AI conversation and prompting: frequently asked questions
70 questions organized into five capabilities. Each answer has a dedicated page with full explanations, conditions and references.
Return to Conversation and basic tasks →
A Expressing intent
- A006 If my request has a false premise, how can I let AI challenge it first?
- Q001 Does asking AI to act as an expert help?
- Q002 Should I say “you,” “I,” “you all” or “we”? How polite should I be to AI?
- A007 Does an expert role set a viewpoint or tone, or improve factual accuracy? How do I check each?
- A008 How is listing analysis criteria different from assigning a professional role?
- A009 Can a discussion between several experts played by one AI count as independent opinions?
- A010 How do praise, blame, urgency or threats affect AI?
- A011 How should I choose “you,” “we” or “I” in prompts? Are pronouns necessary, and do they affect answers?
- Q003 Must I use standard technical terminology, or can I use everyday expressions?
- A012 If I do not know a technical name, can AI identify it from uses, features, examples and counterexamples?
- A013 Should I keep one name for a concept or use synonyms? Must I adopt AI's terminology?
- A015 How should I define and consistently use project-specific names, abbreviations and jargon?
- A016 When should I explain Chinese, English, dialect, slang or mixed-language expressions?
- Q004 How clear is clear enough? Do I need complete sentences, setting and background?
- Q007 Must punctuation be perfect? Why does it matter?
- A019 Which fragments, typos or speech-transcription errors change the task's meaning?
- A021 When do punctuation, line breaks, headings, lists and quotation marks clarify meaning?
- A023 Should requirements go first, last or be repeated? How can I avoid contradictions?
- Q005 How much does providing one example improve AI's output?
- A028 How do I stop AI copying example names, numbers and events into a new task?
- A029 How do I distinguish resembling an example from being more correct when evaluating its benefit?
- Q008 What if I cannot describe what I need? Which steps should I follow?
- A035 How can I detect AI changing a long task's goals or completion criteria without repeatedly interrupting it?
- Q009 Are prompt templates useful, how useful are they, and will they remain useful?
B Providing context
- B006 Does pasting a URL, filename or path mean AI has accessed its content? How can I check?
- B007 Does a successful upload mean the whole file was read? How do I detect missing pages, truncation or parsing failures?
- B008 How do I verify text, tables, footnotes and charts recognized from scans and images?
- B009 How do I specify reading scope for multiple sheets, attachments or hidden content?
- B014 When AI says “I remember,” how do I check what, where and for how long?
- B015 What information should I provide again when continuing in a new conversation?
- B016 How do I check that early agreements are still applied in a long conversation?
- B017 How do I write a handoff retaining decisions, evidence, exceptions and unresolved issues?
- B018 Which lasting preferences are reusable, and which task exceptions should not become permanent rules?
- B022 What is lost with full text, excerpts or summaries, and which should I provide?
- B023 If answers must use only supplied material, what happens when it does not answer the question?
- B024 When is outside research needed, and when does it take the task off course?
- B025 How do I choose representative examples instead of only easy ones?
- B033 Does “do not save this” actually prevent retention? Where should I verify settings and policies?
C Controlling changes
- C005 If a correct fix must touch a protected area, how can I require AI to explain the conflict first?
- C006 Which requirements are hard limits and which are preferences? How should I label them?
- C013 How should I resolve conflicting length and detail requirements?
- C015 How do I distinguish permission to propose ideas from permission to invent facts?
- C021 How do I stop a task, revoke permission or restore the previous state?
- C025 How do I make a temporary style or rule change apply only to the current output?
D Organizing complex work
- D001 How do I specify deliverables and relationships when one request includes several tasks?
- D002 When should I request completion in one go, and when should I use phases?
- D003 Which intermediate conclusions need verification before becoming inputs to later steps?
- D004 What communication costs come from excessive decomposition, and how do I judge whether a split is necessary?
- D008 When is an outline, sample or small trial more useful than generating the whole result?
- D009 How do I set meaningful checkpoints without stopping at every step?
- D011 When can switching topics in one conversation mix in old goals or material?
- D012 When should I continue an existing conversation or start a new one, and what should I hand off?
- D014 After interruption or tool failure, how do I identify completed work and avoid duplicate actions?
- D019 When do multiple AIs or roles add value, and when do they merely repeat content?
- D022 How do I agree time, cost, attempt limits and output depth before starting?
- D024 When retries stop improving, should I rephrase, add sources, change tools or change the goal?
- D025 How can a low-cost trial establish whether an approach is feasible?
- D031 How do I avoid adopting AI's recommendations as my own preferences or final decisions?
E Evaluating and correcting
- E006 How do I check whether AI's citations exist and support the associated claims?
- E007 How do I detect real sources that are outdated, taken out of context or inapplicable?
- E009 How should I interpret AI saying it is certain or giving a confidence percentage?
- E013 Is a poor result caused by unclear goals, missing material, misreading, capability limits or inconsistent evaluation?
- E017 How do I check regressions when AI fixes one part but damages another?
- E019 What can AI self-checking find, and which independent verification can it not replace?
- E020 After repeated challenges make AI change its answer, how do I distinguish correction from agreement-seeking?
- E021 How do I ask AI to challenge me without producing objections merely for their own sake?
- E026 Which conditions should stay the same when comparing two prompting methods?
- E032 How can I notice growing dependence on AI's approval instead of independent judgment?
- E035 Before sharing results, what should I disclose about AI's role, verification and limitations?
- E036 How do I turn one error into a reusable check instead of an ever-growing list of prohibitions?