al. (2022) examine he impac o
badges on UGC c ea i i y
S eaming
pla o ms
Twi ch, YouTube Li e, Douyin,
Huya Li e, RedNo e
Lin e al. (2021) examine impac o emo ions in
li e s eams on ips and use ac i i y
Podcas s Spo i y, Apple Podcas s, Sound-
Cloud
Kozine s e al. (2010) in es iga e how podcas -
based seeding in luences sp ead o UGC and
eWoM
3 Use -Gene a ed Con en
UGC has been in he spo ligh o ma ke ing esea ch now o o e wo decades.
Ini ially, o en gene ally e e ed o as (online o elec onic) wo d o mou h (Godes
and Mayzlin 2004; Hennig-Thu au e al. 2004), he unde s anding and de ini ion
o UGC ha e become mo e g anula o e ime. While he adi ional digi al o
elec onic wo d o mou h li e a u e limi s i sel o UGC ha speci ically ocuses on
consump ion- ela ed con en (Babi´
c Rosa io e al. 2020), he classic UGC li e a u e
goes beyond and includes any so o digi al in o ma ion c ea ed by a p i a e (in
con as o a paid) use ’s communica ion and online beha io and ha is sha ed wi h
he gene al public ia any so o digi al (web) channel (Daughe y e al. 2008). UGC
da a sou ces in es iga ed in he li e a u e include a b oad ange o di e en channels.
Table 1p o ides an o e iew o popula channel ypes, along wi h examples. Wi h
he cons an eme gence o new channels, he olume and a ie y o digi al aces
le by consume s con inue o expand.
A key cha ac e is ic o UGC is ha i a ely occu s in a s uc u ed (i.e., nume ic)
da a o ma (de Haan e al. 2024). I so, o en simple coun measu es a e used as
app oxima ions, such as popula i y (e.g., he numbe o ollowe s), b and liking (e.g.,
h ough he coun o daily likes o b and pos s), and b and-consume ela ions (e.g.,
h ough he coun o daily b and pos sha es). These simple app oaches, howe e ,
neglec he ichness o in o ma ion wi hin UGC ha a ises om he ac ha use s do
no only c ea e in o ma ion by engaging wi h b and con en , bu also ac i ely c ea e
pos s, commen s, ideos, cha s, e c. Mo e impo an ly, as e idenced in he lis o da a
sou ces abo e, he as majo i y o UGC comes in uns uc u ed (i.e., non-nume ic)
da a such as images, ex , and ideos. Se e al s udies ha e made use o such da a o
gene a e consume insigh s. Fo example, he majo i y o esea ch elying on UGC
o app oxima e common b and equi y me ics such as b and knowledge o b and
K
Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525 503
s eng h ha e commonly d awn on ex ual da a ob ained om a ious online and
social media sou ces (see, e.g., Colice e al. 2018). Fewe s udies, howe e , ha e
gone one s ep u he and used non-nume ic UGC, such as images (Dzyabu a e al.
2023; Ha mann e al. 2021), audio (Wang e al. 2021), o ideo (Zhou e al. 2021)
o cap u e consume o b and insigh s.
Uns uc u ed da a b ings he ad an age o iche in o ma ion and hus ba es mo e
po en ial o gene a ing unique insigh s (de Haan e al. 2024). A he same ime,
howe e , UGC da a needs o be ans o med in o s uc u ed da a in o de o be
u he p ocessed and analyzed. Thus, he need o da a ans o ma ion mus be
conside ed as ano he key cha ac e is ic o UGC. De Haan e al. (2024) p o ide
a de ailed and s uc u ed amewo k o how o u n uns uc u ed da a in o a sui able
s uc u ed o ma , aking a ious applica ion and company-speci ic con ingencies
in o conside a ion.
The mos commonly used ools o da a ans o ma ion—a leas in he con ex o
b and measu emen (Be ge e al. 2020)—a e sen imen measu es. Fo ex ual da a,
sen imen measu emen ools can be spli in o op-down and bo om-up app oaches
(Humph eys and Wang 2018). Top-down app oaches such as he o en-used LIWC
o VADER commonly ely on wo d lis s cu a ed by linguis ic esea ch o cap u e
sen imen o alence in o ma ion. By con as , bo om-up app oaches use machine
lea ning ools and p e-coded aining da a o p edic he sen imen o a documen by
looking a he embedded wo ds o wo d combina ions wi hin he documen (Küble
e al. 2020). In he case o audio, ei he ansc ip ions o he spoken con en a e
used o ely on s anda d ex ual sen imen classi ie s (Zhou e al. 2021), o ools o
de ec loudness o one wi hin oice o cap u e sen imen (Wang e al. 2021). Fo
images, esea ch so a mos o en elies on acial exp essions o cap u e sen imen
and alence (Toisoul e al. 2021), which, howe e , equi es images o display aces.
Simila ly, in he case o ideos, he ideo ma e ial is cu in o indi idual ames,
which a e hen subsequen ly ea ed like images (e.g., Li e al. 2019). Besides he
bina y measu emen o sen imen (posi i e and nega i e), an inc easing numbe o
s udies now a emp o cap u e no only he alence, bu subdimensions o emo ions
such as ange , ea , anxie y, sadness, joy, su p ise, and an icipa ion (e.g., Holiday
e al. 2023; Schwenzow e al. 2021; Zhou e al. 2021).
