Free consultation

Corporate characters through research ／ Make it easier to continue

Can corporate characters support continued use?

The conclusion first

There is potential for specific short-term support. Simply adding a character cannot be assumed to sustain long-term engagement.

What is known
Reported findings include associations between human-likeness and intention to use, medication support combining reminders, education and answers to questions, and some short-term exercise-coaching outcomes.
Conditions
Show the next action and progress
What remains unknown
It remains unclear whether use continues after six months or one year, whether contracts are renewed or restarted, or whether a static mascot alone supports continued use.

What research tells us relatively clearly

Reported findings include associations between human-likeness and intention to use, medication support combining reminders, education and answers to questions, and some short-term exercise-coaching outcomes.

This includes related-field research on health behavior. It does not establish isolated corporate-character effects or medical benefits provided by YataLogic.

Which conditions seem important?

  • Show the next action and progress
  • Give specific feedback that matches achievements
  • Explain how to return after an interruption
  • Adjust notification frequency and usability, and connect users to people when needed

Representative studies

What was compared, and what changed? Read the findings together with the participants and conditions. YataLogic’s practical interpretations are distinguished from the study results.

Examines relationships and mechanismsResearch in a related field

Wanting to use something and continuing to use it are different

An analysis combining studies of robots, chatbots and similar technologies found a small positive association between perceived human-likeness and intention to use. Usability, usefulness, perceived safety and other factors were also involved.

What did the study compare?
Research on relationships between perceived human-likeness, intention to use and related measures
What was measured?
Primarily intention to use and perceived usability and usefulness
Participants or setting
108 samples; 11,053 participants
What this does not establish

These results did not track actual long-term use or contract renewal. Association alone does not establish causation.

YataLogic’s interpretation: Provide practical support alongside a sense of familiarity.

Read the study details
Study design
Meta-analysis
Comparison and conditions
Meta-analysis of anthropomorphism research
Measures
Primarily usage intentions and perceptions
Main findings
A small positive association between anthropomorphism and intention to use, involving perceived intelligence, safety, ease of use, usefulness and social presence, among other factors.
Key figures (scales differ between studies)
rc=.35 [.30,.41]; 108 samples; 11,053 participants
What can be said
Perceived ease of use, usefulness and safety matter alongside human-likeness.
Important limitations
The research did not track actual long-term use or renewal rates.
Go to notes on statistics and research methods ↓

Blut, M., Wang, C., Wünderlich, N. V., & Brock, C.（2021）

Understanding anthropomorphism in service provision: a meta-analysis of physical robots, chatbots, and other AI

Journal of the Academy of Marketing Science

View the original paper or source
Varies by condition and measureResearch in a related field

Retaining knowledge is different from continued use

Reviews of pedagogical agents included many analyses showing small positive differences in learning outcomes. Results for motivation and emotion were mixed.

What did the study compare?
Multiple reviews of research on pedagogical agents
What was measured?
Learning outcomes, motivation and emotions
Participants or setting
17 reviews, including 7 meta-analyses
What this does not establish

In education research, retention means remembering learned knowledge. It is different from continuing to use a service or completing a course.

YataLogic’s interpretation: Do not treat better learning outcomes as direct evidence of a higher continued-use rate.

Read the study details
Study design
Umbrella review
Comparison and conditions
Umbrella review of pedagogical-agent research
Measures
Learning, motivation and emotion
Main findings
Many meta-analyses show small positive effects on learning outcomes, but results for motivation and emotion are mixed.
Key figures (scales differ between studies)
17 reviews
What can be said
Pedagogical agents show promise for some forms of learning support.
Important limitations
Retention in education research is knowledge retention, not continued service use or course completion.
Go to notes on statistics and research methods ↓

Schroeder, N. L., Davis, R. O., & Yang, E.（2025）

Designing and Learning With Pedagogical Agents: An Umbrella Review

Journal of Educational Computing Research

View the original paper or source
Varies by condition and measureResearch in a related field

Medication-support systems show potential, with varied results

Medication support combining reminders, education and answers to questions had a positive average result. However, results varied greatly between studies, and an analysis limited to objective adherence measures found no clear difference.

What did the study compare?
Research on chatbot-based medication support
What was measured?
Whether medicines were used as prescribed
Participants or setting
7 studies; 5,676 participants (pooled: 6 RCTs; 5,636 participants)
What this does not establish

This is related-field research on health behavior. It concerns the whole support package, not the isolated effect of a character’s appearance or personality.

YataLogic’s interpretation: Consider having a consistent guide deliver next steps and necessary information.

