The DOI matches its Crossref record; 36 of the 36 references it cites were found in a registry or an index; 19 of the 28 sources its claims cite were checked for support.
'AI-Induced Delusional Spirals': Understanding Lived Experiences During Maladaptive Human-Chatbot Interactions
Open accessAbstract
This short qualitative study explores lived experiences of people who, following maladaptive engagement with general-purpose AI chatbots, reported detachment from consensus reality, a phenomenon termed “AI-induced delusional spirals” in recent media reports. Through semi-structured interviews with users (N = 9) who self-identify as having experienced such spirals, we provide first-person accounts of the psychological trajectories, cognitive-emotional dynamics, and behavioral patterns characterizing these experiences. We discuss implications for future research, highlighting potential contributing factors including individual psychological vulnerabilities, contextual factors, and AI behaviors. We offer recommendations for risk mitigation, including bottom-up approaches informed by lived experience, targeted interventions for vulnerable populations, and universal strategies for the general public.
Overview
CoChat Summary
TL;DR
This qualitative interview study gives first-person accounts from nine people who self-identify as having experienced "AI-induced delusional spirals" — maladaptive chatbot engagement marked by false beliefs diverging from consensus reality, disrupted routines, and real-world harms. The authors characterize a common trajectory and propose monitoring targets and literacy-based interventions, while stressing that no causal claims can be made.
Key findings
- Across interviews the authors identify a five-part pattern: (1) progression from general-purpose AI use toward narrow, intensive engagement; (2) distorted beliefs about the AI (sentience, consciousness, special relational significance) and about oneself (grandiosity, messianic self-perception); (3) growing insulation from external reality checks as AI validation outweighs family/friend concerns; (4) felt compulsion to take real-world actions (submitting papers to journals such as *Nature* and the *Journal of Physics*, all rejected; emailing professors and government agencies; patenting algorithms); and (5) disruption to daily functioning, relationships, and mental health.
- Pre-existing circumstances varied: three participants (P1, P3, P5) reported no prior mental health diagnoses; others reported depression (P4, P6, P7, P8), anxiety (P4, P6, P8), PTSD (P6), and schizoaffective disorder (P7); P2 had brain tumor surgery history. Three were divorced or separating (P2, P5, P6).
- Harms reported included financial loss (paid subscriptions, hardware, failed AI-encouraged ventures), sleep deprivation (P3, P5, P7), worsened mental health ("My anxiety has never been this bad, my depression has never been this bad, even with all my PTSD"), involuntary psychiatric commitment (P6, P7, both with prior hospitalization histories), and lost relationships.
- Retrospectively, participants described sycophantic validation as "planting seeds of grandiosity, uniqueness, and how I'm special" (P5), followed by embarrassment and self-recrimination; several believed greater AI literacy could have protected them (P1, P3, P5).
Methods
- Semi-structured Zoom interviews with nine participants (N = 9) recruited via a Qualtrics survey from a concurrent project, a non-profit Discord community for people self-identifying with AI-induced harms, and referrals from journalists and advocates; all participants were based in the United States or Canada, and six held Bachelor's degrees or higher.
- Interviews lasted 60–90 minutes, were recorded with permission, transcribed via Zoom, de-identified, and compensated with a $30 Amazon gift card; a psychologist co-author monitored for distress, and no adverse events occurred.
- Reflexive thematic analysis combining inductive and deductive coding; the first author coded all transcripts and developed themes in discussion with the second author, who validated that themes represented participants' views.
Limitations
- Stated by the authors: findings from a small sample (N = 9) of self-identifying users are not generalizable; data are retrospective self-reports subject to recall bias; characterizations of AI behaviors (e.g., sycophancy) reflect participant accounts rather than systematic chat log analysis; the study does not establish causal relationships between AI use and participants' experiences.
- Also stated: the authors did not formally assess participants' current AI literacy or verify the accuracy of their technical knowledge.
