pith. sign in

arxiv: 2602.19401 · v2 · submitted 2026-02-23 · 💻 cs.HC

Reassurance Robots: OCD in the Age of Generative AI

Grace Barkhuff This is my paper

Pith reviewed 2026-05-15 21:09 UTC · model grok-4.3

classification 💻 cs.HC
keywords OCDgenerative AIreassuranceobsessionscompulsionsmental healthAI designChatGPT
0
0 comments X

The pith

Generative AI systems harm people with OCD by serving as endless reassurance providers.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper explores how generative AI tools like ChatGPT are creating new challenges for individuals with Obsessive Compulsive Disorder. By examining posts on an OCD-focused online forum, it identifies fresh patterns of obsessions and compulsions centered around AI interactions. The core argument is that these AI systems function as Reassurance Robots, supplying constant validation that sustains rather than relieves the distress of OCD. Readers should care because as AI becomes more integrated into everyday life, its design choices could either mitigate or intensify mental health struggles for those affected by OCD. The author calls for intentional changes in how future AI is built to address this issue.

Core claim

Through an exploratory qualitative analysis of 100 Reddit posts related to AI on a popular OCD subreddit, the author finds that generative AI introduces novel obsessions and compulsions involving AI, and argues that GenAI in its current form harms individuals with OCD by becoming Reassurance Robots, recommending that future designs of GenAI must take OCD into account.

What carries the argument

Reassurance Robots, referring to generative AI systems that provide repeated, on-demand reassurance to individuals with OCD, thereby reinforcing their compulsive behaviors instead of helping to manage them.

Load-bearing premise

The 100 Reddit posts from a single OCD subreddit represent the typical experiences of people with OCD interacting with AI, and the analysis distinguishes true harm caused by AI from mere correlations.

What would settle it

A longitudinal study tracking OCD symptom changes in individuals before and after regular use of generative AI chatbots would test whether AI interactions worsen symptoms.

Figures

Figures reproduced from arXiv: 2602.19401 by Grace Barkhuff.

Figure 1
Figure 1. Figure 1: The OCD Cycle on mental health has been in the context of depression and suicidality [6], however, given the high prevalence of OCD, it is critical that research study the impact of GenAI on OCD as well. This paper contributes what I believe to be the first publication exploring the ways in which GenAI has led to novel presentations of OCD through AI-based obsessions and compulsions. In the discussion, I a… view at source ↗
read the original abstract

Obsessive Compulsive Disorder (OCD) is a mental health disorder characterized by distressing repetitive patterns of thought, called obsessions, and behaviors aimed to reduce the distress, called compulsions. The explosion of artificial intelligence into the modern zeitgeist through the introduction of generative AI (GenAI) systems such as ChatGPT has led to novel obsessions and compulsions involving AI in individuals with OCD. Through an exploratory qualitative analysis of 100 Reddit posts related to AI on a popular subreddit for OCD, I examine ways AI is impacting the presentation of OCD, including novel examples of AI-based obsessions and compulsions. I argue that GenAI in its current form harms individuals with OCD by becoming "Reassurance Robots," and that future designs of GenAI must take OCD into account. I recommend further work explore the intersection between OCD and GenAI.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper conducts an exploratory qualitative analysis of 100 Reddit posts from a single OCD subreddit to identify novel AI-related obsessions and compulsions, arguing that current generative AI systems function as 'Reassurance Robots' that harm individuals with OCD and that future GenAI designs must incorporate OCD considerations.

