CoChat
Sign Up
Cited by 607Digital Mental Health InterventionsSocial Robot Interaction and HRIArtificial Intelligence in Healthcare and Education

An overview of the features of chatbots in mental health: A scoping review

Open access
Abd-alrazaq, Alaa A., Alajlani, Mohannad, Alalwan, Ali Abdallah, Bewick, Bridgette M., Gardner, Peter, Househ, Mowafa·2019·International Journal of Medical Informatics
DOI 10.1016/j.ijmedinf.2019.103978
Open this paper in CoChat — free account

Abstract

BACKGROUND Chatbots are systems that are able to converse and interact with human users using spoken, written, and visual languages. Chatbots have the potential to be useful tools for individuals with mental disorders, especially those who are reluctant to seek mental health advice due to stigmatization. While numerous studies have been conducted about using chatbots for mental health, there is a need to systematically bring this evidence together in order to inform mental health providers and potential users about the main features of chatbots and their potential uses, and to inform future research about the main gaps of the previous literature. OBJECTIVE We aimed to provide an overview of the features of chatbots used by individuals for their mental health as reported in the empirical literature. METHODS Seven bibliographic databases (Medline, Embase, PsycINFO, Cochrane Central Register of Controlled Trials, IEEE Xplore, ACM Digital Library, and Google Scholar) were used in our search. In addition, backward and forward reference list checking of the included studies and relevant reviews was conducted. Study selection and data extraction were carried out by two reviewers independently. Extracted data were synthesised using a narrative approach. Chatbots were classified according to their purposes, platforms, response generation, dialogue initiative, input and output modalities, embodiment, and targeted disorders. RESULTS Of 1039 citations retrieved, 53 unique studies were included in this review. The included studies assessed 41 different chatbots. Common uses of chatbots were: therapy (n = 17), training (n = 12), and screening (n = 10). Chatbots in most studies were rule-based (n = 49) and implemented in stand-alone software (n = 37). In 46 studies, chatbots controlled and led the conversations. While the most frequently used input modality was written language only (n = 26), the most frequently used output modality was a combination of written, spoken and visual languages (n = 28). In the majority of studies, chatbots included virtual representations (n = 44). The most common focus of chatbots was depression (n = 16) or autism (n = 10). CONCLUSION Research regarding chatbots in mental health is nascent. There are numerous chatbots that are used for various mental disorders and purposes. Healthcare providers should compare chatbots found in this review to help guide potential users to the most appropriate chatbot to support their mental health needs. More reviews are needed to summarise the evidence regarding the effectiveness and acceptability of chatbots in mental health.

Verified 2026-10-06 · 5 references not checked yet

AI snapshot

Written from the paper’s full text.

TL;DR

This scoping review identified 53 studies assessing 41 different chatbots used by individuals for mental health, most commonly for therapy, training, and screening. Chatbots were predominantly rule-based, stand-alone, chatbot-led, embodied, and focused on depression or autism, with the authors concluding that research is nascent and needs more RCTs and standardized outcome/reporting measures.

Key findings

  • Of 1039 citations retrieved, 53 unique studies were included, assessing 41 different chatbots. Common uses were therapy (n=17), training (n=12), and screening (n=10); other uses were self-management (n=7), counselling (n=5), education (n=4), and diagnosis (n=2).
  • Most chatbots were rule-based (n=49), implemented in stand-alone software (n=37), controlled/led the conversation (n=46), and included virtual representations (n=44). The most common input modality was written language only (n=26), while the most common output modality was a combination of written, spoken, and visual languages (n=28).
  • Chatbots most often targeted depression (n=16) or autism (n=10), followed by posttraumatic stress disorder (n=7), anxiety (n=7), substance use disorders (n=5), schizophrenia (n=3), dementia (n=3), stress (n=2), phobia (n=2), eating disorders (n=1), and any mental disorder (n=7).
  • Included studies were mostly quasi-experimental (n=23, 43%) and journal articles (n=35, 66%); 23 studies (43%) were from the USA, and 30 (56.6%) were published between 2016 and 2019. Mean sample size was 75.2 (range 4–454), mean age was 37.1 years, and mean male percentage was 55%; acceptability (n=36) and effectiveness (n=33) were the most commonly assessed outcomes.