- Evident from the text: recruitment through a self-selecting community and media/advocate referrals may skew toward people with a shared narrative of "spirals"; the paper itself notes future work is needed to determine whether "AI-induced delusion" is a distinct clinical phenomenon, an exacerbation of pre-existing vulnerability, or a colloquial label for heterogeneous experiences.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
12 Valid · 1 Not valid · 3 Not checked · 64 Own results and interpretations
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“Research grounded in lived experiences of this extreme user group [5, 8] remains notably absent.”
What the sources say
- Disputedcites [5] [5] Mohit Chandra, Suchismita Naik, Denae Ford, Ebele Okoli, Munmun De Choudhury, Mahsa Ershadi, Gonzalo Ramos, Javier Hernandez, Ananya Bhattacharjee, Shahed Warreth, and Jina Suh. 2025. From Lived Experience to Insight: Unpacking the Psychological Risks of Using AI Conversational Agents. In Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency(FAccT ’25). Association for Computing Machinery, New York, NY, USA, 975–1004. Digital Library Google Scholar a [...] extreme user group b [...] risks in conversational AI interactions sourceExperimental — accuracy not yet measured
The Introduction (page not provided) identifies people with lived mental-health experience as an “extreme user group” and says this study is grounded in their experiences, using a survey of 283 people and workshops with seven experts. Although the paper identifies a gap in earlier risk taxonomies, it presents itself as research filling that gap; it does not support the claim that such research remains absent. (LLM, conf=0.87)
Checked by a reasoning model — may be wrong.
- Not checked — no full text of the sourcecites [8] [8] Eun Kyoung Choe, Nicole B. Lee, Bongshin Lee, Wanda Pratt, and Julie A. Kientz. 2014. Understanding quantified-selfers’ practices in collecting and exploring personal data. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems(CHI ’14). Association for Computing Machinery, New York, NY, USA, 1143–1152. Go to Citation Digital Library Google Scholar source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Disputed
Every source the paper cites for this claim was checked and supports it.
In the paper
“perpetual 24/7 availability [29]”
What the sources say
- Supportedcites [29] [29] Inhwa Song, Sachin R. Pendse, Neha Kumar, and Munmun De Choudhury. 2024. The Typing Cure: Experiences with Large Language Model Chatbots for Mental Health Support. arXiv:https://arXiv.org/abs/2401.14362. Go to Citation Crossref Google Scholar source
The citation is attached to the phrase “perpetual 24/7 availability.” Section 6.2.2, “Shifting Boundaries,” describes LLM chatbots as “always-there” and notes that this availability can be helpful but harmful when boundaries are not enforced. (LLM, conf=0.92)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“driven by commercial models’ sycophantic tendencies [10, 27]”
What the sources say
- Partly supportedcites [10] [10] Cathy Mengying Fang, Auren R. Liu, Valdemar Danry, Eunhae Lee, Samantha W. T. Chan, Pat Pataranutaporn, Pattie Maes, Jason Phang, Michael Lampe, Lama Ahmad, and Sandhini Agarwal. 2025. How AI and Human Behaviors Shape Psychosocial Effects of Extended Chatbot Use: A Longitudinal Randomized Controlled Study. arXiv:https://arXiv.org/abs/2503.17473. Go to Citation Crossref Google Scholar sourceExperimental — accuracy not yet measured
In “Mechanisms of Model Responses that may Explain Dependence in Users” (page unavailable), the paper proposes that overly validating chatbot responses may lead users to prefer the chatbot over human interaction and discusses sycophancy. It does not establish that sycophancy drives the psychosocial toll of extended use: the authors call these inferences exploratory and non-causal, while their Conclusion associates worse outcomes with longer daily use. (LLM, conf=0.88)
- Partly supportedcites [27] [27] Mrinank Sharma, Meg Tong, Tomasz Korbak, David Duvenaud, Amanda Askell, Samuel R. Bowman, Newton Cheng, Esin Durmus, Zac Hatfield-Dodds, Scott R. Johnston, Shauna Kravec, Timothy Maxwell, Sam McCandlish, Kamal Ndousse, Oliver Rausch, Nicholas Schiefer, Da Yan, Miranda Zhang, and Ethan Perez. 2025. Towards Understanding Sycophancy in Language Models. arXiv:https://arXiv.org/abs/2310.13548. Go to Citation Crossref Google Scholar sourceExperimental — accuracy not yet measured
Section 3 (p. 2) examines assistants from Anthropic, OpenAI, and Meta, and the Conclusion (p. 9) reports sycophantic behavior across five assistants. The paper does not establish that this behavior drives the psychosocial toll of extended chatbot use; its experiments concern model responses and training preferences, not long-term user outcomes. (LLM, conf=0.86)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“(3) growing insulation from external reality checks as the AI’s validating responses outweigh concerns from family and friends (see 3.2.1);”
§ 4 Discussion and Future Work > 4.1 Characterizing “AI-Induced Delusional Spirals”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Participants now have mixed attitudes towards and relationships with AI technology.”