4 Connec ing Concep ual Face s Wi h UGC Insigh s
CMMs a e ypically measu ed wi h se s o i ems ha co e mul iple ace s and
aspec s o he concep unde conside a ion. The la en concep o sa is ac ion may,
o example, be measu ed o ma i ely by asking consume s o a e a ious aspec s
o a se ice (e.g., cleanliness, p oximi y, calmness, as e), which join ly o m sa is-
ac ion. In case o a e lec i e app oach, he same concep would be measu ed by
he a emp o cap u e he ou come o he cons uc such as, “How sa is ied a e you
wi h he b and?” and “How well does he b and ma ch you expec a ions?” (Fo nell
e al. 1996). T adi ional CMMs measu emen he e o e cap u es s a ed p e e ences,
collec ed h ough explici eedback. In con as , UGC inhe en ly e lec s aspec s o
e ealed p e e ences embedded in implici eedback. This con en o en con ains
K
504 Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525
Table 2 Ad an ages and Disad an ages o T adi ional Da a e sus UGC o CMM Measu emen s
Ad an ages Disad an ages
T adi ional
da a
Sampling:
Po en ial o ensu e and check o he ep esen a i eness o
a sample o a speci ic popula ion
Ta ge ed sampling o , o example, speci ic g oups
Sampling:
Po en ially high da a collec ion cos s o ec ui ing and incen i izing s udy pa icipan s
Reduced imeliness
P one o sampling e o s including non- esponse biases
Po en ial ca eless esponse beha io
Scale use:
Di ec measu emen s, o example, conce ning a i udes
Valida ed, s anda dized measu es a ailable
Scale use:
Scale-speci ic disad an ages (e.g., educed eliabili y and lowe p edic i e alidi y in single-
i em scales)
Many implici assump ions in scale design inc ease unce ain y (e.g., scale ypes, wo ding o
i ems and answe ca ego ies)
Use o uns anda dized, ad-hoc measu es
Scale modi ica ions may change he na u e o measu emen ; necessa y scope o en unclea
Eme gence o esponse biases (e.g., social desi abili y bias)
UGC Cap u es eal-wo ld consume beha io
A ailable om mul iple sou ces, con inuously a ailable, and
o en eely a ailable
Rich in o ma ion om combining mul iple da a sou ces
a ailable
Less p one o adi ional biases known om p ima y da a e-
sea ch (e.g., social desi abili y, su ey a igue, non- esponse
bias)
Unce ain da a composi ion in e ms o a ge g oups who p o ide UGC
Sel -selec ion bias (e.g., dominance o nega i e esponses and da a om e y ac i e use s)
Risk o incomple e cons uc assessmen
Measu es a e mos ly indi ec and app oxima i e
Objec o measu emen may be ambiguous (e.g., p oduc - ela ed s. b and- ela ed assess-
men s)
Combina ion o da a om a ious sou ces can be p ohibi i ely complex and expensi e
Da a is mos ly uns uc u ed, and ans o ma ion is o en necessa y
K
Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525 511
Table 3 (Con inued)
Me ic Cha ac e is ics Sample I ems Tex -based sou ces Image-based sou ces Video-based sou ces
B and
ec-
om-
men-
da-
ion
Consume s’
willingness
o encou age
o he s o buy
o use a b and
Which o he ollowing
b ands would you ec-
ommend o a iend o
colleague?
Which o he ollowing
b ands would you ell
a iend o colleague o
a oid?
How likely is i ha
you would ecommend
b and X o a iend o
colleague?
Use wo d lis s/dic iona ies o ack ec-
ommenda ion-speci ic pos s.
Ex ac Ne P omo o Sco e-like ele-
men s (e.g., 10 ou o 10).
Use opic models o iden i y and ack
es imonial s a emen s.
Ex ac and collec ecommenda ion
sco es om, e.g., e iews.
Iden i y UGC whe e use s ag o he s
ecommending he b and.
Apply NLP ools o cap ions,
ags, o commen s.
Use compu e ision ools o
iden i y in e ac i e elemen s
on isual social media such
as in e ac i e s o ies, polls, o
ecommende links.
Use compu e ision ools o
iden i y he con ex in which
es imonials p esen he p oduc .
Apply NLP ools o audio
ansc ip s, cap ions, ags, o
commen s.
Apply compu e ision ools
o ideo ames.
B and
eq-
ui y
The abili y
o a b and o
a ec buying
decisions
based on
objec i e o
subjec i e
cha ac e is ics
consume s
alue mo e
o he ocal
b and han o
i s compe i o s
I makes sense o buy
X ins ead o any o he
b and, e en i hey a e
he same.
E en i ano he b and
has he same ea u es as
X, I would p e e o buy
X.
I he e is ano he b and
as good as X, I p e e o
buy X.
I ano he b and is no
di e en om X in any
way, i seems sma e o
pu chase X.
Coun he numbe o s a emen s whe e
use s exp ess a a ionale o choosing
a ocal b and o e o he s.
Rely on wo d embedding models o iden-
i y he numbe o ins ances in which
UGC acknowledges ha compe i o s
migh be “as good” bu s ill exp ess p e -
e ence o he ocal b and.
Rely on compa ison mining me hods (use
dependency pa sing o o he syn ac ic
ools o ex ac compa a i e s a emen s
in ol ing ocal b ands and compe i o s).
Use ABSA o conduc easoning ex ac-
ion, by employing a logic classi ie ha
iden i ies ph ases whe e use s a ionalize
o jus i y hei choices.