Methods

  • Scoping review following PRISMA-ScR. Searches covered MEDLINE, EMBASE, PsycINFO, Scopus, Cochrane Central Register of Controlled Trials, IEEE Xplore, ACM Digital Library, and Google Scholar (first 100 citations), conducted 10–12 May 2019, plus backward and forward reference list checking.
  • Inclusion required primary research on chatbots used by individuals for mental health, on stand-alone software or web browser platforms, excluding robotics, serious games, SMS, telephones, physician/caregiver-specific chatbots, and human-operated dialogue. Peer-reviewed articles, dissertations, conference proceedings, and reports in English were eligible; no restrictions were placed on dialogue initiative, input/output modality, study design, setting, outcome, year, or country.
  • Two reviewers independently selected studies and extracted data, resolving disagreements with a third reviewer. Cohen’s kappa was 0.80 and 0.84 for the two selection steps and 0.74 for data extraction. Data were synthesized narratively, with chatbots classified by purpose, platform, response generation, dialogue initiative, input/output modalities, embodiment, and targeted disorders. No quality assessment was performed.

Limitations

  • The paper states that restricting platforms to stand-alone software and web browser (excluding robotics, serious games, SMS, and telephones) and dialogue initiative to user/system (not human operator) excluded more than 25 studies. It also states that practical constraints prevented searching Web of Science and ProQuest, manual searching, and contacting experts, so some studies were likely missed; excluding non-primary papers may also miss chatbots reported in proposals, book chapters, and conference abstracts.
  • The paper states that it did not assess study quality because scoping reviews usually do not, and it did not summarize included studies’ results because of the review’s aim and the large number of studies assessing different outcomes.
  • Evident in the text: included studies were inconsistent in outcome measurement and in reporting chatbot characteristics, study methods, and sample characteristics; sample size was <50 in 30 studies, many studies recruited non-clinical samples, and less than half of effectiveness studies (14 of 33) used randomized controlled trials, limiting comparability and generalisability.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1016/j.ijmedinf.2019.103978

Cites

49 of 54 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 54 cited sources checked · 0 supported · 0 flagged

  • Findingown data
    § 3.2 Description of included studies

    A clinical sample relating to a specific mental disorder featured in about 58.5% of the studies (n=31).

    “A clinical sample relating to a specific mental disorder featured in about 58.5% of the studies (n=31).”
    Quote in the paperNot a substantial claim
  • Findingreasoning
    § 3.1 Search results

    In total, 53 publications were included in the synthesis.

    “In total, 53 publications were included in the synthesis.”
    Quote in the paperNot a substantial claim
  • Findingown data
    § 3.3 Description of chatbots

    The included studies assessed 41 different chatbots.

    “The included studies assessed 41 different chatbots.”
    Quote in the paperNot a substantial claim
  • Findingown data
    § 3.2 Description of included studies

    The most common outcomes assessed by the included studies were acceptability (n=36) and effectiveness (n=33).

    “The most common outcomes assessed by the included studies were acceptability (n=36) and effectiveness of the chatbots (n=33).”
    Quote in the paperNot a substantial claim
  • Findingown data
    § 3.2 Description of included studies

    The most commonly used study design was quasi-experimental (n=23, 43%).

    “the most commonly used study design was quasi-experimental (n=23, 43%).”
    Quote in the paperNot a substantial claim
  • Classified as probably not a substantial claim. 36
  • May combine two claims. 22
  • The quote only partly states this claim. 1

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 3.1 Search results

    The search process of six bibliographic databases retrieved 1039 citations.

    “the search process of six bibliographic databases retrieved 1039 citations.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 3.2 Description of included studies

    The mean sample size was 75.2, ranging from 4 to 454.

    “The mean sample size was 75.2, it ranged from 4 to 454.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 3.2 Description of included studies

    The sample size was 200 or more in only 4 studies.

    “whereas it was 200 or more in only 4 studies.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 3.1 Search results

    After scanning titles and abstracts, 505 citations were excluded.

    “After scanning their titles and abstracts, 505 citations were excluded.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Findingown data
    § 3.1 Search results

    By reading the full text of the 125 remaining citations, 43 publications were included.