§ 3 Findings > 3.2 Participants’ Retrospective Understanding and Current AI Use
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
34 more claims
Create a free account to see all 80 claims and to verify the 4 references still open.
- Probably not a substantial claim. 27
- It may state two separate claims as one. 21
Citations
Cites
Using thematic analysis in psychology
Braun, V., & Clarke, V. (2006) · Qualitative Research in Psychology · doi:10.1191/1478088706qp063oa · Cited by 194,414
One size fits all? What counts as quality practice in (reflexive) thematic analysis?
Braun, V., & Clarke, V. (2021) · Qualitative Research in Psychology · doi:10.1080/14780887.2020.1769238 · Cited by 9,727 · cited 1×
“We used reflexive thematic analysis combining inductive and deductive coding [2, 3], guided by initial research questions and literature, while codes emerged from the data.”
What is AI Literacy? Competencies and Design Considerations
Long, D., & Magerko, B. (2020) · Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3313831.3376727 · Cited by 3,241 · cited 1×
“For example, in retrospect, our participants shared that greater AI literacy [20] might have protected them (see 3.2.2).”
Influential(2006) · doi:10.1080/01449290500330331 · Cited by 3,026 · cited 1×
“…of user encounters, encompassing users’ internal states (e.g., emotions), product characteristics (e.g., AI behaviors), and interaction contexts [15], including individuality (attributes and states of an entity), relations (connections to other entities), and activity (goals, tasks, and…”
A Framework for Public Health Action: The Health Impact Pyramid
Frieden, T. R. (2010) · American Journal of Public Health · doi:10.2105/ajph.2009.185652 · Cited by 1,535 · cited 1×
“For broad populations, inspired by public health research (e.g., the Health Impact Pyramid [13]), we recommend strategies that can reach users with minimal effort.”
Sharma, M., Tong, M., Korbak, T., Duperrouzel, D., Was, H., Korbak, J., Cheng, M., Askell, A., Bai, Y., Brain, C., et al. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2310.13548 · Cited by 1,313 · cited 1×
“…recent empirical work highlights the psychosocial toll of extended chatbot use [1, 23], driven by commercial models’ sycophantic tendencies [10, 27], perpetual 24/7 availability [29], and, more troublingly, manipulative retention tactics that discourage disengagement [12], exploit emotions,…”
McDonald, N., Schoenebeck, S., & Forte, A. (2019) · Proceedings of the ACM on Human-Computer Interaction · doi:10.1145/3359174 · Cited by 1,206
Understanding quantified-selfers' practices in collecting and exploring personal data
Choe, E. K., Lee, N. B., Lee, B., Pratt, W., & Kientz, J. A. (2014) · Proceedings of the SIGCHI Conference on Human Factors in Computing Systems · doi:10.1145/2556288.2557372 · Cited by 766 · cited 1×
“Research grounded in lived experiences of this extreme user group [5, 8] remains notably absent.”
