Speaker Notes

Title Slide

We conducted research on identity disclosure in service chatbots.

Slide 1 — Why Identity Disclosure Matters

Opening: In customer support today, you often talk to AI first — only later does the system route you to a human agent.

This is a growing trend, not an edge case. That's why we're interested in how AI identity disclosure shapes trust and retention.

Say: Read research question verbatim (on slide).

Footer: 1 »

Slide 2 — Theoretical Background

Explain the 2×2 matrix: disclosed vs. concealed AI × good vs. bad service outcome → trust.

Bottom row: bad outcome lowers trust — fairly obvious. The more interesting question is the disclosure axis: even with a good outcome, concealed AI still costs trust.

Beyond trust, we also care about intention — whether users stay or leave (retention).

Our extension: three salience levels instead of binary disclosed/concealed, constant subscription-check scenario.

Footer: 2 »

Slide 3 — Methodological Approach

Static vignettes, no live chatbot — randomized order, within-subjects.

A: robot + AI in chat. B: ‘AI powered’ label only. C: Anna persona, human photo, no AI cue (persona confound).

Retention = mean(_03, _04R) with _04R = 8−_04 — exploratory index, very weak reliability (SB −0.09 to 0.14).

Footer: 3 »

Slide 4 — Results

35 analyzable answers. Scale 1–7. A = radical transparency · B = subtle disclosure · C = concealed AI.

Trust: A and B ~4.2, C ~3.9. Retention: B ~3.9, A and C ~3.6.

Humanness: C ~3.7, B ~3.1, A ~2.8. Comfort: A ~4.1, B ~3.9, C ~3.6.

A leads trust and comfort · B leads retention · C leads humanness.

12 paired t-tests. Only humanness A vs. C stands out: p = 0.017 uncorrected. After Bonferroni: p = 0.204 — not significant.

0 of 12 significant overall. Therefore only exploratory tendencies.

Footer: 4 »

Slide 5 — Discussion

No variant wins on every dimension.

A (radical transparency): good trust, highest comfort. Disclosure is not automatically negative — it can create a sense of security.

B (subtle): highest retention, possible middle ground.

C (concealed): highest humanness, but lower trust and lower comfort. A human impression does not automatically mean more trust.

Limitations: small sample (n = 35 analyzable), C had name and photo (possible persona effect), static vignettes only, no live chat, no manipulation check.

Footer: 5 »

Slide 6 — Conclusion

No clear statistical winner. No significant differences in trust and retention.

Still, recognizable descriptive patterns: concealed AI feels more human; radically transparent AI feels more comfortable. Transparency does not seem inherently harmful.

No general recommendation — design depends on the goal: transparency and honesty → clearer disclosure; human-like effect → more concealed design.

Future research: larger sample, real live chat, better retention scale, separate persona from disclosure.

Closing line: What matters is not just whether AI is disclosed, but how clearly it is disclosed.

Footer: 6 »

Slide 7 — Thank You

Questions on trade-offs, retention proxy, Bonferroni, persona confound, future work (larger n, validated scales, live chatbots, Funke-style outcomes).

Footer: 7 »