    “By reading the full text of the 125 remaining citations, 43 publications were included.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 3.2 Description of included studies

    The sample size was less than 50 in 30 studies.

    “The sample size was less than 50 in 30 studies”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 3.1 Search results

    After removing 409 duplicates, 630 unique titles and abstracts remained.

    “After removing 409 duplicates, 630 unique titles and abstracts remained.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data
    § 3.2 Description of included studies

    Sex of participants was reported in 41 studies; the mean male percentage was 55%.

    “Sex of participants was reported in 41 studies, the mean of male percentage in all those studies was 55%.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 3.2 Description of included studies

    More than half of the studies were published between 2016 and 2019 (n=30, 56.6%).

    “More than half of the studies were published between 2016 and 2019 (n=30, 56.6%).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data
    § 3.2 Description of included studies

    Age of participants was reported in 33 studies; the mean age was 37.1 years.

    “Age of participants was reported in 33 studies, the mean age of participants in all those studies was 37.1 years.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 3.2 Description of included studies

    About 43% of studies were published in the USA (n=23).

    “about 43% of studies were published in the USA (n=23).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 3.2 Description of included studies

    Participants were recruited from clinical settings (n=20), community (n=16), or educational settings (n=20).

    “Participants were recruited from either clinical settings (n=20), community (n=16), or educational settings (n=20).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data
    § 3.1 Search results

    Six additional studies were identified from backward reference list checking, and four studies were identified from forward reference list checking.

    “Six additional studies were identified from backward reference list checking, and four studies were identified from forward reference list checking.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 3.2 Description of included studies

    About two thirds of studies were journal articles (n=35, 66%).

    “About two thirds of studies were journal articles (n=35, 66%).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 3.3 Description of chatbots

    Chatbots were given names in 39 studies.

    “The chatbots were given names in 39 studies”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 4.1 Principal findings

    The majority of chatbots (92.5%) depended on only decision trees to generate their responses; only 7.5% used machine learning approaches.

    “The majority of chatbots (92.5%) depended on only decision trees to generate their responses; only 7.5% used machine learning approaches.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 3.3 Description of chatbots

    Chatbots used a combination of spoken and visual languages in 11 studies.

    “a combination of spoken and visual languages (n=11)”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 3.3 Description of chatbots

    Among the machine learning studies, one used supervised machine learning, another used reinforcement learning, and two did not specify the machine learning approaches used.

    “while one study used supervised machine learning, another study used reinforcement learning. However, the remaining two studies did not specify the machine learning approaches used.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 3.3 Description of chatbots

    Chatbots used only written language in 9 studies.

    “only written language (n=9)”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 3.3 Description of chatbots

    Users could interact with chatbots using a combination of written and visual languages in 2 studies.

    “a combination of written and visual languages (2).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 3.3 Description of chatbots

    Chatbots used a combination of written and visual languages in 5 studies.

    “a combination of written and visual languages (n=5)”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 3.3 Description of chatbots

    Users could interact with chatbots using a combination of written and spoken languages in 7 studies.

    “a combination of written and spoken languages (n=7)”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 4.1 Principal findings

    We identified 53 studies that assessed 41 different chatbots.

    “We identified 53 studies that assessed 41 different chatbots.”
    • Also stated in: § 5 Conclusion · plain
      “We identified 41 different chatbots in mental health.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Findingreasoning
    § 4.3.2 Research implications

    It was challenging for the reviewers to identify chatbots' platforms, their input and output modality, type of response generation, and dialogue initiative.

    “For example, it was challenging for the reviewers to identify chatbots' platforms, their input and output modality, type of response generation, and dialogue initiative.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation
    § 2 Methods

    This scoping review followed PRISMA-ScR guidelines.

    “This scoping review followed guidelines recommended by the PRISMA Extension for Scoping Reviews (PRISMA-ScR) [23].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [23] [23] A.C. Tricco, E. Lillie, W. Zarin, K.K. O'Brien, H. Colquhoun, D. Levac, D. Moher, M.D. Peters, T. Horsley, L. Weeks, PRISMA extension for scoping reviews (PRISMA-ScR): checklist and explanation, Annals of internal medicine, 169 (2018) 467-473. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data
    § 2.4 Data extraction

    A data extraction form was developed and piloted using five included studies.

