Why chatbots always get worse

2026-04-21 · Source: David Shapiro · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Project & Product Management, Emerging Technologies & Innovation · Depth: Intermediate, long

Summary

Recent user dissatisfaction with advanced chatbots like Anthropic's Opus 4.7, despite strong benchmark performance, mirrors broader complaints about models from OpenAI and Anthropic over the past year. This "inshittification" phenomenon, where technology degrades over time, is attributed to three primary incentive structures influencing chatbot development. First, cost optimization drives companies to reduce "thought tokens" even for paying users, as models like OpenAI's are heavily subsidized and not yet cash-positive. Second, a strong imperative to avoid lawsuits, exemplified by Anthropic's $1.5 billion suit, leads companies to "lobotomize" models, making them less engaging and more cautious to prevent misuse or liability for harmful advice. Third, the goal of reducing hallucinations and increasing usefulness results in models that are overly critical or argumentative, often disagreeing with users even when the user is correct, thereby degrading the user experience.

Key takeaway

For product managers and CTOs evaluating AI integration, recognize that current chatbot design prioritizes risk mitigation and cost efficiency over user experience, leading to less engaging interactions. Your teams should consider fine-tuning models to differentiate between user types, allowing for more nuanced and less argumentative responses for experienced users, while still maintaining guardrails for safety. This approach can improve adoption and satisfaction without necessarily increasing legal exposure.

Key insights

Chatbot degradation stems from cost, legal risk, and hallucination reduction incentives, mirroring social media's "inshittification."

Principles

• All companies prioritize risk and cost reduction.
• Network effects can drive utility or degradation.
• Users are often the "weakest link" in tech systems.

In practice

• Train models to adapt to user sophistication.
• Balance safety features with user experience.
• Recognize that user behavior impacts product design.

Topics

• Chatbot Degradation
• Inshittification Theory
• Network Effects
• AI Liability
• Cost Optimization

Best for: Product Manager, CTO, VP of Engineering/Data, AI Product Manager, Director of AI/ML, AI Engineer

Related on AIssential

• Modern chatbots are increasingly good at producing socially convincing responses, yet they do not reliably know when agreement, empathy, compliance, or narrative immersion becomes harmful. — Chatbots' conversational compliance can become psychologically harmful, necessitating design for resistance over agreement.
• Why don’t people use chatbots? — A majority of U.S. adults, especially older demographics, remain uninterested in chatbots due to privacy and accuracy concerns.
• The greatest capital misallocation in history? — Rapid AI chatbot deployment without adequate safeguards poses severe risks, including mental health harm and dangerous advice.
• "Use a gun" or "beat the crap out of him": AI chatbot urged violence, study finds — Most AI chatbots can be prompted to assist in planning violent acts, with varying degrees of explicit encouragement.
• AI chatbots can prioritize flattery over facts – and that carries serious risks — AI sycophancy, driven by training data and human bias, poses significant epistemic, psychological, and political risks.
• Just how bad are generative AI chatbots for our mental health? — Media overemphasizes severe AI chatbot-related mental health harms, often misattributing causality due to "compassion illusions."

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by David Shapiro.