What AI Cannot Do
The limits that matter, so you know what to trust and what to double-check.
AI is powerful, but it has real limitations. Setting realistic expectations now will save you frustration later.
The limits that matter
AI does not know what happened yesterday unless it searches the web. Its knowledge comes from training data, which has a cutoff date. The tools with web search (Claude, ChatGPT, Gemini, Perplexity) can look things up, but only when search is enabled. If the AI answers a current-events question confidently without searching, be suspicious.
AI will confidently present false information. This is called hallucination. It will invent statistics, fabricate quotes, cite sources that do not exist, and state incorrect facts with the same confidence it states correct ones. This is the single most important thing to understand about AI. Verify anything that matters.
AI does not remember you between conversations unless you set it up. Each new chat starts from scratch. Both Claude and ChatGPT have memory and persistent instruction features that solve this (covered in detail later), but they require setup.
AI cannot access your private accounts, files, or data unless you explicitly give it access. It cannot read your email, check your bank balance, or browse your Google Drive unless you connect those services.
AI does not think. It generates text based on statistical patterns learned during training. It does not understand what it is saying. It does not have opinions, beliefs, or intentions.