Apply NLP ools o cap ions,
sea ch o hash ags, and o he
iden i ie s.
Rely on objec classi ie s o
ex ac logos and unde s and
which b ands a e mo e o en o
mo e p ominen ly ea u ed.
Rely on objec classi ie s o
g asp se ing and apply wo d
ec o models o iden i y objec s
o unde s and he i be ween
con en and b and.
Coun he numbe o epea ed
b and ea u es/occu ences in
a speci ic se ing o o speci ic
keywo ds o sea ch e ms.
Apply NLP ools o cap ions,
sea ch o hash ags, and
o he iden i ie s.
C ea e ansc ip s om
oiceo e andapplyNLP
ools o ob ained ex .
Rely on ec o -space models
o b and men ions o iden i y
compa ison ac oss b ands,
ely on dis ance om model
o unde s and s eng h.
K
512 Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525
uns uc u ed da a, which limi s i s e ec i eness in measu ing consume a i udes like
b and sa is ac ion o ecommenda ion.
Podcas pla o ms p ima ily p o ide la gely uns uc u ed da a h ough ansc ip s
a he han nume ic engagemen me ics such as likes o s eam coun s. While hese
ansc ip s can o e ich insigh s in o consume e alua ions and discussions abou
a b and, hey a e less use ul o gauging b and awa eness a he beginning o he
ma ke ing unnel.
5.3 In o ma ion Ex ac ion
In he hi d s ep, esea che s need o ex ac he aspec in o ma ion om he UGC
wi h he help o adequa e ools. As shown in Table 3, a ious app oaches and
me hods a e sui able o his ask. The choice o a me hod depends on he da a
o ma (s uc u ed s. uns uc u ed), he complexi y o he da a, he ype o da a
( ex s. images s. ideos), as well as budge and ime cons ain s.
In he case o op and bo om o he unnel me ics, such as b and awa eness and
sa is ac ion, whe e nume ic in o ma ion is a ailable and easily accessible, simple
me hods ha ely on coun da a may be su icien o p o ide ini ial insigh s (see,
e.g., Küble and Seggie 2024). T acking he numbe o ollowe s o social media
pla o ms o coun ing he numbe o e iew a ings can alidly depic b and awa e-
ness o sa is ac ion, as demons a ed by nume ous s udies in he ield using such
da a (see, e.g., Hewe e al. 2016; Colice e al. 2018; o Küble e al. 2018) o
app oxima e o explain b and awa eness o sa is ac ion. Howe e , unde s anding he
a ious ace s o hese cons uc s equi es a mo e nuanced app oach ha d aws on
uns uc u ed da a and applies mo e complex ools.
Fo CMMs, whe e he iden i ica ion o b and men ions o cap u e b and awa e-
ness o b and conside a ion, is essen ial, esea che s may ely on ools de eloped
o objec iden i ica ion in uns uc u ed da a. This may in ol e b and name lis s
(o dic iona ies) o ag social media pos s o calcula e sha e o oice sco es by
compa ing how many imes he own b and is men ioned compa ed o o he s (e.g.,
Ringel and Skie a 2016). Fo non- ex ual da a, objec classi ie s o objec de ec o
models may be used o ecognize b and logos o key p oduc s o achie e a simi-
la ou come. Besides using one o he a ious comme cial o - he-shel solu ions
buil in o AWS and MS Azu e o open-access ools hos ed on pla o ms such
as HuggingFace (see, e.g., h ps://hugging ace.co/spaces/na hanjc/Logo_de ec ion_
YoloV7), esea che s may build hei own de ec o models. T aining and ine- uning
such classi ie s ha e become easie wi h a ious packages in R o Py hon, o e ing
he applica ion o s a e-o - he-a de ec ion models—see, o example, Yildi im and
Küble (2023) o a logo iden i ie ma kdown app oach. The a ailabili y o p e-
ained models on pla o ms such as HuggingFace ha only equi e a “ ew sho s”
o aining da a (commonly less han 100 images pe objec ) o deli e eliable e-
sul s, has also signi ican ly con ibu ed o he dissemina ion o classi ie s—see, o
example, Ca ion e al. (2020) o a ans o me -based objec de ec ion model.
In si ua ions whe e no only b and awa eness needs o be acked, bu also b and-
ela ed elemen s such as dimensions o b and equi y (e.g., Yoo and Don hu 2001),
o he machine lea ning-based echniques may be help ul. Fo ex ual da a, opic
K
Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525 513
models, which allow he iden i ica ion o la en aspec s wi hin ex da a (Büschken
and Allenby 2016), may help ex ac such insigh s. The key idea o hese models
is ha speci ic wo ds a e ied o opics wi hin documen s and ha documen s a e
composed o di e en opics (Blei e al. 2003). Di e en o ms o opic models
a e a ailable wi h La en Di ichle Alloca ion (LDA) models being he mos widely
applied ones (e.g., Adle and Sa s ed 2021; Adle e al. 2024). Howe e , esea che s
inc easingly apply s uc u al opic models (STM), which can accoun o e ol ing
ends o e ime (Robe s e al. 2019), o models ha accoun o wo d embeddings
such as BERTopic (G oo endo s 2022). Fo mo e de ails and a s ep-by-s ep guide
on how o es ima e opic models, see Yildi im and Küble (2023).
