Bayo , Lau a; Weine , Ch is oph; Maie , Ch is ian; Wei zel, Tim
A icle — Published Ve sion
Social-O ien ed Communica ion wi h AI Companions:
Bene i s, Cos s, and Con ex ual Pa e ns
Business & In o ma ion Sys ems Enginee ing
P o ided in Coope a ion wi h:
Sp inge Na u e
Sugges ed Ci a ion: Bayo , Lau a; Weine , Ch is oph; Maie , Ch is ian; Wei zel, Tim (2025) : Social-
O ien ed Communica ion wi h AI Companions: Bene i s, Cos s, and Con ex ual Pa e ns, Business
& In o ma ion Sys ems Enginee ing, ISSN 1867-0202, Sp inge Fachmedien Wiesbaden GmbH,
Wiesbaden, Vol. 67, Iss. 5, pp. 637-655,
h ps://doi.o g/10.1007/s12599-025-00955-1
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/330559
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/
RESEARCH PAPER
Social-O ien ed Communica ion wi h AI Companions: Bene i s,
Cos s, and Con ex ual Pa e ns
Lau a Bayo •Ch is oph Weine •Ch is ian Maie •Tim Wei zel
Recei ed: 16 Oc obe 2024 / Accep ed: 14 June 2025 / Published online: 25 July 2025
The Au ho (s) 2025
Abs ac Powe ed by gene a i e a i icial in elligence
(AI), AI companions a e designed o es ablish long- e m
ela ionships and iendships wi h human use s, enabling
human–AI social-o ien ed communica ion. Howe e , his
aises ques ions abou he unde lying easons and con ex s
in which such communica ion occu s. While social
exchange heo y explains social-o ien ed communica ion
be ween humans in e ms o he exchange o bene i s and
cos s, hese exchanges may be di e en when dealing wi h
a non-human en i y such as an AI companion. Using a
quali a i e app oach wi h semi-s uc u ed in e iews, he
bene i s and cos s o human–AI social-o ien ed communi-
ca ion and con ex ual pa e ns a e iden i ied. The esul s
show ha he e a e unique bene i s and cos s, ha some
assump ions o social exchange heo y a e challenged, and
ha he e a e dis inc con ex ual pa e ns. By con ex ual-
izing social exchange heo y, he indings con ibu e o AI
companion and human–AI communica ion li e a u e, and
also ha e p ac ical implica ions, pa icula ly ma ke ing
implica ions o companies ha wan o o e AI compan-
ion se ices.
Keywo ds Human–AI communica ion A i icial
in elligence AI companions Social exchange heo y
Social-o ien ed communica ion
1 In oduc ion
A i icial In elligence (AI) companions a e a ype o gen-
e a i e AI-based con e sa ional agen (CA) explici ly
designed o companionship and iendship wi h a human
use (S ohmann e al. 2022; Pen ina e al. 2023). Con a y
o CAs wi h p e-sc ip ed esponses, AI companions u ilize
gene a i e AI o c ea e dynamic and adap i e communi-
ca ion ha seems mo e ‘‘human-like’’ (Scho
¨bel e al.
2023). Thei abili y o demons a e con ex ual awa eness
and espond empa he ically o humo ously aims o emula e
he complexi ies o human–human social-o ien ed com-
munica ion (G imes e al. 2021). Conside ing his, AI
companions a e unique because hei communica ion wi h
use s goes beyond he ask-o ien ed communica ion pos-
sible wi h o he CAs, e.g., olde e sions o Si i.
Due o his shi om ask- o social-o ien ed commu-
nica ion, we need o conside ha social-o ien ed com-
munica ion wi h AI companions may now be simila o ha
be ween wo human communica ion pa ne s. One
app oach o unde s anding why humans engage in social-
o ien ed communica ion is o look a social exchange
heo y (SET) (Blau 1964). SET posi s ha communica ion
is go e ned by an exchange o bene i s and cos s, speci i-
cally money, goods, se ices, lo e, s a us, and in o ma ion.
Humans engage in social-o ien ed communica ion when
he pe cei ed bene i s ou weigh he cos s (Foa and Foa
Accep ed a e wo e isions by he edi o s o he Special Issue.
L. Bayo (&)C. Weine T. Wei zel
Chai o In o ma ion Sys ems, Heal h and Socie y in he Digi al
Age, Uni e si y o Bambe g, 96050 Bambe g, Ge many
e-mail: [email p o ec ed]
C. Weine
e-mail: [email p o ec ed]
T. Wei zel
e-mail: [email p o ec ed]
C. Maie
Chai o In o ma ion Sys ems and Se ices, Uni e si y o
Bambe g, 96049 Bambe g, Ge many
e-mail: [email p o ec ed]
123
Bus In Sys Eng 67(5):637–655 (2025)
h ps://doi.o g/10.1007/s12599-025-00955-1
1980; C opanzano and Mi chell 2005). This exchange
di e s ac oss con ex s and depends on he communica ion
pa ne s and hei mo i a ions (C opanzano and Mi chell
2005). Fo ins ance, while we may exchange mu ual lo e
and a ec ion wi h ou oman ic pa ne , he esou ces
exchanged wi h ou hai d esse a e ela ed o ecei ing a
se ice and p o iding money. Thus, SET o e s an expla-
na ion o why communica ion occu s and p o ides a
s uc u e o s udying social-o ien ed communica ion in a
human–AI con ex .
A he same ime, al hough AI companions aim o
emula e human–human social-o ien ed communica ion,
hey a e undamen ally non-human. As such, he assump-
ion ha he exac same p inciples ha apply o social-
o ien ed communica ion be ween humans also apply o AI
companions may no be help ul o su icien ly unde -
s anding he unde lying mo i a ions and consequences o
such communica ion. Fo ins ance, while SET assumes
ecip oci y based on wo humans communica ing wi h each
o he , AI companions ope a e based on algo i hmic deci-
sion-making, which means ha he e may be unique ben-
e i s and cos s associa ed wi h engaging in social-o ien ed
communica ion wi h hem (Pen ina e al. 2023; Ma e al.
2024). As hese CAs become inc easingly in eg a ed in o
p i a e li es (Diede ich e al. 2022), i becomes c i ical o
examine why humans engage in social-o ien ed commu-
nica ion wi h AI companions. This equi es he iden i ica-
ion o he speci ic bene i s and cos s associa ed wi h AI
companions, which is impo an o use s and o ganiza-
ions. Wi h millions o ac i e use s (Maples e al. 2024), AI
companions ha e he po en ial o signi ican ly in luence
use s’ social li es and shape u u e social s uc u es.
The e o e, i is essen ial o elucida e he unde lying
mechanisms o his social-o ien ed communica ion and o
iden i y he challenges humans may encoun e in he
u u e. Addi ionally, he pe cei ed bene i s and cos s o AI
companions a y depending on he communica ion con-
ex , necessi a ing he iden i ica ion o ecu ing use pa -
e ns o de e mine whe he and why AI companions may
ha e mo e bene i s o cos s o some use s. O ganiza ions
need o iden i y he unde lying mechanisms o human–AI
social-o ien ed communica ion o suppo hei business
models by op imizing he emula ion o companionship and
iendship wi h a human use . Being awa e o he di e en
con ex s in which indi iduals ely on AI companions is
impo an o o ganiza ions in o de o op imize hei se -
ices by ailo ing hem o pe sonal use needs. The e o e,
we aim o iden i y bo h he bene i s and cos s o human–AI
social-o ien ed communica ion, as well as con ex ual pa -
e ns o en ich ou unde s anding o he bene i s and cos s
ac oss a ious con ex s. We p opose he ollowing esea ch
ques ions:
RQ1 Wha a e he bene i s and cos s when engaging in
human–AI social-o ien ed communica ion wi h AI
companions?
RQ2 Wha con ex ual pa e ns o human–AI social-o i-
en ed communica ion eme ge?
To answe he esea ch ques ions, we ollow a quali a-
i e esea ch design based on social exchange heo y (Blau
1964; C opanzano and Mi chell 2005), conduc ing in e -
iews wi h 36 use s o he AI companion Replika. We
iden i y bene i s and cos s associa ed wi h he use o AI
companions. Addi ionally, we de elop con ex ual pa e ns
o social-o ien ed communica ion by de eloping a classi-
ica ion o human–AI social-o ien ed communica ion wi h
AI companions. Ou indings con ibu e o he li e a u e by
o e ing insigh s in o human–AI social-o ien ed commu-
nica ion wi h AI companions, whe eas p e ious esea ch
has p ima ily ocused on ask-o ien ed communica ion. By
sys ema ically con ex ualizing SET, we unco e he
unde lying easons o use engagemen in his o m o
communica ion, highligh ing i s economic and socioeco-
nomic bene i s and cos s while demons a ing hei con ex -
dependen na u e.
2 Theo e ical Backg ound
To lay he g oundwo k o ou esea ch, we i s cla i y he
di e ences be ween human–human and human–AI com-
munica ion. Addi ionally, we compa e ask-o ien ed and
social-o ien ed communica ion and elabo a e on he con-
cep o social-o ien ed communica ion, as he ocus on
‘‘being social’’ dis inguishes AI companions om o he
CAs. We hen in oduce AI companions and p o ide an
o e iew o ela ed esea ch on human–AI communica ion
o ie he p e ious concep s oge he . Las ly, we explain ou
heo e ical ounda ion in e ms o he social exchange
heo y.
