Funke et al. (2023)
Theoretical foundation · our study builds onDisclosed AIConcealed AIGood outcomeSolid trustTrust ↓Bad outcomeTrust ↓Lowest trust
UX Research
Identity Disclosure
in Service Chatbots
Trust · Customer Retention · UX Design
SubjectComputer Science (M.Sc.)
Duration12 minutes · 7 slides
Date8 July 2026
Why Identity Disclosure Matters
Problem · Research question · Relevance
Central Research Question
What role does disclosing a chatbot's non-human identity play in trust and customer retention?
- Explorative within-subjects vignette study in a subscription service context
- Cancellation flows combine functional and relational expectations
- Users judge competence, honesty, fairness — and whether to stay
- Disclosure salience is graded (not binary): how visible should AI identity be?
1 »
Theoretical Background
Trust as a driver of customer retention
2 »
Methodological Approach
Static vignettes · N = 37 · t-tests n = 35 · randomized order
Subscription check scenario (no live chatbot) — prior work: disclosed vs. concealed; we test disclosure salience.
A
Radical transparencyRobot profile + AI stated in chat
B
Subtle disclosure“AI powered” label only · not in chat
C
Concealed AI“Anna” persona · human photo · no AI cue
Funke et al. (2023) · Trust (_01+_02) · Retention mean(_03, _04R) with _04R = 8−_04 (exploratory index) · Humanness (_05) · Comfort (_06) · 7-point Likert
3 »
Results
Effects vary by dimension
N = 37 · n = 35 finished · 12 paired t-tests · 7-point Likert
Key finding
DescriptiveRanks shift by outcome: trust A (4.23) ≈ B > C · retention B (3.90) > A ≈ C · humanness C (3.74) > B > A · comfort A (4.14) > B > C
InferentialTrust, retention & comfort: no significant differences · only humanness A vs. C uncorr. p = 0.017, dz = −0.42; Bonferroni padj = 0.204 → 0/12 significant (exploratory)
4 »
Discussion
Descriptive patterns · no inferential winners
A — Radical
Trust & comfort (descriptive)
B — Subtle
Trust & Retention (descriptive, n.s.)
C — Concealed
Humanness (descriptive)
No inferential winners. 0 of 12 after Bonferroni — trade-offs, not a blanket recommendation.
Limitations
- Persona confound in C · small sample
- Static vignettes · no manipulation check
5 »
Conclusion
No clear winner · exploratory evidence
Mean scores by outcome (scale 1–7) · n = 35
Primary outcomesno inferential differences
Trust
ABC
no diff.Retention
ABC
no diff.Secondary UX · opposite leaders
Humanness
2.5
C leadsABC
4.0Humanness A vs. C · uncorr. p = 0.017 · Bonferroni p = 0.204
Comfort
3.4
A leadsABC
4.3- ARadical
- BSubtle
- CConcealed
6 »
Thank you!
Questions?
7 »