Context anxiety is the informal name for a very real feeling: the creeping worry that your AI assistant is about to forget everything you just told it. It is not a clinical term. It is what many practical users feel during a long working session with a large language model, once the conversation has grown complex and losing track would be costly.

That unease comes from something concrete: the context window. This is the model’s working memory for a session, including your prompt, uploaded files, chat history, tool outputs, and the model’s own replies. As the session grows, all of that material competes for limited room, and older details can become harder for the model to use reliably.

Context anxiety tends to show up during long, multi-step work. The limits are often invisible until you get close to them. Compaction or summarization may help the session continue, but it can also make users wonder whether a crucial constraint was preserved. That is why people sometimes hesitate to paste in more source material, even when the model needs it, because it feels like spending a budget they cannot see.

The fix is mostly discipline. Put important requirements in a working document outside the chat. Ask for checkpoint summaries at natural breakpoints. Split large efforts into phases with clear handoffs. Keep source files stored separately and bring them in only when needed. In other words, treat context as a project resource. That is the practical heart of Context Engineering: managing what the model sees so it can keep doing useful work.

If context anxiety has you rethinking how you talk to AI, the best place to start is with clearer prompts and cleaner structure. Coursera’s Prompt Engineering for ChatGPT course teaches the fundamentals of framing requests well, so each session uses its limited context more effectively.*