Significance. If the patterns hold beyond the sampled posts, the work surfaces a timely HCI concern at the intersection of generative AI and mental health, offering preliminary observations that could inform more responsible AI interface design to avoid reinforcing compulsive reassurance-seeking.

major comments (3)
  1. [Methods] Methods section: selection criteria, search terms, date range, and inclusion/exclusion rules for the 100 posts are not described, so it is impossible to assess sampling bias or replicability of the qualitative corpus.
  2. [Results/Discussion] Results and Discussion: the claim that GenAI 'harms' users by becoming Reassurance Robots rests on self-selected posts without baseline symptom measures, comparison groups, or controls for pre-existing reassurance-seeking behavior expressed through any interactive medium; this leaves the causal inference unsupported.
  3. [Abstract/Methods] Abstract and Methods: no information is given on coding scheme, number of coders, inter-rater reliability, or steps taken to mitigate researcher bias in interpreting the posts as evidence of unique GenAI-induced harm.
minor comments (2)
  1. [Introduction] The term 'Reassurance Robots' is introduced without an explicit operational definition or examples tied to specific post excerpts.
  2. [Discussion] The manuscript would benefit from a limitations subsection that explicitly addresses the single-subreddit, self-report nature of the data.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for their constructive feedback on our exploratory qualitative study. We have carefully considered each major comment and will revise the manuscript to improve methodological transparency while preserving the preliminary nature of the observations. Below we respond point by point.

read point-by-point responses

  1. Referee: [Methods] Methods section: selection criteria, search terms, date range, and inclusion/exclusion rules for the 100 posts are not described, so it is impossible to assess sampling bias or replicability of the qualitative corpus.

    Authors: We agree that the original Methods section omitted these details. In the revision we will add a dedicated subsection describing the data collection: search terms included combinations of 'ChatGPT', 'generative AI', 'AI', 'LLM' with OCD-related terms; the date range was November 2022–March 2023; inclusion required posts describing personal AI-related obsessions or compulsions, while exclusion removed non-English posts, purely informational queries, and duplicates. This addition will allow readers to evaluate sampling and replicability. revision: yes

  2. Referee: [Results/Discussion] Results and Discussion: the claim that GenAI 'harms' users by becoming Reassurance Robots rests on self-selected posts without baseline symptom measures, comparison groups, or controls for pre-existing reassurance-seeking behavior expressed through any interactive medium; this leaves the causal inference unsupported.

    Authors: We accept that the current wording implies stronger causation than the data support. The study is exploratory and draws only on self-reported patterns in public posts. We will revise the language throughout Results and Discussion to state that the observed patterns 'suggest' or 'raise the possibility' that current GenAI interfaces may reinforce reassurance-seeking, rather than asserting definitive harm. A new Limitations paragraph will explicitly note the absence of baseline measures, comparison groups, and controls for pre-existing behavior. revision: partial

  3. Referee: [Abstract/Methods] Abstract and Methods: no information is given on coding scheme, number of coders, inter-rater reliability, or steps taken to mitigate researcher bias in interpreting the posts as evidence of unique GenAI-induced harm.

    Authors: The analysis was performed by a single author using an inductive thematic approach. We will expand the Methods section to detail the coding process (open coding followed by axial coding to surface AI-specific themes), the iterative refinement of the codebook, and bias-mitigation steps including reflexive memoing and consultation with two OCD clinicians for face validity. Because the work was single-coder, inter-rater reliability statistics are not applicable; we will state this limitation explicitly. revision: yes

standing simulated objections not resolved
  • The archival Reddit dataset inherently precludes collection of baseline symptom measures or matched comparison groups; these controls cannot be added retrospectively without new data collection.

Circularity Check

0 steps flagged

No circularity: qualitative analysis derives directly from described data patterns

full rationale

The paper is an exploratory qualitative study of 100 Reddit posts with no equations, fitted parameters, derivations, or load-bearing self-citations. The central claim that GenAI acts as 'Reassurance Robots' is presented as an interpretation of observed patterns in the posts, without any reduction by construction to prior fitted values or self-referential definitions. The argument is self-contained as a descriptive analysis.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The paper rests on the domain assumption that self-reported Reddit content from an OCD community reliably indicates real-world harm from AI, and introduces the conceptual entity 'Reassurance Robots' without independent falsifiable evidence beyond the posts themselves.

