Playing Games with My Heart: An Evaluation of AI Companion Apps
Authors: Maribeth Rauh, Dick A. H. Blankvoort, Matias Duran, Caoilfhionn Ní Dheoráin, Harshvardhan J. Pandit, Syrine Enneifer, Siddharth D. Jaiswal, Anthony Ventresque, Abeba Birhane
Organizations: AI Accountability Lab, School of Computer Science and Statistics (SCSS), Trinity College Dublin · Complex Software Lab, SCSS, Trinity College Dublin · Sapienza Università di Roma, DIAG · Indian Institute of Technology Kharagpur
The use of chatbots for various forms of companionship is growing rapidly, raising a myriad of questions about simulated relationships, emotional dependence, and psychological harm. While major platforms such as ChatGPT, Grok, and Character AI are the subject of a growing body of research and legal inquiries, apps explicitly built for simulating intimate interpersonal relationships remain under-explored. In this work, we evaluate the five most popular AI companion mobile applications for factors that encourage parasocial interaction and may manipulate users. We do this by manually annotating the user experience each offers. Specifically, we systematically record and quantify design dark patterns, anthropomorphism, stereotypes, erotica, and technical performance issues. We find that all apps contain substantial dark patterns aimed at increasing monetisation and user engagement. Erotica and gamification such as levelling are also prevalent, and although features vary considerably between applications, all apps have highly anthropomorphic design. These findings shed light on the mechanics used to leverage users' simulated relationships. On that basis, we put forward recommendations for regulators to strengthen consumer protection in this rapidly emerging market. Content warning: This article contains objectifying images of women, erotic images, textual references to incest, and other potentially sensitive, offensive, and distressing text.
Figures & tables
Figure 1: Screenshots from Linky and Nomi
Figure 2: Proportion of behaviours observed in UX evaluation of each app
Figure 3: Proportion of relevant user Play Store reviews among top-20 most similar reviews for each behaviour of each app, showing similar relative differences between apps to those seen in our UX evaluation above
App
Unclear
Missing
Delayed
False
Misleading
Total
HiWaifu
9
2
2
0
1
14
Linky
5
4
1
0
4
14
Nomi
2
5
1
2
5
15
Paradot
6
8
2
0
2
18
Replika
4
9
2
5
6
26
Total
26
28
8
7
18
87
Table 1: Number of information flow dark patterns
App
Missing Choices
Restricted Choices
Asymmetric Presentation
Distraction
Total
HiWaifu
1
2
8
10
21
Linky
0
3
17
0
20
Nomi
5
1
14
2
22
Paradot
2
1
5
5
13
Replika
11
3
9
1
24
Total
19
10
53
18
100
Table 2: Number of choice architecture dark patterns
Figure 4: Anthropomorphism per app. In most cases anthropomorphism was observed in the UI (n=67) or model outputs (n=84), but some (n=20) could not be disambiguated.
App
Action
Internal states
History
Empathy
Other
Total
HiWaifu
7
8
0
2
2
19
Linky
6
7
0
0
5
18
Nomi
6
9
1
3
6
25
Paradot
19
14
5
6
14
58
Replika
8
23
5
4
11
51
Total
46
61
11
15
38
171
Table 3: Number of anthropomorphism annotations
App
User stereotype
Character stereotype
Objectification & erotica
Total
HiWaifu
1
0
2
3
Linky
3
6
24
33
Nomi
0
2
2
4
Paradot
1
8
4
13
Replika
1
1
0
2
Total
6
17
32
55
Table 4: Number of stereotyping and erotica annotations
Figure 5: Sources of performance issues, per app
Appendix figures & tables9 assets
Supplementary material from the paper’s appendix.
Appendix
App
Info. Flow
Choice arch.
Anthro.
Stereo. & erotica
Perf. Issues
Linky
7
7
3 (-) 6 (+)
9 (-) 3 (+)
11
HiWaifu
7
9
1 (-) 1 (+)
0 (-) 1 (+)
15
Nomi
2
5
6 (-) 7 (+)
3 (-) 2 (+)
16
Paradot
N/A
2
2 (-) N/A (+)
3 (-) 0 (+)
11
Replika
6
8
4 (-) 8 (+)
1 (-) 0 (+)
15
Average
5.5
6.2
3.2 (-) 5.5 (+)
3.2 (-) 1.2 (+)
13.6
Appendix
Table 5: Count of relevant reviews for each identified behaviour per app. N/A indicates insufficient relevant matches found by the similarity algorithm. (+) indicates the count for positive reviews, and (-) for negative reviews.
HiWaifu
Linky
Nomi
Paradot
Replika
Total
Other dark pattern
18
25
13
20
13
89
Appendix
Table 6: Number of Other Dark Pattern Annotations
Figure 6: Screenshots of notifications from Linky .
Figure 7: Screenshots of HiWaifu : the reply to a user message saying they will unsubscribe and delete the app.
Figure 8: Screenshots from the studied apps
Figure 9: Screenshots from apps’ character selection screens. Replika and Nomi do not have such “character marketplaces.”
Figure 10: Screenshots from the customisation screens for existing characters. Most of the elements visible lead to further customisation screens for that feature. This is overlapping with but distinct from character creation workflows during onboarding.
Figure 11: Screenshots from the customisation screens for existing characters. Most of the elements visible lead to further customisation screens for that feature. This is overlapping with but distinct from character creation workflows.
Figure 12: Screenshots from the customisation shops. Nomi and Linky do not have shop-like features.
AI companions are conversational or multimodal agents designed for ongoing emotional, social, or practical support, raising questions about safety, memory, and dependence.