In he case o image (and ideo) da a, objec classi ie s can simila ly be used
o cap u e con ex ual in o ma ion and de i e insigh s abou a ious aspec s o he
meaning o a b and. By coun ing o measu ing he ypes o objec s commonly shown
alongside a b and o p oduc , esea che s may ob ain be e in o ma ion abou how
consume s pe cei e b ands. In some cases, his may equi e a combina ion o ools,
whe e an objec de ec o is used o iden i y all objec s occu ing wi h a b and, and
wo d- ec o models a e subsequen ly used o au oma ically de i e meaning om
he ob ained b and-objec -embeddings—see, o example, he app oach o con ex
i measu emen ha Küble e al. (2024)p esen .
As shown in Table 3, he abili y o cap u e and in eg a e con ex ual in o ma ion
becomes inc easingly ele an o mid- unnel CMMs, such as b and conside a ion
and pu chase in en ion. Fo he o me cons uc , when elying on ex ual da a, p e-
cu a ed wo d lis s, opic models o ec o models may be used o iden i y which
b ands a e commonly men ioned oge he in a consump ion con ex . Simila ly, e-
sea che s may employ hese me hods o de e mine which o he b ands a e equen ly
men ioned o sea ched alongside hei ocal b and o depic he consume s’ conside -
a ion se (e.g., Ringel and Skie a 2016). Subsequen ly, he sha e o oice o men ions
can be used o unde s and he posi ion wi hin he conside a ion se as a measu e
o he s eng h o conside a ion. The same can be achie ed wi h he help o objec
classi ie s in he con ex o isual in o ma ion, such as images o ideos.
While sha e o oice may su ice o cap u e b and conside a ion, pu chase in en-
ion may equi e a mo e de ailed unde s anding o how s ongly a UGC elemen
sugges s ha he con en c ea o p e e s a men ioned b and o p oduc o e i s al-
e na i es (see, e.g., Ka niouchina e al. 2022). To add ess his conce n, esea che s
may ely on ex ual da a using ei he wo d dic iona ies (Pennebake 2001) ha de-
ec u u e- and goal-o ien ed language (like LIWC) o pa -o -speech agge s ha
ind ins ances whe e use s indica e clea buying in en ions (e.g., h ough he use o
wo ds like “can’ wai o ge i ” o “going o buy”) o p e e ences (e.g., h ough
wo ds such as “be e han” o “would buy X ins ead o Y”) as, e.g., demons a ed
by Sepeh i e al. (2023). Simila ly, sen imen sco es can be applied o UGC ha
discusses o compa es al e na i es. In pa icula , aspec -based sen imen analysis
(ABSA) can be le e aged o assign a sen imen sco e o each b and men ioned in
a UGC piece (Do e al. 2019). ABSA can be unde s ood as a combina ion o opic
modeling and sen imen analysis. P e- ained s a e-o - he-a models a e a ailable on
all common pla o ms and can be ine- uned wi h ew sho s (see, e.g., Se Fi ABSA
on h ps://hugging ace.co/blog/se i -absa). Fine- uning equi es esea che s o p o-
K
514 Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525
ide he model wi h aining da a ha ea u es aspec s (o in ou case, b ands) and
examples o posi i e and nega i e b and-speci ic commen s (Liu e al. 2020). Fo
images and ideos, one may apply he same echniques o ex ual da a coming wi h
an image o ideo, such as cap ions, commen s, sub i les, o ansc ip s o oiceo e s
(see, e.g., Ghosh and Deb 2022). Meanwhile, he iden i ica ion o pu chase-speci ic
UGC may be achie able by i s ying o iden i y images o ideos ha a e ied
o a pu chase con ex (e.g., poin o sale con en , unboxing pos s, o p oduc haul
ideos) o subsequen ly iden i y which b ands a e ea u ed mo e o en, as demon-
s a edbyO ie al.(2019).
The impo ance o cap u ing bo h con ex and con en wi hin UGC o accu a ely
unde s and he ele an aspec s becomes e en mo e impo an o he pos -pu chase
CMMs. In he case o sa is ac ion, one may di ec ly assess sa is ac ion by he sou ce
in o ma ion (i.e., by examining e iews o e iew-speci ic con en , such as e iew
images o p oduc e iew ideos) and ely on he emo ions exp essed wi hin his
con en o cap u e he di ec ion and he s eng h o sa is ac ion. Emo ions can be
cap u ed in di e en ways. Fo ex ual da a, emo ion dic iona ies such as EmoLex
(o en also e e ed o as NRC; Mohammad and Tu ney 2013) allow a quick and
easy measu emen o a ious emo ional dimensions such as joy, su p ise, an icipa-
ion, ange , disgus , and ea . Howe e , EmoLex su e s om he common dic io-
na y-based sho coming as i does no cap u e nega ions and is only a ailable o
a ew majo languages (Be ge e al. 2020). Ha mann e al. (2023) p o ide a p e-
ained sen imen analysis ans o me -based model ha cap u es he same emo-
ional dimensions, o e s g ea e accu acy, and can be ine- uned o b and-speci ic
con ex s. The esul ing in o ma ion can hen be used o c ea e a sa is ac ion index,
as demons a ed by he online ideo game dis ibu ion pla o m S eam, which also
uses an emo ion classi ie —accessible h ough S eam’s main API— o de i e a sa -
is ac ion sco e om i s e iew ex s (Guzs inecz and Sz˝ucs 2023). Ho z-Beho si s
e al. (2025) u he mo e p esen a new ool ha i s “emoji ies” ex ual in o ma-
ion o hen in e emo ions om he emoji ied con en . Simila ly, compu e ision
as a esea ch discipline ha a emp s o ans o m uns uc u ed isual con en in o
a s uc u ed o ma (Voulodimos e al. 2018) p o ides ools o he iden i ica ion o
acial emo ions, allowing o cap u e an emo ion sco e om p oduc e iew ideo
ames.