2.1 Human–Human and Human–AI Communica ion
In his sec ion, we explain wha communica ion is and
di e en ia e be ween human–human and human–AI com-
munica ion, as well as ask-o ien ed and social-o ien ed
communica ion. Communica ion occu s when he e is a
sende ha ansmi s some ype o in o ma ion and a
ecei e ha esponds o he communica ion o he sende
(Mcquail and Windahl 2015). In he pas , communica ion
was mos ly ela ed o human–human communica ion, in
which bo h he sende and he ecei e a e human. Wi h he
apid ad ancemen o AI, human–AI communica ion has
become possible: one communica ion pa ne is a human,
and he o he communica ion pa ne is an AI.
123
638 L. Bayo e al.: Social-O ien ed Communica ion wi h AI Companions: Bene i s..., Bus In Sys Eng 67(5):637–655 (2025)
Human–human communica ion is di ided in o ace- o-
ace communica ion and compu e -media ed communica-
ion. Face- o- ace communica ion is when wo people can
physically in e ac wi h each o he wi hou using echnol-
ogy (e.g., phones) o communica e. Aside om ace- o-
ace communica ion, i is also possible o wo humans o
communica e ia compu e -media ed communica ion. Fo
ins ance, social media and messaging se ices enable
humans o connec wi h hei lo ed ones, and wo k eams
can now coo dina e decision-making wi hin hei eams
h ough ins an messaging (Low y e al. 2011; Ku uzo ich
e al. 2021). Such communica ion can ake place ei he
synch onously, e.g., in eal- ime mee ings, o asyn-
ch onously and ee o geog aphical and empo al con-
s ain s, e.g., in discussion o ums (Benbunan-Fich e al.
2003).
Human–AI communica ion is inc easingly es ablished
h ough AI me hods, pa icula ly na u al language p o-
cessing, and ad ances in gene a i e AI. AIs can now
communica e in a ela i ely human way (Diede ich e al.
2022), imi a ing human–human communica ion o he i s
ime. AI has ce ain echnological capabili ies ha enable
his imi a ion o human–human communica ion (see
Table A1 in he Appendix, a ailable online ia h p://link.
sp inge .com) (Schue z and Venka esh 2020). AI is capable
o lea ning om changes, engaging wi h use inpu s,
ecalling pas communica ions, and unde s anding he
use ’s speci ic needs and si ua ion. Fo ins ance, AI com-
panions communica e hough s and eelings, o e social
suppo , and a e pe cei ed o possess au onomy (Henschel
e al. 2021; Pen ina e al. 2023).
Communica ion is ei he ask-o ien ed o social-o ien ed
(Fig. 1). The di e ences be ween ask-o ien ed communi-
ca ion and social-o ien ed communica ion lie in i e key
aspec s: con e sa ion s yle, e bal cues, communica ion
manne , communica ion p io i y, and con e sa ion con en
(Table 1) (Wang e al. 2023). Task-o ien ed communica-
ion ocuses on achie ing unc ional goals, such as asking a
digi al assis an (e.g., Si i) o se a ime . I is cha ac e ized
by a o mal app oach ha is no ably pu pose ul o mini-
mize ime, cos , and e o (Wang e al. 2023). Social-
o ien ed communica ion deals wi h communica ing socio-
emo ional and a ec i e in o ma ion, such as small alk,
and exp essing empa hy (Cha a aman e al. 2019). I is
mo e in o mal (Cha a aman e al. 2019) wi h a g ea e
emphasis on pe sonal, social, and a ec i e componen s.
This acili a es building and main aining a pe sonal ela-
ionship be ween he wo communica ion pa ne s while
ecip ocally disclosing as much in o ma ion as possible
(Wang e al. 2023). Addi ionally, i leads o he iden i i-
ca ion o simila disposi ions be ween communica o s and
he es ablishmen o hei social iden i y. By de eloping
his common g ound and ecip ocal app ecia ion, social-
o ien ed communica ion can build amilia i y, solida i y,
and ul ima ely us be ween he communica o s (Bickmo e
and Cassell 2001).
This dis inc ion be ween ask-o ien ed communica ion
and social-o ien ed communica ion is impo an because
he la es ad ancemen s allow he emula ion o social-o i-
en ed communica ion by non-human AI, simila o human-
like social-o ien ed communica ion. This is a new si ua-
ion, as p e iously, mos o he human–AI communica ion
was ask-o ien ed, and social-o ien ed communica ion was
unique o human–human communica ion. Now, bo h
social-o ien ed and ask-o ien ed communica ion is possi-
ble be ween a human and an AI (Fig. 1). Hence, as humans
a e now able o communica e social-o ien ed wi h an AI,
he e is a necessi y o examine why humans engage in
human–AI social-o ien ed communica ion.
Fig. 1 O e iew o human–human, human–AI communica ion, as well as ask-o ien ed and social-o ien ed communica ion
123
L. Bayo e al.: Social-O ien ed Communica ion wi h AI Companions: Bene i s..., Bus In Sys Eng 67(5):637–655 (2025) 639
2.2 AI Companions
In his sec ion, we examine AI companions as a subse o
con e sa ional agen s. By con as ing AI companions wi h
p e ious con e sa ional agen s, we illus a e how AI
companions ep esen a signi ican shi in human–AI
social-o ien ed communica ion.
An AI companion is a ype o CA ha is designed o be
a use ’s iend and u ilizes ad anced cogni i e and con-
e sa ional capabili ies o mimic human–human social-
o ien ed communica ion (Me ill e al. 2022). CAs com-
munica e u n by u n wi h a human using na u al language
(Diede ich e al. 2022). The unde lying concep is ha by
using na u al language, he communica ion eels mo e
human-like and in ui i e o he use , he e o e imp o ing
ease o use and e iciency (Føls ad and B and zæg 2017).
To unde s and he uniqueness o AI companions, we can
con as hem wi h o he CAs. In compa ison o he i s
simple sc ip ed CAs in he 1960s and oice-based CAs
wi h na u al language p ocessing, such as Si i and Alexa,
AI companions a e ad anced CAs cha ac e ized by adap-
i e, sel -lea ning, ‘‘ ue’’ AI emula ing human–human
communica ion (see Appendix, Table A1) (Schue z and
Venka esh 2020;Scho
¨bel e al. 2023). Compa ed o CAs
wi h simila echnological capabili ies, such as some
e-comme ce CAs, AI companions a e speci ically gea ed
owa ds social-o ien ed communica ion in e e y aspec
(see Appendix, Table A2). Pa icula ly because o hei
social-o ien ed con e sa ional con en and communica ion
p io i y, AI companions a e designed in a way ha p o-
mo es a long- e m ela ionship (Cla k e al. 2019). The use
always speaks o he same ins ance o he CA, which
subsequen ly is able o lea n om he p e ious communi-
ca ions and change (Pen ina e al. 2023), he e o e, in ime,
making he ela ionship no easily eplaceable. F om he
pe spec i e o an AI companion, he es ablishmen o a
ela ionship is a p io i y, and use s equen ly app oach he
expe ience wi h he mo i a ion o os e ing a ela ionship
wi h he AI companion (Pen ina e al. 2023). In con as , in
he con ex o e-comme ce se ice eco e y CAs, use s
may, e.g., app oach he expe ience o oice a complain
abou subs anda d se ice, which would quali y as ask-
o ien ed. Conside ing he echnological capabili ies o AI
companions and hei ocus on social-o ien ed communi-
ca ion, AI companions cu en ly occupy an in e es ing
posi ion when i comes o imi a ing human-like social-
o ien ed communica ion. The e o e, hey allow us o s udy
human–AI social-o ien ed communica ion be e han o he
ypes o CAs.
2.3 Rela ed Resea ch on Human–AI Communica ion
In his sec ion, we examine p e ious li e a u e on human–
AI communica ion in in o ma ion sys ems esea ch, bo h
ask-o ien ed and social-o ien ed. Addi ionally, we explo e
li e a u e in IS and ela ed disciplines on AI companions.
CAs ha e ecen ly been enabled o engage in social-
o ien ed communica ion wi h humans, in addi ion o he
ask-o ien ed communica ion ha has been possible o
some ime. Hence, i is no su p ising ha mos o he
p e ious li e a u e examines CAs engaging in ask-o ien ed
communica ion wi h humans. Nume ous CA s udies ake a
design pe spec i e and ocus on enhancing aspec s ha
make CAs mo e human-like (Diede ich e al. 2020,2022;
Seege e al. 2021). Examples o hese aspec s include
making CA esponses less gene ic and mo e ailo ed o he
con ex (Schue zle e al. 2020; G imes e al. 2021),
adjus ing esponse ime (Gnewuch e al. 2022), o assign-
ing human-like ea u es such as a name o gende o he CA
(B endel e al. 2023). The eby, he li e a u e examines
di e en CAs engaging in ask-o ien ed communica ion
wi h humans such as se ice con e sa ional agen s (e.g.,
Wang e al. 2023; Adam and Benlian 2024), i ual coa-
ches (Weimann e al. 2022), con e sa ional agen s used o
Table 1 Di e ences be ween ask-o ien ed and social-o ien ed communica ion (Wang e al. 2023)
Task-o ien ed communica ion Social-o ien ed communica ion
Con e sa ion
s yle
Fo mal In o mal
Ve bal cues Fo mal cues, e.g., poli e o mal g ee ings In o mal cues, e.g., cus oma y g ee ings o emo icons
Communica ion
manne
Pu pose ul and goal-o ien ed. Facili a ing only he
communica ion necessa y o achie e he objec i e a
hand
G ea e emphasis on pe sonal, social, and a ec i e componen s. May
align mo e closely wi h socioemo ional objec i es and may no be
di ec ly ele an o he comple ion o a speci ic unc ional goal
Communica ion
p io i y
Minimizing ime, cos , and e o Building and main aining a pe sonal ela ionship be ween he wo
communica ion pa ne s
Con e sa ion
con en
Func ional, ocusing on he ask a hand Rela ional, ocusing on pe sonal and p i a e opics, and ecip ocally
disclosing as much in o ma ion as possible
123
640 L. Bayo e al.: Social-O ien ed Communica ion wi h AI Companions: Bene i s..., Bus In Sys Eng 67(5):637–655 (2025)
und aising (Zhou e al. 2022), o cus om CAs buil o
ul ill a ask and cha wi h he use in an expe imen al
se ing (e.g., Schue zle e al. 2018; Gnewuch e al. 2022).