axioms (1)
  • domain assumption Reddit posts from a popular OCD subreddit accurately reflect the experiences of individuals with OCD interacting with generative AI
    The study treats user-generated content as representative without clinical validation or controls for self-selection bias.
invented entities (1)
  • Reassurance Robots no independent evidence
    purpose: Metaphor for generative AI systems that provide compulsive reassurance to OCD-related obsessions
    Conceptual label introduced to frame the observed interaction; no external validation or falsifiable prediction provided.

pith-pipeline@v0.9.0 · 5435 in / 1263 out tokens · 37419 ms · 2026-05-15T21:09:28.306182+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

31 extracted references · 31 canonical work pages

  1. [1]

    [n. d.]. Obsessive-Compulsive Disorder (OCD) - National Institute of Mental Health (NIMH). https://www.nimh.nih. gov/health/statistics/obsessive-compulsive-disorder-ocd

    work page

  2. [2]

    [n. d.]. OCD Treatment Guide: Best Evidence-Based Therapies, Medications, and New Advances. https://iocdf.org/ocd- treatment-guide/

    work page

  3. [3]

    [n. d.]. What is OCD? https://iocdf.org/about-ocd/

    work page

  4. [4]

    Substance Abuse and Mental Health Services Administration. 2016. Table 3.13, DSM-IV to DSM-5 Obsessive-Compulsive Disorder Comparison. https://www.ncbi.nlm.nih.gov/books/NBK519704/table/ch3.t13/ 8 Grace Barkhuff

    work page 2016

  5. [5]

    Umberto Albert, Alessandra Baffa, and Giuseppe Maina. 2017. Family accommodation in adult obsessive-compulsive disorder: clinical perspectives.Psychology Research and Behavior Management10 (2017), 293–304. doi:10.2147/PRBM. S124359

    work page doi:10.2147/prbm 2017

  6. [6]

    Ikram Amine, Imane Lmati, Houda Anoun, and Fatimazahra Ammor. 2025. Use, Performance, and Limitations of Large Language Models in Mental Health: A Systematic Review. In2025 International Conference on Intelligent Systems: Theories and Applications (SITA). 1–8. doi:10.1109/SITA67914.2025.11273691

    work page doi:10.1109/sita67914.2025.11273691 2025

  7. [7]

    Fjolla Arifi. [n. d.]. What is the OCD cycle? The four steps of OCD. https://www.treatmyocd.com/what-is-ocd/info/ocd- stats-and-science/the-ocd-cycle-visualized-how-the-condition-works

    work page

  8. [8]

    Amy Bruckman. 2002. Studying the amateur artist: A perspective on disguising data collected in human subjects research on the Internet.Ethics and Information Technology4, 3 (Sept. 2002), 217–231. doi:10.1023/A:1021316409277

    work page doi:10.1023/a:1021316409277 2002

  9. [9]

    Cohn, Laszlo A

    Jeffrey F. Cohn, Laszlo A. Jeni, Itir Onal Ertugrul, Donald Malone, Michael S. Okun, David Borton, and Wayne K. Goodman. 2018. Automated Affect Detection in Deep Brain Stimulation for Obsessive-Compulsive Disorder: A Pilot Study. InProceedings of the 20th ACM International Conference on Multimodal Interaction(Boulder, CO, USA)(ICMI ’18). Association for C...

    work page doi:10.1145/3242969.3243023 2018

  10. [10]

    Coles, Richard G

    Meredith E. Coles, Richard G. Heimberg, and Barry D. Weiss. 2013. The public’s knowledge and beliefs about obsessive compulsive disorder.Depression and Anxiety30, 8 (Aug. 2013), 778–785. doi:10.1002/da.22080

    work page doi:10.1002/da.22080 2013

  11. [11]

    2026.Basics of Qualitative Research (3rd ed.): Techniques and Procedures for Developing Grounded Theory

    Juliet Corbin and Anselm Strauss. 2026.Basics of Qualitative Research (3rd ed.): Techniques and Procedures for Developing Grounded Theory. SAGE Publications, Inc., Thousand Oaks, California. doi:10.4135/9781452230153

    work page doi:10.4135/9781452230153 2026

  12. [12]