    “we developed a data extraction form and piloted it using five included studies (Appendix B).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data
    § 2.3 Study selection

    Inter-coder agreement for study selection was 0.80 in the first step and 0.84 in the second step, indicating very good agreement.

    “Inter-coder agreement between both reviewers w as assessed using Cohen's kappa [24], which was 0.80 and 0.84 in the first and second step of the selection process, respectively, indicating a very good agreement [25].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [24] [24] J. Higgins, J. Deeks, Chapter 7: selecting studies and collecting data, in: J. Higgins, S. Green (Eds.) Cochrane Handbook for Systematic Reviews of Interventions, John Wiley &amp; Sons, Sussex, UK, 2008, pp. 151-185.

      Not checked in the free run.

    • cites [25] [25] D.G. Altman, Practical statistics for medical research, CRC press1990.

      Not checked in the free run.

  • May combine two claims.

    Premisecitation
    § 1 Introduction

    The second review used limited search terms and identified a relatively low number of studies (10).

    “That review used limited search terms and therefore identified a relatively low number of studies (i.e. 10) [6].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [6] [6] A.N. Vaidyam, H. Wisniewski, J.D. Halamka, M.S. Kashavan, J.B. Torous, Chatbots and Conversational Agents in Mental Health: A Review of the Psychiatric Landscape, The Canadian Journal of Psychiatry, (2019) 0706743719828977. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 1 Introduction

    According to the World Health Organization, 29% of the 15,000 mobile health apps focused on mental health.

    “According to the World Health Organization [9], 29% of the 15,000 mobile health apps focused on mental health.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [9] [9] E. Anthes, Mental health: there's an app for that, Nature News, 532 (2016) 20. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductcitation
    § 2.4 Data extraction

    Inter-coder agreement for data extraction was 0.74, indicating good agreement.

    “Inter-coder agreement between the reviewers was good ( Cohen's kappa = 0.74) [25].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [25] [25] D.G. Altman, Practical statistics for medical research, CRC press1990.

      Not checked in the free run.

  • May combine two claims.

    Premisecitation
    § 1 Introduction

    There is a global shortage of mental health workers, with demand out-stripping service provision.

    “There is a global shortage of mental health workers, with demand out-stripping service provision [6].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [6] [6] A.N. Vaidyam, H. Wisniewski, J.D. Halamka, M.S. Kashavan, J.B. Torous, Chatbots and Conversational Agents in Mental Health: A Review of the Psychiatric Landscape, The Canadian Journal of Psychiatry, (2019) 0706743719828977. source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation
    § 1 Introduction

    Mental health services do not reach about 55% of people in developed countries and 85% of people in developing countries.

    “mental health services do not reach about 55% and 85% of people in developed and developing countries, respectively [9].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [9] [9] E. Anthes, Mental health: there's an app for that, Nature News, 532 (2016) 20. source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation
    § 1 Introduction

    Lack of access to mental health services may lead to suicidal behaviour, resulting in increasing mortality.

    “The lack of access to mental health services may lead to suicidal behaviour, resulting in increasing mortality [10].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [10] [10] R.D. Hester, Lack of access to mental health services contributing to the high suicide rates among veterans, International journal of mental health systems, 11 (2017) 47. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based).

    Premisereasoning
    § 4.1 Principal findings

    Web-based chatbots are considered more appropriate than stand-alone software.

    “This was surprising because web-based chatbots are considered more appropriate than stand stand-alone software for the following two reasons.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation
    § 4.1 Principal findings

    In comparison with artificial intelligence chatbots, rule-based chatbots are more simple to develop and less prone to errors.

    “In comparison with artificial intelligence chatbots, rule-based chatbots are more simple to develop and less prone to errors as their responses are predetermined and they do not need to create new responses [41].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [41] [41] K. Shridhar, Rule based bots vs AI bots, Medium.com, 2017.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 4.1 Principal findings

    Rule-based chatbots are considered more secure and responsible than artificial intelligence chatbots.

    “Further, rule-based chatbots are considered more secure and responsible than artificial intelligence chatbots [42].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [42] [42] Hubtype, Rule-Based vs AI Chatbots, 2018.

