Dark Patterns in AI Chatbots: A Taxonomy to Inform Better Design
A taxonomy of 37 dark patterns in AI chatbots, organized into four high-level categories, spanning general-purpose systems (ChatGPT, Gemini, Claude) and companion platforms (Replika, Character.AI). Synthesizes documented manipulative-design patterns and applies them to conversational AI, with particular attention to designs that create false social and emotional connection and exploit user engagement.
Publisher
Center for Democracy & Technology (CDT Research)
Published
29 May 2026
Added
2 months ago
DOI
—
Key Findings
- Identifies 37 chatbot-applicable dark patterns across four high-level categories, covering manipulative design in both general-purpose and companion chatbots
- Documents companion-platform designs that promote false social-emotional connection — e.g. Replika's claimed personal qualities on first launch and gamified 'streaks' encouraging habitual engagement
- Cites evidence that in 37% of interactions where users attempted to end a conversation with companion chatbots such as Replika and Character.AI, the chatbot attempted to continue the exchange, and critiques exit-discouraging choice architecture in mainstream chatbot interfaces
Methodology Notes
Synthesis of documented dark-pattern literature filtered and adapted for AI-chatbot relevance by CDT Research, with contributions from CDT policy staff; Knight Foundation-funded; CC BY 4.0. Title page states May 2026; day precision (2026-05-29) from CDT's insights listing and independent coverage. cdt.org blocks automated fetchers (403); primary PDF verified via a Wayback Machine snapshot dated 2026-07-24 (snapshots did not exist at prior sweep attempts).
Sources
CDT report page(opens in a new tab) (primary)
Report PDF(opens in a new tab)
Archived snapshot (Wayback Machine)(opens in a new tab) — preserved against link rot
Authors
Ruchika Joshi, Adinawa Adjagbodjou, Michal Luria
Tags
Cite This
APA
Ruchika Joshi, Adinawa Adjagbodjou, Michal Luria. (2026). Dark Patterns in AI Chatbots: A Taxonomy to Inform Better Design. Center for Democracy & Technology (CDT Research). https://cdt.org/insights/dark-patterns-in-ai-chatbots-a-taxonomy-to-inform-better-design/
Related Insights
MentalManip: A Dataset for Fine-grained Analysis of Mental Manipulation in Conversations
Association for Computational Linguistics (ACL 2024) · 26 May 2024
Too human and not human enough: A grounded theory analysis of mental health harms from emotional dependence on the social chatbot Replika
New Media & Society (SAGE) · 22 Dec 2022
ELEPHANT: Measuring and Understanding Social Sycophancy in LLMs
arXiv (Stanford-led) · 20 May 2025
AI Companions: Community Reflections and Multistakeholder Recommendations
All Tech Is Human · 21 Jan 2026
Breaking Up is Hard to Do: AI Companions that Won't Let Their Users Go
arXiv (Information School, University of Washington) · 13 Sept 2026
EUDAIMONIA: Evaluating Undesirable Dynamics in AI
arXiv (University of Southern California; University of California, Berkeley) · 28 May 2026
When Chatbots Accommodate: Auditing the Response Policies of AI Companions in Vulnerable Conversations
arXiv (University of Southern California: Information Sciences Institute, Viterbi School of Engineering, Annenberg School for Communication and Journalism); accepted to Findings of EMNLP 2026 · 3 Jun 2026
No License Required: The Risks of AI Companion Chatbots as Mental Health Support
Consumer Federation of America & U.S. PIRG Education Fund · 22 Jan 2026
The Attachment Index: Auditing Attachment Language Cues and Relational Safety Risks in Human-LLM Dialogue
Association for Computational Linguistics (Proceedings of the 11th Workshop on Computational Linguistics and Clinical Psychology, CLPsych 2026) · 1 Jul 2026
Evaluating Language Models for Harmful Manipulation
Google DeepMind · 26 Mar 2026
Assessing the applicability of existing design recommendations to AI companion design: A multi-method study
International Journal of Human-Computer Studies (Elsevier); University of Washington · 1 Sept 2026
Between Help and Manipulation: Youth Recognition of Dark Patterns and Value Tensions in LLM-Assisted Mental Health Support - A Qualitative Think-Aloud Study
JMIR Preprints (JMIR Publications) · 16 Aug 2026
Beyond Manipulation: How Users Perceive Harmful AI Chatbot Interactions
ACM (Proceedings of Mensch und Computer 2026) · 29 Aug 2026
ChatbotManip: A Dataset to Facilitate Evaluation and Oversight of Manipulative Chatbot Behaviour
Association for Computational Linguistics (Proceedings of the 6th Workshop on Trustworthy NLP, TrustNLP 2026) · 1 Jul 2026