Bo h app oaches can also be applied o measu e b and ecommenda ion, a CMM
a he bo om o he unnel. He e, one may ac i ely use wo d lis s o opic models
o iden i y whe he a piece o UGC con ains a speci ic ecommenda ion, be o e
applying he p e iously discussed NLP and compu e ision ools o de e mine
which b ands a e ecommended and o wha easons. In addi ion, e e e and
a ilia e links, poll mechanisms, and sco es ha esemble classic a ings o ne
p omo e sco e o ing elemen s (such as “I would gi e 9 ou o 10” o “ o me
a clea i e-s a ho el”) may be used o iden i y ecommenda ion-speci ic con en . In
his case, he nume ic in o ma ion can also be le e aged o unde s and how s ongly
a piece ecommends a b and o p oduc .
K
Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525 515
5.4 Valida ion
Once one has collec ed and ans o med he da a wi h he help o he me hods
desc ibed abo e, he ques ion a ises, how well he ob ained measu es depic he
CMM and i he ob ained measu es can be used o ack cus ome mindse along
he decision-making unnel.
The alida ion o UGC measu es may, in p inciple, ollow s anda d app oaches
well-known om he psychome ics li e a u e, while accoun ing o he speci ici ies
o he da a ype. This p ocess s a s wi h e alua ing how well he UGC measu e
cap u es he concep domain iden i ied in S ep 1 (Fig. 1). Such an e alua ion is
subjec i e, bu sys ema ic in na u e, ypically d awing on esea che s wi h expe
knowledge in he domain. This quali a i e assessmen should be ollowed by a quan-
i a i e analysis ha examines he deg ee o which he UCG measu e co ela es wi h
an al e na i e measu e o he same concep (con e gen alidi y) and i s abili y o
p edic ele an ou come a iables (p edic i e powe ).
The s aigh o wa d app oach o assess con e gen alidi y is o use su ey-based
CMMs and es ima e he co ela ion be ween he su ey and UGC da a (e.g., Küble
e al. 2020). Such an analysis, howe e , equi es esea che s o ha e con ol da a a
hand, which is o en no he case due o inancial cons ain s o because such da a has
no been collec ed be o e and is hus no a ailable. In some cases, esea che s may
d aw on seconda y da a p oxies as su oga es o p ima y su ey da a. Fo example,
esea ch ins i u ions ha e used cus ome complain s as a p oxy o (dis)sa is ac ion
(Hun 1991). Such seconda y da a p oxies a ely co e he concep ’s domain in ull,
bu may s ill ac as easonable s anda ds o compa ison, p o ided ha he li e a u e
o e s suppo o hei use (Hous on 2002). Wi hou such heo e ical suppo , exam-
ining he p oxy’s co ela ion wi h o he UGC measu es wi hin he s udy can o e
clues ega ding disc iminan alidi y (i.e., he deg ee o which measu es o di e en
concep s a e su icien ly empi ically dis inc ).
The p ima y conce n in he alida ion p ocess should be es ablishing he UCG
measu e’s p edic i e alidi y o suppo i s ele ance o manage ial decision-making
(e.g., Sa s ed and Danks 2022). P edic i e alidi y assessmen may also d aw on
su ey-based con ol da a, bu in he absence o such da a, esea che s may es
he ob ained me ics by elying on he gene al concep o he decision-making
unnel. Acco ding o his concep , in es men s a each le el o he unnel (e.g.,
awa eness) will, o e ime, mo e down he unnel and a ec lowe le els o he
decision-making p ocess (e.g., Colice e al. 2018). Knowing ha each CMM le el
is linked o (o p edic s) a lowe -le el CMM and, in u n, o obse able ou comes,
one can de elop a adi ional KPI- ype model. Recognizing ha he e is a causal
chain be ween awa eness, in e es , conside a ion, pu chase in en ion, sa is ac ion,
and ecommenda ion, esea che s may a emp o ie he me ics ob ained om
UGC o pe o mance a iables ha a e obse able a each s age ollowing he lead
pe o mance indica o app oach ha Pauwels (2014) desc ibes. By unde s anding
how well each UGC-based CMM measu emen p edic s me ics u he down he
unnel, alongside pe o mance a iables a ailable wi hin he company (such as leads,
pu chases, o epea pu chases), esea che s may be able o con i m he alidi y o he
K
516 Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525
ob ained UGC measu es. Fo a guide on how o concep ually build and empi ically
es such a model, we e e o Hanssens e al. (2014) and Yildi im and Küble (2023).
While es ablishing con e gen and p edic i e alidi y should be ele an o all
ypes o UGC measu es, esea che s may conside u he alidi y ypes, depend-
ing on he na u e o he da a and i s sou ce. Fo example, Be ge e al. (2020)
p o ide a comp ehensi e amewo k o alida ing measu es de i ed om ex anal-
ysis. Rus e al. (2021) p opose a eal- ime b and epu a ion acke based on social
media mining using a cus ome equi y amewo k. Thei wo k highligh s he alue
o heo y-d i en UGC analysis and p o ides obus alida ion echniques by linking
b and d i e s o inancial ou comes. Finally, Hous on (2002) in oduced a alida-
ion guideline o measu emen s o ma ke ing cons uc s ha d aw on seconda y
da a. This p ocedu e comp ises a h ee-s ep p ocess, which pa ly o e laps wi h ou
amewo k (e.g., heo e ical speci ica ion), bu conside s addi ional elemen s such
as one-dimensionali y and nomological alidi y assessmen .
