A fundamental question in discussing artificial intelligence is what prompts are—instructions users provide to specify what they wish to obtain from the algorithm. The issue extends beyond teaching prompt engineering to teaching people to recognize when AI systems cannot provide an adequate response, why this happens, and how to identify such situations. The key issue is developing critical and lateral thinking, which is needed to understand when results are reliable and when they are not. Not all types of problems and cognitive analyses are handled with the same degree of reliability. The limits of reliability are constantly evolving, but available evidence highlights categories where AI systems prove particularly reliable, including text summarization of classical texts, analysis of regulatory or case-law corpora, review of key contributions from schools of thought, and analysis of large quantities of accounting documents or statistical data. Asking AI systems to summarize long, complex texts, identify key points, answer questions about content and relationships, or present information with varying levels of summarization is one of the most established and reliable capabilities. Certain types of analysis can now be easily delegated, often with surprising results.