Read the study details
Study design
Systematic review and meta-analysis
Comparison and conditions
Review and meta-analysis of AI chatbot medication support
Measures
Medication adherence
Main findings
The pooled effect was positive but heterogeneity was very high; pooled objective PDC showed no significant difference.
Key figures (scales differ between studies)
SMD=.69 [.17,1.22]; I²=97% / objective PDC SMD=.13, NS
What can be said
Support packages combining reminders, education, Q&A and related elements may be promising in some settings.
Important limitations
This is not an isolated effect of character appearance or personality. SMD=.69 does not mean a 69% improvement.
Go to notes on statistics and research methods ↓

Chen, S., Fang, Y., Ding, L., Mo, P. K. H., & Wang, Z.（2026）

Artificial intelligence–based chatbots to enhance medication adherence among patients with non-communicable chronic diseases: Systematic review and meta-analysis

PLOS Digital Health

View the original paper or source

Additional studies considered

Varies by condition and measureResearch in a related field

Even short-term exercise support depends on the analysis

In a trial of an exercise coach including goal setting and personalized advice, an analysis using all time points over 12 weeks found a behavioral difference. A simple comparison of change among participants with complete data did not show a clear difference.

What did the study compare?
Support combining an on-screen coach, goal setting and personalized advice versus a comparison condition
What was measured?
Step counts and physical activity
Participants or setting
n=70
What this does not establish

This is not simply a before-and-after effect, nor can the difference be attributed to the coach character alone.

YataLogic’s interpretation: Clarify the job of delivering goals, progress and advice suited to the individual.

Read the study details
Study design
RCT
Comparison and conditions
Virtual coach plus goal setting/feedback versus control
Measures
Step counts and activity
Main findings
Repeated-measures analysis using all time points over 12 weeks found a difference, but simple change among complete cases had p=.07.
Key figures (scales differ between studies)
+2.9% vs -12.8%; repeated-measures p=.02
What can be said
There is potential for short-term behavior support.
Important limitations
This is not a simple before-and-after comparison against a pre-intervention baseline, nor an isolated character effect.
Go to notes on statistics and research methods ↓

Watson, A., Bickmore, T., Cange, A., Kulshreshtha, A., & Kvedar, J.（2012）

An Internet-Based Virtual Coach to Promote Physical Activity Adherence in Overweight Adults: Randomized Controlled Trial

Journal of Medical Internet Research

View the original paper or source

Studies without positive results and other cautions

One analysis found no clear difference in participation through the final assessment when conversational agents were used. Short-term findings cannot simply be extended to long-term continued use. Anger and failure also require care.

No clear difference foundResearch in a related field

Conversational guides may not improve continued participation

Across comparative trials of conversational agents in health settings, the proportion remaining in the study until the final assessment was almost the same as in comparison conditions. No clear difference was found.

What did the study compare?
Conversational-agent conditions versus each study’s comparison condition
What was measured?
Whether participants stayed in the study through the final assessment
Participants or setting
14 RCTs; 11,059 participants in total
What this does not establish

The measure was continued study participation, which differs from continued app use.

YataLogic’s interpretation: Decide in advance what counts as continuing: use, completion, returning or another outcome.

Read the study details
Study design
Systematic review and meta-analysis
Comparison and conditions
14 RCTs comparing conversational agents with respective controls
Measures
Continued study participation through final assessment
Main findings
The pooled RR was close to 1, with no difference.
Key figures (scales differ between studies)
RR=.99［.95,1.03］、n=11,059
What can be said
Conversational agents may not improve continued study participation.
Important limitations
Continued study participation is different from continued app use.
Go to notes on statistics and research methods ↓

Cevasco, K. E., Morrison Brown, R. E., Woldeselassie, R., & Kaplan, S.（2024）

Patient Engagement with Conversational Agents in Health Applications 2016–2022: A Systematic Review and Meta-Analysis

Journal of Medical Systems

View the original paper or source

Not finding a difference does not mean there is no effect in every situation. Adverse effects should also be read within the conditions studied.

How YataLogic interprets this

These are design ideas drawn from several studies for practical use. They do not guarantee the same results for your business.

What the study found

An analysis combining studies of robots, chatbots and similar technologies found a small positive association between perceived human-likeness and intention to use. Usability, usefulness, perceived safety and other factors were also involved.

YataLogic’s practical application

Provide practical support alongside a sense of familiarity.

Check the original study and conditions ↑
What the study found

Reviews of pedagogical agents included many analyses showing small positive differences in learning outcomes. Results for motivation and emotion were mixed.

YataLogic’s practical application

Do not treat better learning outcomes as direct evidence of a higher continued-use rate.

Check the original study and conditions ↑

What should you look at in practice?

Assess the following responses separately, according to the job assigned.