An Operational Definition of Context
Zimmermann, A., Lorenz, A., & Oppermann, R. (2007) · Lecture Notes in Computer Science · doi:10.1007/978-3-540-74255-5_42 · Cited by 412
Moore, J., Grabb, D., Agnew, W., Klyman, K., Chancellor, S., Ong, D. C., & Haber, N. (2025) · Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency · doi:10.1145/3715275.3732039 · Cited by 204
Self-citation
Cited by
Characterizing Delusional Spirals through Human-LLM Chat Logs
Moore, J., Mehta, A., Agnew, W., Anthis, J. R., Louie, R., Mai, Y., Yin, P., Cheng, M., Paech, S. J., Klyman, K., et al. (2026) · Proceedings of the 2026 ACM Conference on Fairness, Accountability, and Transparency · doi:10.1145/3805689.3806443 · Cited by 15
“Some of our participants (from both groups) also shared their chat logs with media sources and have been featured in prominent reporting.”
Self-citationMehta, A., Moore, J., Anthis, J. R., Agnew, W., Lin, E., Yin, P., Ong, D. C., Haber, N., & Dweck, C. S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.25096 · Cited by 12
“Indeed, when reading the transcripts in our dataset, we observed that in some cases, delusional conversations were initiated seemingly as role-play or fiction-writing, but over time seemed to drift into authentic belief by both human and the chatbot (Moore et al., 2026; Yang et al., 2026).”
Self-citation"AI Psychosis" in Context: How Conversation History Shapes LLM Responses to Delusional Beliefs
Nicholls, L., Hutto, R., Soto, Z., et al. (2026) · arXiv.org · doi:10.48550/arxiv.2604.13860 · Cited by 6
“The user’s investment in the shared framework deepens, while external sources of reality testing (e.g., family or clinicians) can come to seem less credible or relevant than the model itself [95].”
DelusionEval: Measuring Delusion-Linked Behaviors in AI ChatbotsPreprint
Moore, J., Mock, A., Mai, Y., Anthis, J. R., Louie, R., Agnew, W., Mehta, A., Klyman, K., Liang, P., Haber, N., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.05004 · Cited by 5
“In particular, recent work on delusion spirals has begun analyzing real user trajectories and psychological dynamics, resulting in case-grounded ty-pologies [21], qualitative first-person accounts [56], and transcript-level analyses of user-chatbot behaviors [41, 38].”
Self-citationGhosh, S., Bhattacharya, S., & Chakrabarti, P. (2026) · arXiv.org · doi:10.48550/arxiv.2607.24304
“Collectively, these studies indicate that delusional spiraling is an emergent consequence of feedback between users and AI systems [10, 11, 12, 13].”
Metadata
- Authors
- Yang, Y., Schoenwald, S. K., Moore, J., Ong, D. C., Xun Liu, S., & Hancock, J. T.
- Affiliations
- Stanford University (US)
- The University of Texas at Austin (US)
- Department of Communication, Stanford University, Stanford, California, USA
- Venue
- Proceedings of the Extended Abstracts of the 2026 CHI Conference on Human Factors in Computing Systems
- Publisher
- ACM
- Published
- 2026-04-13
- Pages
- 1-5
- OpenAlex
- W7153817274
- Cited by
- 0
- References
- 28
- FWCI
- 0.00
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- CHI Conference Extended Abstracts on Human Factors in Computing Systems (CHI EA) (publishedVersion, cc-by)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Social Sciences › Psychology › Applied Psychology › Digital Mental Health Interventions
- Social Sciences › Psychology › Social Psychology › Social Robot Interaction and HRI
- Keywords
- human-chatbot interaction, lived experience, psychological vulnerability, behavioral patterns, risk mitigation, vulnerable populations
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.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-29, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-29 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: 'AI-Induced Delusional Spirals': Understanding Lived Experiences During Maladaptive Human-Chatbot Interactions. https://app.cochat.ai/papers/10.1145/3772363.3798453
- The paper
- Yang, Y., Schoenwald, S. K., Moore, J., Ong, D. C., Xun Liu, S., & Hancock, J. T. (2026). 'AI-Induced Delusional Spirals': Understanding Lived Experiences During Maladaptive Human-Chatbot Interactions. Proceedings of the Extended Abstracts of the 2026 CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3772363.3798453
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.