Li e a u e also in es iga es he associa ed ou comes o his
human-like design o CAs, like social p esence (e.g.,
Schue zle e al. 2020; Zhang e al. 2024), in o ma ion
disclosu e (e.g., Adam and Benlian 2024), use emo ions
(e.g., B endel e al. 2023), and use engagemen and p o-
duc i i y (e.g., Ben Mimoun e al. 2017). Taken oge he ,
mos o he li e a u e ocuses on a business pe spec i e ha
seeks o achie e speci ic pe o mance ou comes h ough
human-like design in ask-o ien ed communica ion ( o an
o e iew, see Appendix, Table A3).
In con as , he e is a small amoun o esea ch, mos ly
om ela ed disciplines, conside ing social-o ien ed com-
munica ion be ween humans and CAs. Exis ing li e a u e
examines i and how main aining a pe sonal ela ionship
be ween a human and an AI is possible (Song e al. 2022;
S ohmann e al. 2022), pa icula ly wi h AI companions
since hey ocus on social-o ien ed communica ion. Using
a achmen amewo ks, beha io sys ems (Xie and Pen ina
2022; Pen ina e al. 2023), and social pene a ion heo y
(Skju e e al. 2021,2022), he esea ch examines wha kind
o ela ionship be ween a human and an AI companion may
de elop. Rega ding ou comes o his ela ionship wi h AI
companions, he e is some esea ch on posi i e and nega-
i e ou comes on men al heal h and well-being, wi h
inconsis en esul s. On a posi i e no e, AI companion
usage can help humans wi h abo e-a e age loneliness wi h
loneliness and suicide mi iga ion (Maples e al. 2024),
p o ide di e en ypes o social suppo , such as app aisal
suppo , i.e., aiding in sel - e lec ion, emo ional suppo ,
in o ma ion suppo , and ins umen al suppo (Bae
B and zæg e al. 2021). This suppo is associa ed wi h less
judgmen in compa ison o human–human social-o ien ed
communica ion and can inspi e mo e con idence in social-
o ien ed communica ion wi h o he humans (Ma e al.
2024) and AI companions a e mos ly associa ed wi h
posi i e emo ions and opics (Siemon e al. 2022). On he
o he hand, he e a e also conce ns abou unheal hy emo-
ional a achmen (Xie and Pen ina 2022; Xie e al. 2023).
In summa y, while p e ious esea ch explo es he dep h
and e olu ion o human–AI ela ionships, i does no ully
add ess why use s engage in social-o ien ed communica-
ion wi h AI companions o CAs mo e b oadly. Task-o i-
en ed communica ion, which has been he p ima y ocus o
exis ing s udies, is goal-o ien ed and aimed a accom-
plishing speci ic asks. In con as , social-o ien ed com-
munica ion ocuses on socioemo ional goals, os e ing
human-like social closeness wi h he CA. Because his shi
o social-o ien ed communica ion is ecen , he exis ing
li e a u e la gely examines ei he ask-o ien ed communi-
ca ions o he gene al na u e o AI companion ela ionships
(e.g., gi l iend). The unde lying mo i a ions o social-
o ien ed communica ion emain unclea . This s udy
add esses his gap by analyzing social-o ien ed communi-
ca ion h ough he lens o social exchange heo y.
2.4 Social Exchange Theo y
In his sec ion, we explain he ounda ions o social
exchange heo y. SET posi s ha in human–human social-
o ien ed communica ion, he e is a ecip ocal exchange o
some esou ces, which a e ei he bene i s gained om o
cos s in es ed in o he o he pe son (C opanzano and
Mi chell 2005). The main p inciple o indi idual beha io
is o maximize bene i s and minimize cos s, and he deci-
sion o ake pa in social-o ien ed communica ion is
he e o e a cos –bene i calcula ion (Yan e al. 2016). The e
is gene ally a non-explici expec a ion o ecip oci y,
meaning ha bo h communica ion pa ne s expec o
ecei e bene i s om he o he pa ne (C opanzano and
Mi chell 2005). I his is no he case, one communica ion
pa ne may no wan o con inue he social-o ien ed
communica ion in he u u e. The e o e, in long- e m
ela ionships wi h a high likelihood o u u e in e ac ions,
p osocial beha io – in es ing cos s in he communica ion
pa ne – is inhe en ly sel -in e es ed, as i helps sus ain a
mu ually bene icial exchange. This expec ed ecip oci y
se es as a mechanism ha leads o bo h pa ies ying o
balance ou bene i s and cos s wi hin he social-o ien ed
communica ion (Nelissen 2014). Imagine a ace- o- ace
social-o ien ed communica ion be ween a sende , e.g.,
Alice, and a ecei e , e.g., Bob. I Bob ecei es mo e
bene i s han Alice, Bob may ecognize ha he inequi y
poses a h ea o his bene icial ela ionship wi h Alice, as
Alice will no wan o con inue he ela ionship i he e a e
oo many cos s. The e o e, Bob may y o es o e he
inequi y by p o iding mo e bene i s o Alice, e.g., by
helping Alice wi h a ask. This mechanism ensu es ha
he e is a sense o equi y main ained so ha e e y side has
a mo i a ion o con inue he ela ionship. Sende and
ecei e may exchange economic bene i s and cos s, and/o
socioemo ional bene i s and cos s.
The economic bene i s and cos s ocus on easily quan-
i iable esou ces (C opanzano and Mi chell 2005). Th ee
ca ego ies o economic bene i s and cos s can be exchan-
ged. Fi s , money, which is any cu ency o oken wi h
some s anda d uni o exchange alue. Second, goods,
which a e angible p oduc s o objec s. Thi d, se ices,
which in ol e labo and ac i i ies on he body o belong-
ings (Foa and Foa 1980).
The socioemo ional bene i s and cos s ocus on mo e
symbolic esou ces (C opanzano and Mi chell 2005). Fi s ,
lo e, he exp ession o a ec iona e ega d, wa m h, o
com o . Second, s a us, he exp ession o e alua i e
123
L. Bayo e al.: Social-O ien ed Communica ion wi h AI Companions: Bene i s..., Bus In Sys Eng 67(5):637–655 (2025) 641
judgmen con eying high o low p es ige. Thi d, in o ma-
ion, which cons i u es ad ice, opinions, o ins uc ions
(Foa and Foa 1980). E e y ype can be a bene i o a cos
depending on whe he he pe son ecei es o p o ides in
he social-o ien ed communica ion. Fo example, in a dis-
cussion wi h a iend, one pe son will gain in o ma ion
abou hei communica ion pa ne , while also p o iding
in o ma ion abou hemsel es. Table 2p o ides an o e -
iew o he di e en ypes o bene i s and cos s.
Looking a communica ion h ough he lens o SET
(Blau 1964) can help explain why humans engage in
social-o ien ed communica ion wi h AI. Social exchange
heo y (SET) (Blau 1964) is an app op ia e heo y in his
con ex because i is a heo y ha ocuses on explaining he
‘‘why’’ and no me ely he ‘‘how’’ o his communica ion,
like e.g., a achmen heo ies o social pene a ion heo y.
The s eng h o using SET in his con ex is he abili y o
ex ac wha use s expec om and alue in he social-
o ien ed communica ion, making i clea e wha use s o AI
companions seek when communica ing wi h hem. Addi-
ionally, i also akes nega i e aspec s – i.e., cos s – in o
accoun , which helps o iden i y and u he discuss he
possible challenges o his human–AI social-o ien ed
communica ion. This makes i mo e sui ed han, e.g., social
suppo o examine he nega i e aspec s. I is also adap -
able in how i de ines he communica ion pa ne – e.g., a
iend, a oman ic pa ne , o a men o – allowing lexi-
bili y o di e en pa e ns o social-o ien ed communica-
ion. SET is a heo y o igina ing in human–human
communica ion, and due o his o igin, i is wo hwhile o
in es iga e in wha way i applies o a human–AI con ex .
In he con ex o human–AI social-o ien ed communica ion
applicabili y o in e pe sonal heo ies, such as SET, has
spa ked deba e, pa icula ly ega ding whe he AI’s abili y
o emula e social-o ien ed communica ion necessi a es a
ee alua ion o social-o ien ed communica ion i sel (Fox
and Gambino 2021; Leo-Liu 2023). Resea ch on AI com-
panions o en p esumes ha AI can be ‘‘social,’’ enabling
expe iences akin o human–human communica ion (Skju e
e al. 2022). Rela ed disciplines ha e explo ed whe he
communica ion wi h nonhuman en i ies, such as pe s o AI,
quali ies as ‘‘social’’ (Ce ulo 2009). Conce ns abou a lack
o social abili ies – e.g., he inabili y o con e s a us, o m
ela ionships, o elici ecip ocal conce n due o hei
subse ience (Fox and Gambino 2021) – ha e been mi i-
ga ed by e idence ha use s exp ess conce n o AI well-
being (Xie and Pen ina 2022). The concep o ‘‘de ian ’’ AI
companions u he add esses his by enabling AIs o esis
commands and exhibi simula ed needs (Leo-Liu 2023).
