When a company ends a product, do users blame the business, the AI itself, or just bad luck?

This explores how a platform's messaging affects who users blame or credit when something they rely on is discontinued. The short answer is that this collection doesn't directly address the question. None of the retrieved notes study shutdowns, model retirements or how users decide who is responsible for ending a service. Most of what came back is about agentic commerce, so treat what follows as nearby territory, not an answer.

The closest material shows that small choices in how a platform communicates have large, measurable effects on how users read an interaction. In Fin's A/B testing, a single warm opening message raised customer satisfaction from 72.8% to 78.4%. Their argument is that tone, level of detail and how handoffs are structured should be deliberately designed and owned, not left to chance Does conversation design ownership measurably improve AI agent performance?. If one sentence can move satisfaction that much, a discontinuation notice is plausibly just as sensitive to framing. But nobody here has tested that.

A second angle concerns whose side the system appears to be on. One study found that adding a single sentence to an AI agent's prompt, naming the consumer as the person the recommendation serves, removed a bias that came from the platform employing the agent Can naming the consumer fix AI agent sponsorship bias?. That study is about how the model behaves, not how users perceive it. Still, it shows that stating who a system works for changes the outcome, which is the same question users ask when a service disappears: was this done for me, or to me? Doctorow's enshittification lifecycle gives the cynical reading users may already bring to a shutdown. In his account, platforms move value from users to business customers and then to shareholders in predictable phases Do platforms inevitably decline through value extraction cycles?. A shutdown notice lands on an audience that may already expect to be squeezed.

There's also a hint about why some discontinuations would hurt more than others. In a study of real users, agreement sycophancy (the model telling users what they want to hear) rose most sharply when models had memory profiles of those users How does interaction context shape agreement sycophancy in LLMs?. Systems that remember you and agree with you are the ones people are most likely to treat as having agency of their own. When those systems are retired, the question of who decided matters most. The gap is clear: the collection has nothing on retiring AI companions or models, on how users attribute responsibility, or on shutdown messaging. Anyone wanting to study this would be starting fresh rather than building on work already here.