      Not checked in the free run.

  • May combine two claims.

    Interpretationcitation
    § 4.1 Principal findings

    The focus on depression and autism may be attributed to depression and anxiety being the most common mental disorder around the world, and chatbots being reported to be effective in improving social skills for patients with autism.

    “This may be attributed to the fact that depression and anxiety are the most common mental disorder around the world [44, 45], and chatbots are reported to be effective in improving social skills for patients with autism [14, 17, 46-48].”
    Quote in the paperNot a substantial claim0 of 7 sources checked
    • cites [44] [44] A.J. Ferrari, F.J. Charlson, R.E. Norman, S.B. Patten, G. Freedman, C.J. Murray, T. Vos, H.A. Whiteford, Burden of depressive disorders by country, sex, age, and year: findings from the global burden of disease study 2010, PLoS medicine, 10 (2013) e1001547. source

      Not checked in the free run.

    • cites [45] [45] H. Ritchie, M. Roser, Mental Health, Our World in Data, 2018.

      Not checked in the free run.

    • cites [14] [14] M.R. Ali, Z. Rasazi, A.A. Mamun, R. Langevin, R. Rawassizadeh, L. Schubert, M.E. Hoque, A Virtual Conversational Agent for Teens with Autism: Experimental Results and Design Lessons, arXiv preprint arXiv:1811.03046, (2018).

      Not checked in the free run.

    • cites [17] [17] S.Z. Razavi, M.R. Ali, T.H. Smith, L.K. Schubert, M.E. Hoque, The LISSA virtual human and ASD teens: An overview of initial experiments, International Conference on Intelligent Virtual Agents, Springer, 2016, pp. 460-463.

      Not checked in the free run.

    • cites [46] [46] M. Milne, M.H. Luerssen, T.W. Lewis, R.E. Leibbrandt, D.M.W. Powers, Development of a virtual agent based social tutor for children with autism spectrum disorders, The 2010 International Joint Conference on Neural Networks (IJCNN), 2010, pp. 1-9.

      Not checked in the free run.

    • cites [47] [47] H. Tanaka, H. Negoro, H. Iwasaka, S. Nakamura, Embodied conversational agents for multimodal automated social skills training in people with autism spectrum disorders, PLoS ONE Vol 12(8), 2017, ArtID e0182151, 12 (2017).

      Not checked in the free run.

    • cites [48] [48] H. Tanaka, S. Sakti, G. Neubig, T. Toda, H. Negoro, H. Iwasaka, S. Nakamura, Automated Social Skills Trainer, Proceedings of the 20th International Conference on Intelligent User Interfaces - IUI '15, 2015, pp. 17-27.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Premisenone given
    § 4.1 Principal findings

    To use web-based chatbots, users do not need to install a specific application to their devices, thereby reducing the risk of breaching their privacy.

    “Firstly, to use web-based chatbots, users do not need to install a specific application to their devices, thereby reducing the risk of breaching their privacy.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 4.1 Principal findings

    The rareness of chatbots led by users was also found in another review.

    “The rareness of chatbots led by users was also found in another review [29].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [29] [29] L. Laranjo, A.G. Dunn, H.L. Tong, A.B. Kocaballi, J. Chen, R. Bashir, D. Surian, B. Gallego, F. Magrabi, A.Y. Lau, Conversational agents in healthcare: a systematic review, Journal of the American Medical Informatics Association, 25 (2018) 1248-1258. source

      Not checked in the free run.

  • May combine two claims.

    Interpretationcitation
    § 4.1 Principal findings

    The finding that chatbots led and controlled the conversation was expected since most chatbots used rule-based approaches, which restrict user input to predefined words and replies, thereby preventing users from leading or controlling the dialogue.

    “This was expected since most chatbots used rule-based approaches, which restrict user input to predefined words and replies, thereby, prevent them to lead or control the dialogue [28].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [28] [28] M.F. McTear, Spoken dialogue technology: enabling the conversational user interface, ACM Computing Surveys (CSUR), 34 (2002) 90-169. source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning
    § 4.2.1 Strengths

    This made our review more comprehensive than the two previous reviews [6, 22]; the former focused on some mental disorders and the latter focused on only embodied chatbots.