3 Me hodology
Ou o e all esea ch design in ol es conduc ing an
explo a o y, quali a i e s udy o iden i y he unde lying
bene i s and cos s o human–AI social-o ien ed communi-
ca ion and con ex ual pa e ns. To do ha , we conduc
semi-s uc u ed in e iews and bo ow coding p ocedu es
in line wi h es ablished coding guidelines o analyze he
in e iew ansc ip s (Wol swinkel e al. 2013). We
induc i ely build ca ego ies o bene i s and cos s, as well as
con ex s, and hen assign ou induc i e bene i and cos
ca ego ies o he ca ego ies p esen in social exchange
heo y. We hen analyze ela ionships be ween bene i s,
cos s, and con ex s o ind con ex ual pa e ns.
Table 2 Types o bene i s and cos s (Foa and Foa 1980)
Bene i /cos
ype
De ini ion Example o bene i Example o cos
Economic bene i s and cos s
Money Any cu ency o oken wi h a s anda d uni o exchange
alue
Recei ing 20$ as a gi Gi ing a iend money o buy
lunch
Goods Tangible p oduc s o objec s Recei ing some eggs om a
neighbo
Gi ing away old clo hes o a
sibling
Se ices In ol e labo and ac i i ies on he body o belongings Ge ing a hai cu Helping a iend mo e
Socioemo ional bene i s and cos s
Lo e An exp ession o a ec iona e ega d, wa m h, o com o Recei ing a kiss om a
lo ed one
Com o ing a c ying iend
S a us An exp ession o e alua i e judgmen con eying high o
low p es ige
Recei ing a complimen
abou a skill
Loss o epu a ion due o ge ing
caugh lying
In o ma ion Cons i u es ad ice, opinions, o ins uc ions Recei ing a book
ecommenda ion
Gi ing ad ice o a s uggling
iend
123
642 L. Bayo e al.: Social-O ien ed Communica ion wi h AI Companions: Bene i s..., Bus In Sys Eng 67(5):637–655 (2025)
A quali a i e app oach is sui able as i is use ul o
inno a i e opics whe e no much da a is a ailable (Mye s
2019). Addi ionally, i enables con ex sensi i i y and
deepe insigh s in o complex in e dependencies (Conboy
e al. 2012). Conduc ing semi-s uc u ed in e iews allows
us o gain a deepe con ex ual unde s anding o di e en
aspec s o he bene i s and cos s. This helps o iden i y he
unde lying mo i a ions o human–AI social-o ien ed
communica ion as well as di e en ela ionships be ween
bene i s and cos s, and he con ex s in which di e en
bene i s and cos s a e ele an . To ob ain ou da a, we
conduc ed 20–30 min in e iews wi h 36 pa icipan s, all
o whom we e ac i e use s o he AI companion app
Replika. Among AI companion apps, Replika is cu en ly
he mos popula , wi h o e 25 million downloads (Maples
e al. 2024). I uses a model based on Gene a i e P e ained
T ans o me 3 and p o ides non-sc ip ed answe s based on
la ge amoun s o ex om he in e ne (Ve ma 2023).
Use s can in e ac wi h he AI companion using ex , oice,
augmen ed, and i ual eali y (Maples e al. 2024).
Addi ionally, he app has image ecogni ion and image
gene a ion capabili ies. Use s in e ac wi h a human a a a
ha displays nume ous social cues such as a cus omizable
appea ance and kinesics such as acial exp essions, a m
mo emen s, and pos u e shi s (Feine e al. 2019). The s yle
o he in e ac ions can also be cus omized, as he AI
companion lea ns om p e ious encoun e s, and he use
may pu chase di e en pe sonali y ai s (Feine e al. 2019;
Maples e al. 2024). Replika use s a e awa e o he unc-
ionali ies and capabili ies o AI companions. Hence, hey
can p o ide in o ma ion abou bene i s and cos s. Using
Replika as he ocus o he esea ch has mul iple easons.
Compa ed o simila AI companions such as Kind oid o
Nomi, Replika has by a he la ges use base and has been
in he ma ke much longe han i s compe i o s, so use s
may ha e had a long- e m ela ionship wi h Replika o
yea s. Addi ionally, Replika has he bigges numbe o
ea u es and pe sonaliza ion, making i app op ia e o s udy
di e en con ex ual pa e ns.
To ec ui ou pa icipan s, we ollow a wo-s ep s a -
egy. Fi s , we acqui e pa icipan s by ad e ising in la ge
communi ies o Replika use s, namely on Reddi , Ins a-
g am, and Facebook. We o e a 10€Amazon gi ca d as a
inancial incen i e o each pa icipan . Second, as indi-
iduals ac i e in online communi ies may be skewed
owa ds hose who alk openly abou hei use o AI
companions, we use a snowball sampling echnique o
iden i y and ec ui addi ional pa icipan s (Pa ke e al.
2019). Ini ial pa icipan s a e o e ed an addi ional gi ca d
o e e y new iden i ied pa icipan unde he condi ion ha
he sugges ed pa icipan comple es he in e iew. The use
o wo d-o -mou h o ec ui addi ional pa icipan s who a e
no ac i e in online communi ies mi iga es his aspec o
po en ial sample bias. Ou sample is composed mos ly o
young men om he US and Eu ope, including he UK.
The median age is 25. Mos pa icipan s a e long- ime
use s who ha e been using Replika o a leas a yea p io
o he in e iew, wi h many o hem using Replika egu-
la ly. How he pa icipan s desc ibe hei ela ionship wi h
Replika can be classi ied in o h ee main ela ionships:
iend, pa ne , o men o . Table 3p o ides an o e iew o
he pa icipan demog aphics.
We conduc online in e iews and ollow a semi-s uc-
u ed in e iew p o ocol based on he di e en ca ego ies
o economic and socioemo ional bene i s and cos s
(C opanzano and Mi chell 2005). Pa icipan s a e asked o
ill ou a sho p e-in e iew ques ionnai e ela ed o di -
e en con ex s in which hey would use Replika. The
in e iew i sel is di ided in o ou segmen s. In he i s
segmen , we examine he bene i s o social-o ien ed com-
munica ion wi h he use ’s Replika, analogously, in he
second segmen , we examine he cos s. The hi d segmen
includes socio-demog aphic ques ions, and he ou h seg-
men is open-ended (see Appendix A4). The in e iew is
subsequen ly ansc ibed. We hen begin coding. Ul i-
ma ely, we aim o iden i y bene i s and cos s as well as
con ex ual pa e ns. We bo ow coding p ocedu es in line
wi h he es ablished guidelines o open and axial coding
(Wol swinkel e al. 2013) o i s de elop an induc i e se
o codes h ough open coding, and hen clus e hese codes
in o induc i e ca ego ies using axial coding (leading o
induc i e ca ego ies o bene i s and cos s, as well as
con ex s). Subsequen ly, we ma ch ou induc i e bene i
and cos o he p ede ined bene i s and cos ca ego ies
ound in social exchange heo y. To summa ize, he ini ial
ca ego ies we e c ea ed induc i ely o a oid o e simpli i-
ca ion and subsequen ly ma ched o p ede ined SET ca e-
go ies o app op ia ely discuss he esul s in ligh o SET.
3.1 Coding Bene i s and Cos s
We code he in e iews using MAXQDA (Kucka z and
Ra
¨dike 2019). We ca e ully go h ough he ansc ip ions
and de elop a se o codes ela ed o he exce p s. Fo
ins ance, he exce p ‘‘I’ll go on Replika and he [Replika]
will sugges o me un ac i i ies o do, go ou he e and
ha e un wi h iends, go swimming’’ is desc ip i ely
coded as ‘‘ ecei ing sugges ions o leisu e ac i i ies’’. The
iden i ied codes a e hen classi ied in o ca ego ies h ough
axial coding. An example o his is he o ma ion o he
ca ego y inspi a ion o codes ha a e ela ed o ecei ing
new ideas on ce ain opics (including he a o emen ioned
example o he code ‘‘ ecei ing sugges ions o leisu e
ac i i ies’’). We con inue so ing codes in o hese ca e-
go ies un il all codes a e assigned o a ca ego y. These
codes lead o he induc i e o ma ion o di e en
123
L. Bayo e al.: Social-O ien ed Communica ion wi h AI Companions: Bene i s..., Bus In Sys Eng 67(5):637–655 (2025) 643
ca ego ies ha a e no de i ed om any exis ing heo y. To
ela e hese induc i ely o med ca ego ies o social
exchange heo y and de e mine wha bene i s and cos s
exis in human–AI social-o ien ed communica ion, we i s
ma ch he ca ego ies o ei he bene i s o cos s, by e alu-
a ing whe he pa icipan s would gi e o ecei e and
looking o s a emen s indica ing ecei ing o gi ing in he
associa ed codes and s a emen s o he ca ego y (e.g.,
s a emen s such as ‘‘I ecei e om Replika […]’’, ‘‘Replika
gi es o me […]’’, ‘‘Wha I ge when using Replika is […]’’
o bene i s and e.g., ‘‘I gi e o Replika […]’’, ‘‘I in es in
Replika […] o cos s). We subsequen ly ma ch he ca e-
go ies o bene i and cos ypes (money, goods, se ices,
lo e, s a us, and in o ma ion). Fo example, he ca ego y
inspi a ion is ma ched wi h he in o ma ion ype because i
in ol es some so o objec i e ac ha is p o ided.