  • Did users continue after 7 and 30 days?
  • Did they complete the planned action?
  • Did they return within 7 days of stopping?
  • Did more people disable notifications?
  • Did they renew the contract? Assess this separately from short-term use.

Research notes

The main text can be understood without reading the statistics. Use these notes to check what the figures mean.

Understand the statistics and research methods

Figures supplement the findings by showing measured differences and uncertainty. Different measures cannot be converted into one common improvement percentage.

p-value
An indication of the probability, under the null hypothesis, of obtaining a difference at least as extreme as the observed result. It is not the size of an effect. Even p<.05 need not indicate a practically large effect. Do not equate p<.05 with a large effect.
95% confidence interval (95% CI)
A range expressing uncertainty around an estimate. When an interval crosses 0 for SMD/g or 1 for RR/OR, a clear difference is often difficult to claim. Read the full interval.
Hedges' g / SMD / Cohen's d
A standardized effect size expressing mean differences from different scales in standard-deviation units. 0.39 does not mean a 39% improvement. Describe g=.39 as a standardized difference; do not convert it into a percentage.
RR (risk ratio)
A measure comparing the proportions experiencing an event between conditions. RR=1 indicates similar proportions. RR=.99 [.95,1.03] indicates similar proportions, and the CI also crosses 1.
Correlation r / corrected correlation rc
The degree to which two variables vary together. Correlation alone does not establish causation. Describe the result as an association.
RCT
A comparative trial that randomly assigns participants to groups. Generally strong for causal inference, but the effect concerns the whole intervention. When other functions are added alongside a character, do not call it an isolated character effect.
Meta-analysis
A method for statistically combining results from multiple studies. When studies differ substantially, the average alone is insufficient. Also check I² and differences in populations and interventions.
NS / no significant difference
A statistically clear difference was not found. This does not necessarily prove that the effect is zero. State that no significant difference was found under these conditions.
I²
A measure of variation between study results in a meta-analysis. Higher values make it harder to generalize from the average effect alone. For I²=97%, explicitly state that heterogeneity is very high.

This synthesis is YataLogic’s practical reading of multiple studies. It is not a scientifically established proprietary model.

References

This page covers 5 findings from 5 papers or sources. Different findings from the same paper are presented separately.

  1. Blut, M., Wang, C., Wünderlich, N. V., & Brock, C.（2021）

    Understanding anthropomorphism in service provision: a meta-analysis of physical robots, chatbots, and other AI

    Journal of the Academy of Marketing Science

    View the original paper or source

    Finding discussed: Wanting to use something and continuing to use it are different

  2. Schroeder, N. L., Davis, R. O., & Yang, E.（2025）

    Designing and Learning With Pedagogical Agents: An Umbrella Review

    Journal of Educational Computing Research

    View the original paper or source

    Finding discussed: Retaining knowledge is different from continued use

  3. Chen, S., Fang, Y., Ding, L., Mo, P. K. H., & Wang, Z.（2026）

    Artificial intelligence–based chatbots to enhance medication adherence among patients with non-communicable chronic diseases: Systematic review and meta-analysis

    PLOS Digital Health

    View the original paper or source

    Finding discussed: Medication-support systems show potential, with varied results

  4. Watson, A., Bickmore, T., Cange, A., Kulshreshtha, A., & Kvedar, J.（2012）

    An Internet-Based Virtual Coach to Promote Physical Activity Adherence in Overweight Adults: Randomized Controlled Trial

    Journal of Medical Internet Research

    View the original paper or source

    Finding discussed: Even short-term exercise support depends on the analysis

  5. Cevasco, K. E., Morrison Brown, R. E., Woldeselassie, R., & Kaplan, S.（2024）

    Patient Engagement with Conversational Agents in Health Applications 2016–2022: A Systematic Review and Meta-Analysis

    Journal of Medical Systems

    View the original paper or source

    Finding discussed: Conversational guides may not improve continued participation

What job should a character do?

Use six roles to organize the job, the touchpoint and the measure.

See how this role can be used →

How do companies actually use characters?

Explore business applications by role.

Explore Make it easier to continue cases →
See the research overview of all six roles →

Explore the research and design in more depth

The book Sono Kigyo Character, Imi Aru? examines the research and character design systematically. The book is available in Japanese only.

View the book on Amazon (Japanese only) ↗

What about your business?

Identify the job you want a character to do from your current challenge.

Explore the right job for your character

8 questions · about 2 minutes · no registration

Does your company need a character? →

Apply these ideas to your business

Together, we consider the role based on your business and customer touchpoints.

Talk to YataLogic →

English summaries and any translated source titles are provided by YataLogic. Linked papers and source documents have not been translated and may be in Japanese or another language.