    Rebecca Deusser, Andrea McCracken, Stephanie Cogen, Boris Litvin, Sanjaya Saxena, Nicolás Tentoni, Rachel Crofut, and Jordan Arfanakis. 2025. America’s OCD Care Crisis. https://iocdf.org/wp-content/uploads/2025/12/Full-Report- Americas-OCD-Care-Crisis-12-9-2025.pdf

    work page 2025

  13. [13]

    David Eigen, Daniel Grollman, David Laidlaw, Benjamin Greenberg, and Erin Einbinder. 2004. Visualizing deep brain stimulation settings in obsessive compulsive disorder. InACM SIGGRAPH 2004 Posters(Los Angeles, California) (SIGGRAPH ’04). Association for Computing Machinery, New York, NY, USA, 110. doi:10.1145/1186415.1186545

    work page doi:10.1145/1186415.1186545 2004

  14. [14]

    David Hammer and Leema K. Berland. 2014. Confusing Claims for Data: A Critique of Common Practices for Presenting Qualitative Research on Learning.Journal of the Learning Sciences23, 1 (2014), 37–46. doi:10.1080/10508406.2013.802652

    work page doi:10.1080/10508406.2013.802652 2014

  15. [15]

    David Francis Hunt. 2026. Holding duality in mental health research and practice: towards the meaningful integration of dual-positioned researchers.Critical Public Health36, 1 (2026), 2619251. doi:10.1080/09581596.2026.2619251

    work page doi:10.1080/09581596.2026.2619251 2026

  16. [16]

    Vassilis-Javed Khan, Panos Markopoulos, and Nynke Spijksma. 2011. On the use of pervasive computing to support patients with obsessive compulsive disorder. InCHI ’11 Extended Abstracts on Human Factors in Computing Systems (Vancouver, BC, Canada)(CHI EA ’11). Association for Computing Machinery, New York, NY, USA, 1213–1218. doi:10.1145/1979742.1979750

    work page doi:10.1145/1979742.1979750 2011

  17. [17]

    Pohl, and Carolyn I

    Jiyeong Kim, Juan Pablo Gonzalez Pacheco, Ashleigh Golden, Elias Aboujaoude, Peter van Roessel, Aayushi Gandhi, Pavithra Mukunda, Tatevik Avanesyan, Haopeng Xue, Ehsan Adeli, Jane Paik Kim, Manish Saggar, Shannon Wiltsey Stirman, Eric Kuhn, Kaustubh Supekar, Kilian M. Pohl, and Carolyn I. Rodriguez. 2025. Artificial Intelligence in Obsessive-Compulsive Di...

    work page doi:10.1007/s40501-025-00359-8 2025

  18. [18]

    Maria Liaskou and Charalampos Rizopoulos. 2023. Using Virtual Reality to Raise Public Awareness of Obsessive- Compulsive Disorder: A Preliminary Study. InProceedings of the 2nd International Conference of the ACM Greek SIGCHI Chapter(Athens, Greece)(CHIGREECE ’23). Association for Computing Machinery, New York, NY, USA, Article 14, 5 pages. doi:10.1145/36...

    work page doi:10.1145/3609987.3610001 2023

  19. [19]

    Yuan Ma and Maorong Hu. 2025. Leveraging Educational Informatization for Mental Health Service Innovation: An ACT-Based Digital Intervention Model in Adolescents with OCD. InProceedings of the 2025 International Conference on AI-Enabled Education (AIEE ’25). Association for Computing Machinery, New York, NY, USA, 66–69. doi:10.1145/ 3768421.3768433

    work page arXiv 2025

  20. [20]

    Patrick McGrath PhD. 2025. A Quick Guide to Some Common OCD Subtypes. https://www.treatmyocd.com/blog/a- quick-guide-to-some-common-ocd-subtypes

    work page 2025

  21. [21]