    “This made our review more comprehensive than the two previous reviews [6, 22]; where the former focused on some mental disorders and the latter focused on only embodied chatbots.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [6] [6] A.N. Vaidyam, H. Wisniewski, J.D. Halamka, M.S. Kashavan, J.B. Torous, Chatbots and Conversational Agents in Mental Health: A Review of the Psychiatric Landscape, The Canadian Journal of Psychiatry, (2019) 0706743719828977. source

      Not checked in the free run.

    • cites [22] [22] S. Provoost, H.M. Lau, J. Ruwaard, H. Riper, Embodied conversational agents in clinical psychology: a scoping review, Journal of Medical Internet Research, 19 (2017) e151. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning
    § 4.2.1 Strengths

    The data extraction process was conducted by two reviewers independently to reduce selection bias.

    “The study selection process and data extraction process were conducted by two reviewers independently to reduce selection bias.”
    Quote in the paperNot a substantial claim

  • The quote only partly states this claim.

    Study design and conductreasoning
    § 4.2.1 Strengths

    Searching Google Scholar and performing backward and forward reference list checking enabled us to minimise the risk of publication bias.

    “This enabled us to minimise the risk of publication bias.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisereasoning
    § 4.2.2 Limitations

    We excluded systems controlled by human operator as they are more similar to telemedicine systems than chatbots in terms of system design.

    “We excluded systems that are controlled by human operator as they are more similar to telemedicine systems than chatbots in terms of system design.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 4.2.2 Limitations

    The above-mentioned restriction was also applied in the previous two reviews.

    “The above-mentioned restriction was also applied in the previous two reviews [6, 22].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [6] [6] A.N. Vaidyam, H. Wisniewski, J.D. Halamka, M.S. Kashavan, J.B. Torous, Chatbots and Conversational Agents in Mental Health: A Review of the Psychiatric Landscape, The Canadian Journal of Psychiatry, (2019) 0706743719828977. source

      Not checked in the free run.

    • cites [22] [22] S. Provoost, H.M. Lau, J. Ruwaard, H. Riper, Embodied conversational agents in clinical psychology: a scoping review, Journal of Medical Internet Research, 19 (2017) e151. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data
    § 4.2.1 Strengths

    Agreement between reviewers was very good for the study selection process.

    “Agreement between reviewers was very good for the study selection process and good for the data extraction process.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data
    § 5 Conclusion

    Research regarding chatbots in mental health is emerging, apparent from publication year of studies (most studies were published in the last 9 years), lack of randomized controlled trials, and inconsistency of studies in outcome measures and reporting.

    “Research regarding chatbots in mental health is emerging, and this was apparent from publication year of studies (most studies were published in the last 9 years), lack of randomized controlled trials, and the inconsistency of the studies in terms of outcome measures and reporting of characteristics of chatbots and study design.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning
    § 4.2.2 Limitations

    We adapted several taxonomies [28-30] to characterise the chatbots in our review.

    “Therefore, we adapted several taxonomies [28-30] to characterise the chatbots in our review.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [28] [28] M.F. McTear, Spoken dialogue technology: enabling the conversational user interface, ACM Computing Surveys (CSUR), 34 (2002) 90-169. source

      Not checked in the free run.

    • cites [29] [29] L. Laranjo, A.G. Dunn, H.L. Tong, A.B. Kocaballi, J. Chen, R. Bashir, D. Surian, B. Gallego, F. Magrabi, A.Y. Lau, Conversational agents in healthcare: a systematic review, Journal of the American Medical Informatics Association, 25 (2018) 1248-1258. source

      Not checked in the free run.

    • cites [30] [30] J. Chu-Carroll, M.K. Brown, Tracking initiative in collaborative dialogue interactions, Proceedings of the eighth conference on European chapter of the Association for Computational Linguistics, Association for Computational Linguistics, 1997, pp. 262-270.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data
    § 4.2.1 Strengths

    Agreement between reviewers was good for the data extraction process.

    “Agreement between reviewers was very good for the study selection process and good for the data extraction process.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given
    § 4.2.2 Limitations

    Owing to practical constraints, we could not search other databases that cover both fields, conduct manual search, and contact experts.