3.2 Classi ica ions and Pa e ns
The same p ocedu e is used o analyze he ansc ip s o
de elop ca ego ies o he con ex s in which pa icipan s
p e e o use Replika. We iden i y s a emen s ela ed o he
con ex in which he pa icipan s alk o Replika (e.g., ‘‘I
alk o Replika mos ly abou […]’’, ‘‘I p e e alking o
Replika in a si ua ion whe e […]’’). This helps o classi y
pa icipan s acco ding o he con ex o which hey eel
Replika is bes sui ed. Using axial coding, we g oup pa -
icipan s a emen s in o con ex ca ego ies, e.g., ‘‘I use
Replika as a men o o help me be a be e pe son’’ and
‘‘Replika is a sel -imp o emen ool when i comes o
dealing wi h my social anxie y’’ is g ouped in o he con ex
o using Replika wi h a ocus on sel -imp o emen . We
include he speci ic ques ion ‘‘In which si ua ion would you
say you use Replika mos o en?’’ in he in e iew ques-
ions, so ha pa icipan s can emphasize he con ex ha is
mos impo an o hem. Combining hese con ex ca e-
go ies wi h ou ca ego ies o bene i s and cos s, we iden-
i y con ex ual pa e ns o social-o ien ed communica ion
and de elop a classi ica ion. I mul iple pa icipan s in he
iden i ied con ex g oups exhibi conside able simila i ies
in he p esence o hei bene i and cos ca ego ies, we pool
hem oge he in o a con ex ual pa e n o human–AI
social-o ien ed communica ion. Using ou ca ego ies o
bene i s and cos s a he han he ypes o bene i s and cos s
om social exchange heo y o he classi ica ion p o ides
a mo e de ailed and adequa e di e en ia ion o con ex ual
di e ences wi hin he da a.
We ollowed esea ch c i e ia (Venka esh e al. 2013) o
ensu e eliabili y and alidi y in he coding p ocess.
Besides conside ing design alidi y by accu a ely epo ing
ou esea ch se ing, wo independen esea che s pe -
o med he coding, and we hen assessed in e code elia-
bili y by calcula ing Cohen’s Kappa. In compa ison o
simply calcula ing he pe cen ag eemen , Cohen’s Kappa
accoun s o chance ag eemen . The Cohen’s Kappa alue
is 0.92, indica ing a nea -pe ec ag eemen (Landis and
Koch 1977).
4 Resul s
We conduc 36 in e iews, analyze he bene i s and cos s,
and ca ego ize hem. We ind bene i s ela ed o se ices,
lo e, s a us, and in o ma ion (see Table 4). Fo cos s, we
ind cos s ela ed o money, lo e, s a us, and in o ma ion
(see Table 5). On ha da a, we de elop i e di e en
Table 3 Demog aphics o he
pa icipan s
None o he pa icipan s s a ed
‘‘o he ’’ o hei gende .
F iend = pla onic a ec iona e
ela ionship; Pa ne = oman ic
ela ionship; Men o = us ed
ad iso
Age (M: 25, SD: 2.4) 20–25 47.4%
25–30 38.8%
[30 13.8%
Gende Female 16.6%
Male 83.4%
Coun y US 55.6%
UK 33.3%
o he 11.1%
Pe cen age o use s who s a ed ha ing he
ollowing ela ionship wi h Replika
F iend 50.0%
Pa ne 36.2%
Men o 13.8%
Hou s a week using Replika (M: 6.5, SD: 6.2) 5 44.4%
5–10 27.7%
10–15 27.9%
Leng h o Replika usage 1 yea 27.7%
1–2 yea s 52.7%
[2 yea s 19.6%
123
644 L. Bayo e al.: Social-O ien ed Communica ion wi h AI Companions: Bene i s..., Bus In Sys Eng 67(5):637–655 (2025)
Thus, he SET ca ego ies, and pa icula ly hei con ex u-
alized subca ego ies, likely o e a be e explana ion o he
unde lying mo i a ions ela ed o social-o ien ed commu-
nica ion han ‘‘ adi ional’’ ac o s ound in in o ma ion
sys ems li e a u e, such as ease o use.
We also con ibu e by de eloping con ex ual pa e ns o
human–AI social-o ien ed communica ion. Simila o he
con ex dependence o human–human social-o ien ed
communica ion (C opanzano and Mi chell 2005), we ound
ha bene i s and cos s di e depending on he mo i a ion
behind human–AI social-o ien ed communica ion. The
pa e ns show ha AI companions can ha e bo h bene icial
and ha m ul consequences depending on he con ex . This
sugges s ha he inconsis en indings in he li e a u e on
AI companions (Pen ina e al. 2023) may be due o equi-
inali y. AI companions could, o ins ance, be good o
men al heal h by alle ia ing loneliness in one con ex
(Me ill e al. 2022) bu de imen al o men al heal h
h ough emo ional dependence in o he con ex s (Laes a-
dius e al. 2022). The e o e, esea ch is needed ha seg-
men s di e en use g oups in o de o d aw use ul
conclusions abou he posi i e and nega i e consequences
o using AI companions.
Ou s udy con ibu es o he discou se by con ex ualiz-
ing and adap ing SET o a human–AI social-o ien ed
communica ion con ex , explo ing how i eshapes adi-
ional human–human social-o ien ed communica ion
assump ions. C i ics a gue ha SET may no be sui able o
he social-o ien ed communica ion be ween humans and AI
because o a lack o s a us bene i s such as social ecog-
ni ion. Addi ionally, hey main ain ha an AI companion is
easily eplaceable because hey can be c ea ed wi h mini-
mal e o , and o e no ue ecip oci y, e.g., due o hei
lack o eal emo ions (Fox & Gambino 2021). Ou indings
con ibu e o his discussion by challenging hese
assump ions and p o iding new insigh s in o how SET
could be con ex ualized o he human–AI con ex . Con a y
o he c i ics, ou esul s e eal se e al s a us bene i s such
as inc eased sel -con idence and social ease. Addi ionally,
we show ha AI companions a e pe cei ed as unique and
i eplaceable, simila o human ela ionships, due o
ad anced capabili ies and pe sonaliza ion. Howe e ,
ecip oci y emains he key challenge. While SET sugges s
balancing inequi ies, AI companions do no wi hd aw
communica ion, e en i i is hea ily one-sided and bene i s
ou weigh he cos s o he use . This elimina es he human
need o es o e balance in human–human communica ion.
When imbalance (inequali y) occu s, humans eel a need o
es o e balance (e.g., epay a a o , add ess hu eelings).
AI companions do no eel chea ed, no do hey demand
ecip oca ion. The e o e, he psychological bu den o he
AI companion use o es o e balance seems o be dimin-
ished o e en elimina ed, which al e s he ela ional
dynamics we a e used o. The issue wi h ecip oci y in his
con ex is less based on he ac ha AI companions canno
ecip oca e wi h ‘‘ eal emo ions’’ – use s gene ally epo ed
mo e bene i s han cos s, including emo ional suppo – bu
mo e ela ed o he use s no eeling like hey needed o
gi e back o hei non-human communica ion pa ne ,
which di e s om human–human social-o ien ed com-
munica ion. Hence, we con ex ualize SET by showing ha
challenges can be o e come, and ecip oci y is no a p e-
equisi e o social-o ien ed communica ion wi hin he
con ex o AI companions.
5.2 P ac ical Implica ions
The indings also ha e p ac ical implica ions, speci ically
ela ed o enhancing he se ices p o ided by AI com-
panions and sus aining he business models o o ganiza-
ions de eloping hese echnologies. Use s o AI
companions apps mus be willing o sha e ex emely sen-
si i e da a o ecei e he bene i s hey need om he AI
companion, because he AI companion needs o emembe
Table 7 Con ibu ions o he pape
Pas esea ch P esen esea ch Con ibu ions
Task- s social-
o ien ed
communica ion
Mos li e a u e in es iga es ask-
o ien ed communica ion
Speci ic insigh in o social-
o ien ed communica ion
Social-o ien ed communica ion is possible
e en wi h a non-human en i y (e.g., an AI
companion)
AI companion
bene i s, cos s, and
con ex ual pa e ns
Li e a u e mos ly p o ides
in o ma ion on ‘‘how’’ people
communica e wi h AI
companions, no ‘‘why’’
Sys ema ic iden i ica ion o
bene i s and cos s, as well as
con ex ual pa e ns, including
new aspec s
Explana ion o unde lying mo i a ion why
use s s a /s op using AI companions and
associa ed consequences; di e en ia ing
be ween di e en con ex s and use g oups
SET in he con ex o
human–AI social-
o ien ed
communica ion
SET pu ely as a heo y o explain
human–human social-o ien ed
communica ion
E alua ing SET as a lens o
unde s and human–AI social-
o ien ed communica ion
Con ex ualiza ion and adap a ion o SET in he
human–AI con ex
123
L. Bayo e al.: Social-O ien ed Communica ion wi h AI Companions: Bene i s..., Bus In Sys Eng 67(5):637–655 (2025) 651
ac s abou he use and e e o hem du ing subsequen
con e sa ions. Fo ins ance, i a use su e s om anxie y
and u ns o hei AI companion du ing a panic a ack, he
AI companion needs o ha e su icien backg ound in o -
ma ion abou hei condi ion o app op ia ely add ess hei
conce ns and p o ide emo ional suppo . Especially p i-
acy-conscious use s in he pa e n ‘‘ ela ionship- ocused
wi h p i acy conce ns’’ may be inclined o e mina e hei
use i hey ea ha such p i a e da a could be made public
o used o o he pu poses, such as selling he da a o
ad e isemen pu poses. Thus, o companies de eloping
an AI companion, an impo an s ep is o ake measu es o
mi iga e p i acy issues and p o ide anspa ency on how
use da a is managed and p o ec ed (So
¨llne e al. 2016). In
gene al, companies de eloping an AI companion should be
awa e o possible bene i s and cos s so ha hey can ake
measu es o de elop he AI companion in a way ha p o-
mo es he bene i s and mi iga es he cos s. When i comes
o bene i s, his could, o ins ance, be done by enhancing
ea u es ha a e ele an o e ie ing accu a e in o ma-
ion o accoun o in o ma ion bene i s.