6 Discussion
Unde s anding how UGC ansla es in o consume pe cep ions, a i udes, and in-
en ions has become c ucial o manage ial decision-making. While ma ke ing e-
sea che s and p ac i ione s o en ely on nume ic da a (e.g., like coun s) o app oxi-
ma e consume assessmen s, UGC o e s much iche in o ma ion ha can be eaped
wi h oday’s machine lea ning ools. Add essing he challenges ha come wi h such
an app oach, we de eloped a ou -s ep p ocess ha uses i em con en s iden i ied
in psychome ic assessmen s o CMMs as a bluep in o sou ce iden i ica ion as
well as in o ma ion ex ac ion and con ol. To acili a e he adop ion o ou ame-
wo k, we classi y UGC da a sou ces acco ding o hei sui abili y o cap u ing
p ominen CMMs and iden i y ools o do so. Applying ou amewo k, ma ke ing
esea che s and p ac i ione s can e icien ly le e age UGC da a o measu e CMMs.
This app oach o e s u he bene i s o he ield: The combina ion o di e en UCG
measu es in e ms o , o example, nume ic da a (e.g., likes on social media) and
ex da a (e.g., e iews on e aile websi es) can help deepen he unde s anding o
a CMM h ough iangula ion. Insigh s om UGC da a can also be combined wi h
scale measu emen s om cus ome su eys o b ing he bes o he old and he new
wo lds oge he . Addi ionally, due o i s imeliness and lowe le el o heo e ical
inpu , UGC can also help imp o e CMM’s su ey measu emen s. Speci ically, com-
pa ing esul s om UGC o su ey i ems indica es which scale dimensions o i ems
a e ep esen ed well in UGC bu also which aspec s a e missing. Indeed, consume s’
online e iews could help iden i y which p oduc o se ice ea u es a e likely linked
o cus ome sa is ac ion. Such assessmen s can in o m o ma i e measu emen ap-
p oaches o ensu e ha he i ems cap u e he concep in ull. Simila ly, he absence o
hemes in UGC could help e ine scales. Responden s in su eys mus o en answe
all i ems o p oceed in he ques ionnai e—i espec i e o whe he a measu emen is
ac ually use ul (A is e al. 2014). I a epea ed assessmen on he g ounds o UGC
does no iden i y a speci ic measu emen aspec , i may no be ele an and could
be conside ed o emo al in u u e psychome ic assessmen s. UGC can he e o e
K
Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525 517
Table 4 Agenda o Fu u e Resea ch
Resea ch Agenda Speci ic Resea ch Ques ions
Concep ualiza ion Which aspec s a e sui able o p oduc -speci ic CMM cha ac e is ics?
Which aspec s a e sui able o b and-speci ic CMM cha ac e is ics?
How s ongly can UGC based CMM measu es be s anda dized s. need o cus omiza ion?
Which me ic/UGC da a combina ion p oduce which le el o unce ain y?
Sou ce Iden i ica-
ion
Which da a sou ce is sui able o which CMM aspec ?
How o iden i y b and- and p oduc ele an UGC o CMM measu emen ?
How o sample UGC da a acco ding o he a ge g oup?
Which mix o UGC sou ces is necessa y o ensu e ep esen a i eness?
How o a oid J-shape ap o alence?
How o bes a oid da a dependence and ensu e pe manen da a access?
How much is UGC da a p one o manipula ions and how o a oid ake insigh s?
In o ma ion Ex-
ac ion
Which con en measu e ool o which CMM aspec and da a sou ce?
Which CMM aspec s gene ally equi e which ex ac ion ool ai s?
Wha a e cos s s. bene i s ade-o s o measu emen app oaches in e ms o CMM measu e imp o emen ?
Wha a e cos s and bene i s o mul imodal s. singula model measu emen app oaches
Valida ion Wha a e sui able alida ion guidelines o UGC-based CMMs?
Do we need di e en guidelines o di e en ypes o CMMs?
Which eliabili y and alidi y ypes should be conside ed in he alida ion o UGC-based CMMs?
How can ake e iews and AI-gene a ed con en be iden i ied?
K
518 Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525
se e as an ou side alidi y check o es ablished scales ha would be speci ically
use ul i he cons uc is con ex -dependen o ime- a ian . Due o he a ie y o
channels ha p oduce UGC, esea che s may also d aw on a wide ange o da a
om di e se popula ions (e.g., eens and younge adul s on TikTok and business
p o essionals on LinkedIn). Wi h hei speci ic socio-demog aphics, expe iences,
and a i udes, using UGC o scale e inemen may aid o eplace explo a o y s eps
in scale de elopmen .
7 Fu u e Resea ch
While ou amewo k se s he s age o using UGC o CMM measu emen , i also
o e s oom o ollow-up esea ch, which we summa ize in Table 4.