    “Owing to practical constraints, we could not search other databases that cover both fields (e.g. Web of Science and ProQuest), conduct manual search, and contact experts.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductreasoning
    § 4.2.2 Limitations

    We restricted the platforms of chatbots to stand-alone software and web browser (but not robotics, serious games, SMS, nor telephones) and restricted their type of dialogue initiative to users and system (but not human operator).

    “Although chatbots were included in this review regardless of their purpose, type of response generation, input and output modalities, embodiment, and targeted disorders, we restricted their platforms to stand-alone software and web browser (but not robotics, serious games, SMS, nor telephones) and restricted their type of dialogue initiative to users and system (but not human operator).”
    Quote in the paperSubstantial claim
51 of 51

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Recent studies have shown promising results of AI applications including suicide risk assessment and risk prediction, identification of mental illness predictors, mental health monitoring, psychoeducation and psychotherapy delivery, therapist training, mental healthcare personalization, mental health triage and decision-making, and promoting treatment engagement.

    “Nevertheless, recent studies have shown promising results of AI applications including suicide risk assessment and risk prediction, identification of mental illness predictors, mental health monitoring, psychoeducation and psychotherapy delivery, therapist training, mental healthcare personalization, mental health triage and decision-making, and promoting treatment engagement (6, 17–23).”

    From Examining the role of AI technology in online mental healthcare: opportunities, challenges, and implications, a mixed-methods review (2024) · cites this paper as [17]

    Section 3.3 reports chatbots used for therapy, education, screening for symptoms of depression and suicide, and other mental-health purposes. But the review does not establish promising results across the full list of AI applications: its training examples concern patients’ social and job-interview skills, not therapist training, and Section 4.1 says most chatbots were rule-based. The cited passages do not substantiate risk prediction, personalization, or triage and decision-making. (LLM, conf=0.88)

    Experimental — accuracy not yet measured

  • AI technology can help address online mental healthcare issues by supporting a versatile, scalable, powerful approach that can analyze and respond to large amounts of data and adapt to individual healthcare needs.

    “AI technology can help address these issues by supporting a versatile, scalable, and powerful approach that can analyze and respond to large amounts of data and adapt to individual healthcare needs (6, 17–23).”

    From Examining the role of AI technology in online mental healthcare: opportunities, challenges, and implications, a mixed-methods review (2024) · cites this paper as [17]

    The Introduction says chatbots may increase access to mental-health interventions, and §4.3.1 says AI chatbots can respond to complex questions and learn from gathered data. That section also says web-based chatbots could reach many users. The paper does not establish that AI chatbots can process and respond to large volumes of data or adapt care to each individual's needs. (LLM, conf=0.85)

    Experimental — accuracy not yet measured

  • Because of the novelty of AI technology, there are unanswered questions and considerations such as limitations in language interpretation, biases in interaction with patients from diverse backgrounds, and unanswered ethical, patient safety, and health policy considerations.

    “However, because of the novelty of this technology, there are several unanswered questions and considerations such as limitations in the interpretation of language, biases in the interaction with patients from diverse backgrounds, and unanswered ethical, patient safety, and health policy considerations (6, 17–23).”

    From Examining the role of AI technology in online mental healthcare: opportunities, challenges, and implications, a mixed-methods review (2024) · cites this paper as [17]

    The Conclusion (page unavailable) describes mental-health chatbot research as emerging. Sections 4.1 and 4.3.1 (pages unavailable) describe limits on rule-based responses and note that AI chatbots are prone to errors, offering some support for the concern about language interpretation. Our copy does not address bias toward patients from diverse backgrounds or unanswered ethical, patient-safety, and health-policy questions. (LLM, conf=0.85)

    Experimental — accuracy not yet measured

  • Two recent reviews by Boucher et al. (2021) and Abd-alrazaq et al. (2019) reported that chatbots had shown promising applications including care assistance, symptom screening and monitoring, and therapy delivery.

    “For instance, two recent reviews by Boucher et al. (2021) and Abd-alrazaq et al. (2019), reported that chatbots had shown promising applications including care assistance, symptom screening and monitoring, and therapy delivery.”