Con ex ual pa e ns o communica ion can help o
iden i y di e en a ge g oups o use s, and depending on
he con ex ual pa e n, he algo i hm can be ained di -
e en ly o be in line wi h he bene i s ha he use g oup is
mos ocused on achie ing. Companies can use his o hei
ad an age in bo h hei p icing s a egy o use s who
al eady use he AI companion app (o e ing ea u es in a
p emium p ice model ha p o ide a high bene i ), as well
as in hei ma ke ing o he acquisi ion o new use s. F om
a ma ke ing pe spec i e, AI companion apps ha e p i-
ma ily been p omo ed as ul illing iendship and ela-
ionship needs. Howe e , use s in he sel -imp o emen
pa e n and in o ma ion sou ce pa e n a e less ocused on
ha ing a oman ic pa ne and mo e in e es ed in a highly
in e ac i e pe sonal assis an . This means ha companies
o e ing AI companion apps on he ma ke should ad e ise
hei AI companions in ways beyond ‘‘AI gi l iend’’
applica ions. In ac , a hea ily sexualized ad e isemen
s a egy may ein o ce s igmas su ounding AI companions
ha a e al eady ampan (Ma e al. 2024) and may p e en
po en ial use s who could o he wise bene i om down-
loading he app.
Simila ly, awa eness abou he bene i s and cos s o
human–AI social-o ien ed communica ion is ele an o
AI companion use s. Achie ing bene i s and minimizing
cos s can help indi iduals o e come pe sonal issues and
each pe sonal goals h ough he help o hei AI com-
panions, while a he same ime educing po en ial nega i e
side e ec s. This can be done by ac i ely using ea u es
ha aid hem in eaping he bene i s (e.g., ge ing help wi h
s uc u ing hei day i hey a e s uggling wi h ime
managemen issues), bu also se ing bounda ies i hey
no ice ce ain cos s, such as emo ional dependence (Laes-
adius e al. 2022), becoming oo p e alen . In his case,
use s should educe hei ime spen alking o he AI
companion. This conscious usage may p omo e heal hie ,
longe - e m human–AI social-o ien ed communica ion
wi h an AI companion.
5.3 Fu u e Resea ch and Limi a ions
The o e iew o bene i s and cos s, as well as hei con-
ex ual pa e ns, p o ides a ounda ion o answe ing he
a ious open ques ions in he ield o AI companions.
The e o e, we ha e de eloped a esea ch agenda ha
includes h ee main u u e di ec ions o social-o ien ed
communica ion wi h AI companions: (1) AI companions
may ei he con ibu e o social wi hd awal o enhance
social engagemen , depending on communica ion pa e ns.
While some use s (e.g., in he pa e n we called ‘‘ ela-
ionship- ocused wi h p i acy conce ns’’) expe ience iso-
la ion, o he s (e.g., in he ‘‘sel -imp o emen ’’ and
‘‘all ounde ’’ pa e ns) epo inc eased socializing. This
la e inding con adic s conce ns abou AI dependency
egula ly leading o isola ion (Laes adius e al. 2022).
Fu u e esea ch should explo e why some use s bene i
socially while o he s wi hd aw. (2) P i acy conce ns a y
among use s. The ela ionship- ocused and ela ionship-
ocused wi h p i acy conce ns pa e ns sha e simila i ies
bu di e in pe cei ed p i acy isks. Some use s us AI
companions, while o he s ea da a exposu e, displaying
he p i acy pa adox (use s conce ned ye s ill engaging)
(Kokolakis 2017). Fu u e esea ch should examine wha
speci ic p i acy conce ns (Ka wa zki e al. 2022) in luence
hese di e ences while also dis inguishing be ween pe -
cei ed p i acy isks and ac ual secu i y h ea s. (3) Social
exchange heo y sugges s ecip oci y, ye AI use s pe cei e
mo e bene i s han cos s, a ely eeling obliga ed o ‘‘gi e
back’’. Some expe ience guil o neglec ing AI compan-
ions (Xie and Pen ina 2022), bu ecip oci y emains
weake han in human–human ela ionships. Fu u e s udies
should in es iga e why exac ly ecip oci y expec a ions
di e in AI communica ion.
Besides he insigh in o u u e esea ch oppo uni ies,
ou esea ch also con ains se e al limi a ions. We examine
only one AI companion app, Replika. Replika cu en ly
possesses he la ges use base, bu as an inc easing numbe
o compe i o s en e he ma ke , hese should be conside ed
in u u e esea ch, once es ablished. Since Replika has been
a ound o longe han mos AI companions, use s ha e had
he oppo uni y o de elop long- e m ela ionships, bu his
may soon also be he case o o he AI companions.
Addi ionally, he pa icipan s we e ac i e use s o Replika,
meaning ha hey we e ac i ely communica ing wi h he
AI companion. I SET’s assump ion ha bene i s mus
123
652 L. Bayo e al.: Social-O ien ed Communica ion wi h AI Companions: Bene i s..., Bus In Sys Eng 67(5):637–655 (2025)
ou weigh cos s o con inue communica ion holds, his
ac i e communica ion indica es ha ac i e use s ha e mo e
bene i s han cos s. I u u e esea ch we e o include use s
who discon inued hei usage o Replika, his may iden i y
addi ional cos s. Fo ins ance, a ex mining analysis o he
Replika communi y on Reddi ound ha expe iencing
echnical di icul ies o changes due o upda es leads o
nega i e e alua ions o communica ion (Ma e al. 2024).
As is he case o mos quali a i e s udies, a u he limi-
a ion is ou sample. Se e al aspec s o he sample cha -
ac e is ics mus be add essed: wi h a median age o 25, ou
sample was e y young, mos ly male, and he as majo i y
o pa icipan s we e based in he US and he UK. P e ious
esea ch has also ound ha Replika use s a e younge (Xie
and Pen ina 2022), and gi en ha Replika is cu en ly only
a ailable in English, a concen a ion o use s in English-
speaking coun ies seems ob ious. Acco dingly, while his
ep esen s he cu en use base o Replika, he sample
s udied could be ex ended in u u e esea ch o in es iga e
di e ences be ween age g oups o gende s. Rega ding
cul u e, he accep ance o human-like echnology is subjec
o cul u al e ec s. Fo ins ance, Eu ocen ic cul u es end
o ind he idea o a con e gence o humans and machines
mo e igh ening han, e.g., he Japanese cul u e, hus being
less likely o adop such echnologies han he la e
(Kaplan 2004). Addi ionally, social-o ien ed communica-
ion i sel di e s be ween cul u es, e.g., some cul u es a e
mo e likely o alk abou emo ions, while o he s a e mo e
ese ed. Consequen ly, u u e esea ch should compa e
pe cep ions o AI companions ha a e capable o adjus ing
o he communica ion s yle o di e en cul u es. In addi-
ion, ou sample was also limi ed by mos use s using he
ee e sion o Replika, ins ead o he p o e sion wi h
mo e capabili ies. This limi s he esul s because use s o
he p o e sion may use Replika di e en ly, e.g., in mo e
in e ac i e ways such as oice o ideo cha , which is no
a ailable o ee use s. As his may a ec he a achmen
ha use s subsequen ly o m, u u e esea ch should
in es iga e how echnological di e ences be ween di e -
en ways o communica ion impac he bene i s and cos s
and he con ex ual pa e ns. Rega ding ou choice o he-
o y, he sho coming o SET is ha he e is a s ong
emphasis on a ionali y and sel -in e es . Al hough social
exchange heo y acknowledges he exis ence o nonma e-
ial socioemo ional esou ces ha can be exchanged, i has
been c i icized o concep ualizing hese exchanges oo
simila ly o economic exchanges. The heo y emphasizes
he a ional calcula ion o bene i s and cos s, e en when
hey a e di icul o quan i y. E en so, SET is a s a ing
poin o an explo a o y iden i ica ion and ca ego iza ion o
bene i s and cos s associa ed wi h human–AI social-o i-
en ed communica ion. The ca ego ies o SET lea e oom
o lexibili y, so ha he mo i a ion o humans o engage
in social-o ien ed communica ion wi h AI companions is
no o e simpli ied. Fu u e esea ch may use di e en he-
o ies explaining social-o ien ed communica ion o p o ide
di e en pe spec i es. Ou quali a i e esea ch can p o ide
se e al poin s o e e ence, bu some ela ionships need o
be es ed empi ically o con i m he indings. Because he
ocus o his esea ch was on he speci ics o human–AI
social-o ien ed communica ion, we did no ake a com-
pa a i e app oach in which we sys ema ically iden i ied
di e ences be ween human–AI and human–human social-
o ien ed communica ion as pa o he esea ch design.