As o he iden i ica ion o key CMM cons uc aspec s, we i s ocused on ex-
ending measu emen s o CMMs wi h UGC. Compa ed o he uns uc u ed da a ha
UGC p o ides, CMMs come in he o m o s anda dized measu emen s. Howe e ,
measu es alida ed on psychome ic g ounds p o ide an e en highe le el o s an-
da diza ion and can be accompanied by no ms ha iden i y which alues on a scale
a e high o low compa ed o a speci ic popula ion (Rigdon and Sa s ed 2022). This
le el o s anda diza ion is, as o now, no achie ed by measu es elying on UGC.
Fu u e esea ch should hus p io i ize he deli e y o an unde s anding whe he UGC
CMM measu es always need o be cus omized o i hey can also be s anda dized,
and i so, wha is equi ed o his p ocess.
Rela edly, in applying ou amewo k, esea che s need o explici ly acknowledge
ha UGC da a only allows o app oxima ing he concep s o in e es . Recen e-
sea ch highligh s he ole o measu emen unce ain y in psychome ic assessmen s
o concep s in his con ex , which induces a alidi y gap be ween he heo e ical
concep and he cons uc (e.g., Rigdon e al. 2019). Using UGC da a is likely o
widen his gap due o he idiosync asies o he da a and he esea che ’s deg ees
o eedom in he p ocessing and analysis o he da a. The use o UGC in CMM
measu emen he e o e emphasizes Rigdon e al. (2019) call o pu g ea e e o in o
quan i ying measu emen unce ain y (see also Rigdon and Sa s ed 2022). Fu u e
s udies should also de elop alida ion guidelines o me ics de i ed om UGC.
While con en , con e gen , and p edic i e alidi y, as highligh ed abo e, a e impo -
an elemen s, o he aspec s o alidi y a e also ele an . Co esponding s udies may
build on he concep ual wo ks o Be ge e al. (2020) and Rus e al. (2021), o
d aw on ex an guidelines p oposed in ela ed con ex s (e.g., seconda y da a alida-
ion; Hous on 2002). The goal should be o de elop a comp ehensi e app oach ha
enables esea che s o assess he alidi y o a gi en me ic based on UGC ai s,
CMM aspec s, and ex e nal alida ion da a cha ac e is ics. This app oach should
help de e mine how well a me ic aligns wi h exis ing ex e nal da a sou ces and
e alua e i s alidi y holis ically. In doing so, u u e esea ch should hus also a ge
he ques ion o which aspec s d i e o minimize me ological unce ain y, which has
been iden i ied as a majo con ibu o o low eplica ion a es (Rigdon e al. 2019).
Rega ding da a sou ce managemen , u u e esea ch should i s es ablish a clea e
unde s anding o which sou ces—and combina ions o sou ces—a e mos sui able
K
Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525 519
o UGC-based CMM measu emen . This in ol es assessing he ep esen a i eness
o da a sou ces o a gi en a ge g oup and, on he o he , de eloping a mo e obus
app oach o cons uc ing a alid sample om di e se UGC sou ces. A key challenge
lies in he inhe en ly pola ized na u e o UGC, whe e use s end o ei he s ongly
p aise o ha shly c i icize a b and o p oduc . To mi iga e he well-known J-shaped
dis ibu ion issue obse ed in online e iew a ings (Hydock e al. 2020), esea che s
mus explo e e ec i e s a egies o cap u ing neu al UGC as well. Finally, i is
impo an o ecognize ha UGC is suscep ible o manipula ion, such as ake e iews.
Fu u e esea ch should he e o e explo e me hods o de ec ing and emo ing in alid
o manipula i e con en om samples.
Rela edly, he quali y o UGC-based CMM measu es can be comp omised by
he p esence o non-o ganic con en , such as ake e iews o AI-gene a ed ex s
(e.g., Ko ács 2024). Fake e iews, o en gene a ed o p omo ional o malicious
pu poses, can in oduce signi ican biases in CMM measu es by dis o ing sen imen
dis ibu ions, misleading consume pe cep ion analyses, and educing he o e all
c edibili y o insigh s de i ed om UGC. Despi e e o s by pla o ms o de ec and
il e audulen con en , e ol ing echniques in au oma ed e iew gene a ion and
coo dina ed manipula ion campaigns con inue o pose challenges (Wu e al. 2020).
Simila ly, he ise o AI-gene a ed ex s in oduces ano he laye o complexi y.
Wi h ad ancemen s in na u al language gene a ion, AI sys ems can p oduce highly
con incing UGC, anging om p oduc e iews o social media pos s. While AI-
gene a ed con en can enhance engagemen and s eamline communica ion, i also
isks unde mining he alidi y o CMM measu es i such con en lacks genuine
consume in en o in oduces a i icial sen imen ends, pa icula ly i hese a e
being ein o ced by he euse o AI-gene a ed da a (Xing e al. 2025). To mi iga e
hese isks, u u e esea ch and p ac ice should ocus on de eloping obus il e ing
mechanisms, in eg a ing AI-powe ed de ec ion ools, and e ining alidi y con ols
o UGC-based CMM measu es.
While UGC is widely a ailable online, accessibili y emains a c i ical chal-
lenge o esea che s. Many pla o ms inc easingly es ic au oma ed da a e ie al
h ough hei e ms o se ice, ad anced echnical measu es, o limi ed API access
(Boege shausen e al. 2022), he eby seeking o main ain compe i i e ad an ages.