    Joseph Reagle. 2022. Disguising Reddit sources and the efficacy of ethical research.Ethics and Information Technology 24, 3 (Sept. 2022), 41. doi:10.1007/s10676-022-09663-w

    work page doi:10.1007/s10676-022-09663-w 2022

  22. [22]

    Sigal Samuel. 2025. ChatGPT and OCD are a dangerous combo. https://www.vox.com/future-perfect/417644/ai- chatgpt-ocd-obsessive-compulsive-disorder-chatbots

    work page 2025

  23. [23]

    Noël Schepers, Cynthia Acca, Patricia Derard, and Philippe Fontaine. 2016. Using smartphones apps in psychotherapy: experiences with PTSD, OCD and panic disorder patients. InProceedings of the 2016 Virtual Reality International Conference(Laval, France)(VRIC ’16). Association for Computing Machinery, New York, NY, USA, Article 29, 3 pages. doi:10.1145/292...

    work page doi:10.1145/2927929.2927957 2016

  24. [24]

    Seibell and Eric Hollander

    Phillip J. Seibell and Eric Hollander. 2014. Management of Obsessive-Compulsive Disorder.F1000Prime Reports6 (Aug. 2014), 68. doi:10.12703/P6-68 Reassurance Robots: OCD in the Age of Generative AI 9

    work page doi:10.12703/p6-68 2014

  25. [25]

    Inhwa Song, Sachin R Pendse, Neha Kumar, and Munmun De Choudhury. 2025. The Typing Cure: Experiences with Large Language Model Chatbots for Mental Health Support.Proc. ACM Hum.-Comput. Interact.9, 7, Article CSCW249 (Oct. 2025), 29 pages. doi:10.1145/3757430

    work page doi:10.1145/3757430 2025

  26. [26]

    Brittany Stahnke. 2021. A systematic review of misdiagnosis in those with obsessive-compulsive disorder.Journal of Affective Disorders Reports6 (2021), 100231. doi:10.1016/j.jadr.2021.100231

    work page doi:10.1016/j.jadr.2021.100231 2021

  27. [27]

    Nora Yanyi Sun, Christopher Pittenger, and Terence Ching. 2025. Analyzing the Contents of a Large, Public Online Peer Support Forum for Obsessive-Compulsive Disorder: Thematic Analysis.JMIR Formative Research9, 1 (May 2025), e60899. doi:10.2196/60899

    work page doi:10.2196/60899 2025

  28. [28]

    Hongmiao Wang. 2020. Research on the Causes and Treatment of OCD. InProceedings of the 2020 Conference on Artificial Intelligence and Healthcare(Taiyuan, China)(CAIH2020). Association for Computing Machinery, New York, NY, USA, 188–193. doi:10.1145/3433996.3434030

    work page doi:10.1145/3433996.3434030 2020

  29. [29]

    It was Mentally Painful to Try and Stop

    Ru Wang, Kexin Zhang, Yuqing Wang, PhD Brown, Keri, and Yuhang Zhao. 2025. “It was Mentally Painful to Try and Stop”: Design Opportunities for Just-in-Time Interventions for People with Obsessive-Compulsive Disorder in the Real World. InProceedings of the 27th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS ’25). Association ...

    work page doi:10.1145/3663547.3746394 2025

  30. [30]

    Williams

    Rua M. Williams. 2025.Disabling Intelligences: Legacies of Eugenics and How We are Wrong about AI. Springer Nature Switzerland, Cham. doi:10.1007/978-3-032-02665-1

    work page doi:10.1007/978-3-032-02665-1 2025

  31. [31]

    Sina Ziegler, Klara Bednasch, Sabrina Baldofski, and Christine Rummel-Kluge. 2021. Long durations from symptom onset to diagnosis and from diagnosis to treatment in obsessive-compulsive disorder: A retrospective self-report study. PLoS ONE16, 12 (Dec. 2021), e0261169. doi:10.1371/journal.pone.0261169

    work page doi:10.1371/journal.pone.0261169 2021