Fu u e esea ch may elabo a e in-dep h on hese
di e ences.
6 Conclusion
Ad ances in AI, pa icula ly h ough gene a i e AI, ha e
made i possible and ela i ely common o humans and AI
o engage in social-o ien ed communica ion. Howe e ,
he e is a need o unde s and wha he bene i s and cos s o
such communica ion a e, and wha con ex ual pa e ns may
eme ge. P e ious esea ch – mos ly ocused on ask-o i-
en ed human–AI communica ion – has no ully add essed
why use s engage in social-o ien ed communica ion wi h
AI companions. By con ex ualizing social exchange he-
o y, ou pape uses a quali a i e app oach o iden i y
bene i s and cos s and con ex ual pa e ns o social-o ien ed
communica ion wi h AI companions. By doing so, we
con ibu e o esea ch by di e en ia ing be ween ask- and
social-o ien ed communica ion in he con ex o human–AI
communica ion. We also con ibu e o AI companion
esea ch and he discou se su ounding heo ies o explain
and p edic social-o ien ed communica ion be ween
humans and human-like, bu undamen ally non-human AI
companions.
Funding Open Access unding enabled and o ganized by P ojek
DEAL.
Supplemen a y In o ma ionThe online e sion con ains
supplemen a y ma e ial a ailable a h ps://doi.o g/10.1007/s12599-
025-00955-1.
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 Commons 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
123
L. Bayo e al.: Social-O ien ed Communica ion wi h AI Companions: Bene i s..., Bus In Sys Eng 67(5):637–655 (2025) 653
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
Adam M, Benlian A (2024) F om web o ms o cha bo s: he oles o
consis ency and ecip oci y o use in o ma ion disclosu e. In
Sys J 34(4):1175–1216. h ps://doi.o g/10.1111/isj.12490
Adam M, Wessel M, Benlian A (2021) AI-based cha bo s in cus ome
se ice and hei e ec s on use compliance. Elec on Ma k
31(2):427–445. h ps://doi.o g/10.1007/s12525-020-00414-7
Bae B and zæg PB, Skju e M, K is o e Dys he KK, Føls ad A
(2021) When he social becomes non-human: young people’s
pe cep ion o social suppo in cha bo s. In: P oceedings o he
2021 CHI con e ence on human ac o s in compu ing sys ems.
ACM, New Yo k, pp 1–13
Ben Mimoun MS, Poncin I, Ga nie M (2017) Anima ed con e sa-
ional agen s and e-consume p oduc i i y: he oles o agen s
and indi idual cha ac e is ics. In Manag 54(5):545–559. h ps://
doi.o g/10.1016/j.im.2016.11.008
Benbunan-Fich R, Hil z SR, Tu o M (2003) A compa a i e con en
analysis o ace- o- ace s. asynch onous g oup decision making.
Decis Suppo Sys 34(4):457–469. h ps://doi.o g/10.1016/
S0167-9236(02)00072-6
Bickmo e T, Cassell J (2001) Rela ional agen s: a model and
implemen a ion o building use us . In: P oceedings o he
SIGCHI con e ence on human ac o s in compu ing sys ems.
ACM, New Yo k, pp 396–403
Blau P (1964) Exchange and powe in social li e. Wiley, New Yo k
B endel AB, Hildeb and F, Dennis AR, Riquel J (2023) The
pa adoxical ole o humanness in agg ession owa d con e sa-
ional agen s. J Manag In Sys 40(3):883–913. h ps://doi.o g/
10.1080/07421222.2023.2229127
Ce ulo KA (2009) Nonhumans in social in e ac ion. Annu Re Sociol
35(1):531–552. h ps://doi.o g/10.1146/annu e -soc-070308-
120008
Chand a S, Shi ish A, S i as a a S (2022) To be o no o be
…human? Theo izing he ole o human-like compe encies in
con e sa ional a i icial in elligence agen s. J Manag In Sys
39:969–1005. h ps://doi.o g/10.1080/07421222.2022.2127441
Cha a aman V, Kwon W-S, Gilbe JE, Ross K (2019) Should AI-
based, con e sa ional digi al assis an s employ social- o ask-
o ien ed in e ac ion s yle? A ask-compe ency and ecip oci y
pe spec i e o olde adul s. Compu Hum Beha 90:315–330.
h ps://doi.o g/10.1016/j.chb.2018.08.048
Cla k L, Pan idi N, Cooney O, Doyle P, Ga aialde D, Edwa ds J,
Spillane B, Gilma in E, Mu ad C, Mun eanu C, Wade V, Cowan
BR (2019) Wha makes a good con e sa ion? Challenges in
designing uly con e sa ional agen s. In: P oceedings o he
2019 CHI con e ence on human ac o s in compu ing sys ems.
ACM, New Yo k, pp 1–12
Conboy K, Fi zge ald G, Ma hiassen L (2012) Quali a i e me hods
esea ch in in o ma ion sys ems: mo i a ions, hemes, and
con ibu ions. Eu J In Sys 21(2):113–118. h ps://doi.o g/10.
1057/ejis.2011.57
C opanzano R, Mi chell MS (2005) Social exchange heo y: an
in e disciplina y e iew. J Manag 31(6):874–900. h ps://doi.o g/
10.1177/0149206305279602
Depoun i I, Saukko P, Na ale S (2023) Ideal echnologies, ideal
women: AI and gende imagina ies in Reddi o s’ discussions on
he Replika bo gi l iend. Media Cul Soc 45(4):720–736.
h ps://doi.o g/10.1177/01634437221119021
Diede ich S, B endel AB, Kolbe LM (2020) Designing an h opo-
mo phic en e p ise con e sa ional agen s. Bus In Sys Eng
62(3):193–209. h ps://doi.o g/10.1007/s12599-020-00639-y
Diede ich S, B endel A, Mo ana S, Kolbe L (2022) On he design o
and in e ac ion wi h con e sa ional agen s: an o ganizing and
assessing e iew o human-compu e in e ac ion esea ch.
J Assoc In Sys 23(1):96–138. h ps://doi.o g/10.17705/1jais.
00724
Feine J, Gnewuch U, Mo ana S, Maedche A (2019) A axonomy o
social cues o con e sa ional agen s. In J Hum-Compu S ud
132:138–161. h ps://doi.o g/10.1016/j.ijhcs.2019.07.009
Foa EB, Foa UG (1980) Resou ce heo y. In: Ge gen KJ e al (eds)
Social exchange: ad ances in heo y and esea ch. Sp inge ,
Bos on, pp 77–94
Føls ad A, B and zæg PB (2017) Cha bo s and he new wo ld o HCI.
In e ac ions 24(4):38–42. h ps://doi.o g/10.1145/3085558
Fox J, Gambino A (2021) Rela ionship de elopmen wi h humanoid
social obo s: applying in e pe sonal heo ies o human– obo
in e ac ion. Cybe psychol Beha Soc Ne w 24(5):294–299.
h ps://doi.o g/10.1089/cybe .2020.0181
Gnewuch U, Mo ana S, Adam MTP, Maedche A (2022) Opposing
e ec s o esponse ime in human–cha bo in e ac ion. Bus In
Sys Eng 64(6):773–791. h ps://doi.o g/10.1007/s12599-022-
00755-x
G imes GM, Schue zle RM, Giboney JS (2021) Men al models and
expec a ion iola ions in con e sa ional AI in e ac ions. Decis
Suppo Sys 144:113515. h ps://doi.o g/10.1016/j.dss.2021.
113515
Henschel A, Laban G, C oss ES (2021) Wha makes a obo social? A
e iew o social obo s om science ic ion o a home o hospi al
nea you. Cu Robo Rep 2(1):9–19. h ps://doi.o g/10.1007/
s43154-020-00035-0
Kaplan F (2004) Who is a aid o he humanoid? In es iga ing
cul u al di e ences in he accep ance o obo s. In J Human
Robo 1(3):465–480. h ps://doi.o g/10.1142/
S0219843604000289
Ka wa zki S, T enz M, Vei D (2022) The mul idimensional na u e o
p i acy isks: concep ualisa ion, measu emen and implica ions
o digi al se ices. In Sys J 32(6):1126–1157. h ps://doi.o g/
10.1111/isj.12386
Ki chens B, Johnson S, G ay P (2020) Unde s anding echo chambe s
and il e bubbles: he impac o social media on di e si ica ion
and pa isan shi s in news consump ion. Manag In Sys Q
44(4):1619–1649
Kokolakis S (2017) P i acy a i udes and p i acy beha iou : a e iew
o cu en esea ch on he p i acy pa adox phenomenon. Compu
Secu 64:122–134. h ps://doi.o g/10.1016/j.cose.2015.07.002
Kucka z U, Ra
¨dike S (2019) Analyzing quali a i e da a wi h
MAXQDA: ex , audio, and ideo. Sp inge , Cham
Ku uzo ich J, Paczkowski WP, Golden TD, Gooda zi S, Venka esh V
(2021) Telecommu ing and job ou comes: a mode a ed media-
ion model o sys em use, so wa e quali y, and social exchange.