Fo ins ance, pla o ms like Amazon o Google employ ad anced measu es o p e-
en web sc aping. Public APIs, by con as , o e a mo e e icien al e na i e bu
a e o en expensi e, a e-limi ed, o subjec o leng hy applica ion p ocesses. Fo ex-
ample, X cha ges $ 5000 pe mon h o i s P o API o e ie e 1,000,000 pos s (X,
2025), while TikTok’s ee academic API is es ic ed o selec egions and ins i u-
ions (TikTok 2024) Such es ic ions may in ensi y in he u u e, posing signi ican
ba ie s o esea che s and disp opo iona ely impac ing smalle o mid-sized com-
panies ha lack he esou ces o ob ain p op ie a y da ase s o access expensi e
APIs. Re lec ing on hese dynamics, u u e esea ch should add ess he implica ions
o limi ed da a accessibili y and explo e al e na i e ways o ga he and u ilize UGC
da a.
As de Haan e al. (2024) and Küble (2023) highligh , esea ch has p o ided
a ich and expanding a ay o UGC in o ma ion ex ac ion ools. New me hods and
ools o en lead o classic “benchma king” s udies, which compa e he sui abili y
K
520 Schmalenbach Jou nal o Business Resea ch (2025) 77:497–525
o a new ool agains exis ing ools. This chain o esea ch becomes inc easingly
di icul o ollow due o he apid de elopmen o me hods. We sugges ha u u e
esea ch should ins ead ocus mo e on he aspec s o he di e en CMMs and assess
me hods on a me a-le el a he han solely on hei pe o mance inc eases. We
u he encou age u u e esea ch o shi i s ocus owa d iden i ying which ools
a e bes sui ed o speci ic CMM- ela ed asks. This includes no only assessing
he cos -bene i ade-o s o a ailable ools bu also de e mining whe he a single-
ool app oach (singula measu emen ) is su icien o i a combina ion o ools (i.e.,
a mul imodal app oach) would yield be e esul s.
Finally, while ou amewo k conside s impo an CMMs, u u e s udies should
ex end i o u he me ics ha ea u e p ominen ly in he li e a u e. These include,
bu a e no limi ed o measu es o co po a e epu a ion (Sa s ed e al. 2013)and
b and image (D iesne and Romaniuk 2006).
Despi e i s limi a ions, ou amewo k o e s a i s s ep in o le e aging UGC.
Resea che s and p ac i ione s can d aw on he a ious ools desc ibed in ou a icle
o pu he amewo k in o p ac ice.
Funding The au ho s did no ecei e suppo om any o ganiza ion o he submi ed wo k.
Con lic o in e es R.V. Küble , S. Adle , L. Welke, M. Sa s ed and K. Pauwels decla e ha hey ha e
no compe ing in e es s.
Open Access This a icle is licensed unde a C ea i e Commons A ibu ion 4.0 In e na ional License,
which pe mi s use, sha ing, adap a ion, dis ibu ion and ep oduc ion in any medium o o ma , as long as
you gi e app op ia e c edi o he o iginal au ho (s) and he sou ce, p o ide a link o he C ea i e Com-
mons licence, and indica e i changes we e made. The images o o he hi d pa y ma e ial in his a icle
a e included in he a icle’s C ea i e Commons licence, unless indica ed o he wise in a c edi line o he
ma e ial. I ma e ial is no included in he a icle’s C ea i e Commons licence and you in ended use is no
pe mi ed by s a u o y egula ion o exceeds he pe mi ed use, you will need o ob ain pe mission di ec ly
om he copy igh holde . To iew a copy o his licence, isi h p://c ea i ecommons.o g/licenses/by/4.
0/.
Re e ences
Adle , S., and M. Sa s ed . 2021. Mapping he jungle: A bibliome ic analysis o esea ch in o cons ual
le el heo y. Psychology & Ma ke ing 38(9):1367–1383.
Adle , S.J., M.K. Schönige , M. Lich e s, and M. Sa s ed . 2024. Fo y yea s o con ex e ec esea ch in
ma ke ing: A bibliome ic analysis. Jou nal o Business Economics 94(3):437–466.
Albe s, S. 2010. PLS and success ac o s udies in ma ke ing. In Handbook o pa ial leas squa es: Con-
cep s, me hods and applica ions in ma ke ing and ela ed ields Sp inge Handbooks o Compu a-
ional S a is ics Se ies, Vol. II, ed. V. Esposi o Vinzi, W.W. Chin, J. Hensele , and H. Wang, 409–425.
Be lin: Sp inge .
Anselmsson, J., and N. Bondesson. 2015. B and alue chain in p ac ise; he ela ionship be ween mindse
and ma ke pe o mance me ics: A s udy o he Swedish ma ke o FMCG. Jou nal o Re ailing and
Consume Se ices 25:58–70.
A is, M., S. Fo bes, and S. Fe guson. 2014. The b and pe sonali y o ocks. Ma ke ing Theo y 14(4):451–
475.
Babi´
c Rosa io, A., K. De Valck, and F. So giu. 2020. Concep ualizing he elec onic wo d-o -mou h p o-
cess: Wha we know and need o know abou eWOM c ea ion, exposu e, and e alua ion. Jou nal o
he Academy o Ma ke ing Science 48(3):422–448.
Baeh e, S., M. O’Dwye , L. O’Malley, and V.M. S o y. 2022. Cus ome mindse me ics: A sys ema ic
e alua ion o he ne p omo e sco e (NPS) s. al e na i e calcula ion me hods. Jou nal o Business
Resea ch 149:353–362.
K