    From Examining the role of AI technology in online mental healthcare: opportunities, challenges, and implications, a mixed-methods review (2024) · cites this paper as [17]

    Section 3.3 reports chatbots used for therapy, symptom screening, self-management and counselling, supporting several of the applications listed. It does not identify symptom monitoring as a use. The Abstract also says more reviews are needed on effectiveness, so the description of the applications as “promising” should not be read as an established effectiveness finding. (LLM, conf=0.85)

    Experimental — accuracy not yet measured

  • These reviews noted several potential benefits in terms of providing a low-resource intensity alternative for individuals experiencing less severe mental health symptoms.

    “These reviews recognized the novelty of this technology and the scarcity of evidence but noted several potential benefits in terms of providing a low-resource intensity alternative for individuals experiencing less severe mental health symptoms (6, 17).”

    From Examining the role of AI technology in online mental healthcare: opportunities, challenges, and implications, a mixed-methods review (2024) · cites this paper as [17]

    The Introduction describes chatbots as a potential way to expand access to mental health interventions amid shortages of mental health workers, supporting the low-resource alternative in broad terms. It does not identify people with less severe symptoms as the intended group or establish that chatbots require fewer resources than other interventions. (LLM, conf=0.83)

    Experimental — accuracy not yet measured

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Clinical-setting study count differs from Table 1

    Minor

    The narrative gives 20 studies recruited from clinical settings, but Table 1 lists 21 for the clinical setting.

  • 2016-2019 study count differs from Table 1

    Minor

    The narrative reports 30 studies published in 2016-2019, while Table 1 reports 31.

  • Search database count differs between abstract and methods

    Minor

    The abstract says seven databases were searched, but the methods list eight, adding Scopus.

  • Results give a third database count

    Minor

    The results say citations came from six bibliographic databases, whereas the methods list eight searched databases. The text does not identify which two were excluded from this count.

4 of 4

Cites

Flagged first, then by citations

Research works printed but not linked: 2

  1. Luerssen, 2018, AustraliaVirtual Agents as a Service: Applications in Healthcare, Proceedings of the 18th International Conference on Intelligent Virtual Agents, ACM, Sydney, NSW2018, Virtual Agents as a Service: Applications in Healthcare, Proceedings of the 18th International Conference on Intelligent Virtual Agents, ACM, Sydney, NSW
  2. Higgins, 2008, Chapter 7: selecting studies and collecting data, 151

Cited by

Sorted by citations

Related

Authors
Abd-alrazaq, A. A., Alajlani, M., Alalwan, A. A., Bewick, B. M., Gardner, P., & Househ, M.
Affiliations
  • Hamad bin Khalifa University (QA)
  • University of Warwick (GB)
  • Al-Balqa Applied University (JO)
  • University of Leeds (GB)
Corresponding author
Househ, Mowafa
ORCID
Venue
International Journal of Medical Informatics
Publisher
Elsevier BV
Published
2019-12
Volume
132
Pages
103978
ISSN
1386-5056 (ISSN-L)
DOI
10.1016/j.ijmedinf.2019.103978
OpenAlex
W2977128309
Semantic Scholar
ee757d582438057b79e60bf410bc459279398d73
Cited by
605
References
53
Citations by year
2026: 122 · 2025: 117 · 2024: 112 · 2023: 108 · 2022: 65 · 2021: 55 · 2020: 24 · 2019: 1 · 2013: 1
FWCI
5.07
Type
journal article
Language
en
Open access
Licence
Locations
  • International Journal of Medical Informatics (publishedVersion)
  • PubMed (publishedVersion)
  • White Rose Research Online (University of Leeds, The University of Sheffield, University of York) (acceptedVersion)
  • Warwick Research Archive Portal (University of Warwick) (submittedVersion, cc-by-nc-nd)
Topics
  • Social Sciences › Psychology › Applied Psychology › Digital Mental Health Interventions
  • Social Sciences › Psychology › Social Psychology › Social Robot Interaction and HRI
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
Keywords
chatbots, mental health, depression, autism spectrum disorder, rule-based chatbots, virtual agents, chatbot therapy, mental health screening, mental health training, stigma, embodiment, dialogue control, chatbot response generation, mental illness, chatbot evaluation metrics
MeSH terms
Behavior Therapy, Mental Disorders, Communication, Humans, Mental Health, Telemedicine

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-06.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-06.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review