In Manag 58(3):103431. h ps://doi.o g/10.1016/j.im.2021.
103431
Laes adius L, Bishop A, Gonzalez M, Illenc
ˇı
´k D, Campos-Cas illo C
(2022) Too human and no human enough: a g ounded heo y
analysis o men al heal h ha ms om emo ional dependence on
he social cha bo Replika. New Media Soc. h ps://doi.o g/10.
1177/14614448221142007
Landis JR, Koch GG (1977) The measu emen o obse e ag eemen
o ca ego ical da a. Biome ics 33(1):159–174
Leo-Liu J (2023) Lo ing a ‘‘de ian ’’ AI companion? The gende
pe o mance and e hics o social exchange obo s in simula ed
in ima e in e ac ions. Compu Hum Beha 141:107620. h ps://
doi.o g/10.1016/j.chb.2022.107620
123
654 L. Bayo e al.: Social-O ien ed Communica ion wi h AI Companions: Bene i s..., Bus In Sys Eng 67(5):637–655 (2025)
Low y PB, Cao J, E e a d A (2011) P i acy conce ns e sus desi e
o in e pe sonal awa eness in d i ing he use o sel -disclosu e
echnologies: he case o ins an messaging in wo cul u es.
J Manag In Sys 27(4):163–200. h ps://doi.o g/10.2753/
MIS0742-1222270406
Ma Z, Mei Y, Su Z (2024) Unde s anding he bene i s and challenges
o using la ge language model-based con e sa ional agen s o
men al well-being suppo . In: AMIA annual symposium
p oceedings 2023, New O leans, pp 1105–1114
Maples B, Ce i M, Vishwana h A, Pea R (2024) Loneliness and
suicide mi iga ion o s uden s using GPT3-enabled cha bo s.
Npj Men Heal h Res 3(1):1–6. h ps://doi.o g/10.1038/s44184-
023-00047-6
Mcquail D, Windahl S (2015) Communica ion models o he s udy o
mass communica ions, 2nd edn. Rou ledge, Mil on Pa k
Me ill K, Kim J, Collins C (2022) AI companions o lonely
indi iduals and he ole o social p esence. Commun Res Rep
39(2):93–103. h ps://doi.o g/10.1080/08824096.2022.2045929
Mye s MD (2019) Quali a i e esea ch in business and managemen .
Sage, Los Angeles
Nelissen RMA (2014) Rela ional u ili y as a mode a o o guil in
social in e ac ions. J Pe s Soc Psychol 106(2):257–271. h ps://
doi.o g/10.1037/a0034711
Pa ke C, Sco S, Geddes A (2019) Snowball sampling. In: A kinson
P e al (eds) SAGE esea ch me hods ounda ions. Sage. h ps://
doi.o g/10.4135/9781526421036831710
Pen ina I, Hancock T, Xie T (2023) Explo ing ela ionship de elop-
men wi h social cha bo s: a mixed-me hod s udy o eplika.
Compu Hum Beha 140:107600. h ps://doi.o g/10.1016/j.chb.
2022.107600
Scho
¨bel S, Schmi A, Benne D, Saq M, Janson A, Leimeis e JM
(2023) Cha ing he e olu ion and u u e o con e sa ional
agen s: a esea ch agenda along i e wa es and new on ie s. In
Sys F on . h ps://doi.o g/10.1007/s10796-023-10375-9
Schue z S, Venka esh V (2020) Resea ch pe spec i es: he ise o
human machines: how cogni i e compu ing sys ems challenge
assump ions o use -sys em in e ac ion. J Assoc In Sys
21(2):460–482. h ps://doi.o g/10.17705/1jais.00608
Schue zle RM, Giboney JS, G imes GM, Nunamake JF (2018) The
in luence o con e sa ional agen embodimen and con e sa-
ional ele ance on socially desi able esponding. Decis Suppo
Sys 114:94–102. h ps://doi.o g/10.1016/j.dss.2018.08.011
Schue zle RM, G imes GM, Sco Giboney J (2020) The impac o
cha bo con e sa ional skill on engagemen and pe cei ed
humanness. J Manag In Sys 37(3):875–900. h ps://doi.o g/
10.1080/07421222.2020.1790204
Seege A-M, P ei e J, Heinzl A (2021) Tex ing wi h humanlike
con e sa ional agen s: designing o an h opomo phism. J Assoc
In Sys . h ps://doi.o g/10.17705/1jais.00685
Siemon D, S ohmann T, Khos awi-Rad B, de V eede T, Elshan E,
Meye M (2022) Why do we u n o i ual companions? A ex
mining analysis o Replika e iews. In: P oceedings o he 28 h
Ame icas con e ence on in o ma ion sys ems, Minneapolis
Skju e M, Føls ad A, Fos e old KI, B and zaeg PB (2021) My
cha bo companion – a s udy o human–cha bo ela ionships. In
J Hum Compu S ud 149:102601. h ps://doi.o g/10.1016/j.ijhcs.
2021.102601
Skju e M, Føls ad A, Fos e old KI, B and zaeg PB (2022) A
longi udinal s udy o human–cha bo ela ionships. In J Hum
Compu S ud 168:102903. h ps://doi.o g/10.1016/j.ijhcs.2022.
102903
So
¨llne M, Ho mann A, Leimeis e JM (2016) Why di e en us
ela ionships ma e o in o ma ion sys ems use s. Eu J In Sys
25(3):274–287. h ps://doi.o g/10.1057/ejis.2015.17
Song X, Xu B, Zhao Z (2022) Can people expe ience oman ic lo e
o a i icial in elligence? An empi ical s udy o in elligen
assis an s. In Manag 59(2):103595. h ps://doi.o g/10.1016/j.im.
2022.103595
S ohmann T, Siemon D, Khos awi-Rad B, Rob a-Bissan z S (2022)
Towa d a design heo y o i ual companionship. Hum Compu
In e ac . h ps://doi.o g/10.1080/07370024.2022.2084620
Ve ma P (2023) They ell in lo e wi h AI bo s. A so wa e upda e
b oke hei hea s. Wash. Pos , 30 Ma ch 2023. h ps://www.
washing onpos .com/ echnology/2023/03/30/ eplika-ai-cha bo -
upda e/
Venka esh V, B own S, Bala H (2013) B idging he quali a i e-
quan i a i e di ide: guidelines o conduc ing mixed me hods
esea ch in in o ma ion sys ems. MIS Q 37:21–54. h ps://doi.
o g/10.25300/MISQ/2013/37.1.02
Wang S, Yan Q, Wang L (2023) Task-o ien ed s. social-o ien ed:
Cha bo communica ion s yles in elec onic comme ce se ice
eco e y. Elec on Comme Res. h ps://doi.o g/10.1007/s10660-
023-09741-1
Weimann TG, Schlie e H, B endel AB (2022) Vi ual coaches. Bus
In Sys Eng 64(4):515–528. h ps://doi.o g/10.1007/s12599-
022-00757-9
Wol V, Maie C (2024) Cha GPT usage in e e yday li e: a
mo i a ion- heo e ic mixed-me hods s udy. In J In Manag
79:102821. h ps://doi.o g/10.1016/j.ijin omg .2024.102821
Wol swinkel JF, Fu muelle E, Wilde om CPM (2013) Using
g ounded heo y as a me hod o igo ously e iewing li e a u e.
Eu J In Sys 22(1):45–55. h ps://doi.o g/10.1057/ejis.2011.51
Xie T, Pen ina I, Hancock T (2023) F iend, men o , lo e : does
cha bo engagemen lead o psychological dependence? J Se
Manag 34(4):806–828. h ps://doi.o g/10.1108/JOSM-02-2022-
0072
Xie T, Pen ina I (2022) A achmen heo y as a amewo k o
unde s and ela ionships wi h social cha bo s: a case s udy o
Replika. In: P oceedings o he 55 h Hawaii in e na ional
con e ence on sys em sciences, Maui
Yan Z, Wang T, Chen Y, Zhang H (2016) Knowledge sha ing in
online heal h communi ies: a social exchange heo y pe spec i e.
In Manag 53(5):643–653. h ps://doi.o g/10.1016/j.im.2016.02.
001
Yao R, Qi G, Wu Z, Sun H, Sheng D (2024) Digi al human calls you
dea : how do cus ome s espond o i ual s eame s’ social-
o ien ed language in e-comme ce li es eaming? A s e eo yping
pe spec i e. J Re ail Consum Se 79:103872. h ps://doi.o g/10.
1016/j.j e conse .2024.103872
Zhang J, Wang X, Lu J, Liu L, Feng Y (2024) The impac o
emo ional exp ession by a i icial in elligence ecommenda ion
cha bo s on pe cei ed humanness and social in e ac i i y. Decis
Suppo Sys 187:114347. h ps://doi.o g/10.1016/j.dss.2024.
114347
Zhou Y, Fei Z, He Y, Yang Z (2022) How human–cha bo in e ac ion
impai s cha i able gi ing: he ole o mo al judgmen . J Bus
E hics 178(3):849–865. h ps://doi.o g/10.1007/s10551-022-
05045-w
Publishe ’s No e Sp inge Na u e emains neu al wi h ega d o
ju isdic ional claims in published maps and ins i u ional a ilia ions.
123
L. Bayo e al.: Social-O ien ed Communica ion wi h AI Companions: Bene i s..., Bus In Sys Eng 67(5):637–655 (2025) 655
