Microsoft Copilot AI Prompts Guidebook (AI Bookstore)（Z-Library）

# Microsoft Copilot AI Prompts Guidebook ## 【One-Line Pitch】 A practical flight manual for Microsoft Copilot, teaching you to move from passive search-bar user to active commander of AI by mastering prompt frameworks, interface modes, and responsible usage. Essential reading for students, professionals, and creators who want to leverage AI as a true thinking partner rather than a fancy search engine. ## 【Book Arc】 - **Opening (~0%–9%)**: Establishes the paradigm shift from command-based computing to conversational AI, introducing Copilot as a "reasoning engine" that unifies GPT-4's language understanding, Bing's live web access, and DALL-E's image generation into one interface. The "Copilot vs. Autopilot" philosophy is introduced—you remain the pilot in command. - **Early (~9%–19%)**: Sets expectations for what the book will teach: mastering Creative/Balanced/Precise modes, the proprietary C.R.A.F.T. prompt framework, multimodal fluency (text + images + data), ecosystem integration with Edge/Windows/Microsoft 365, and ethical guardrails around hallucinations and privacy. - **Middle (~19%–28%)**: Walks through the "cockpit" setup—the different access points (web, Edge sidebar, Windows integration, mobile app, Microsoft 365) and their distinct missions. Covers the anatomy of the interface: prompt box, microphone, image upload, and the critical "New Topic" reset button for managing context windows. - **Late (~28%–38%)**: Dives into the three conversation modes (Creative, Balanced, Precise) with clear guidance on when to use each—never code in Creative, never poetry in Precise. Explains authentication tiers (anonymous, personal, work/school) and the crucial "green shield" Commercial Data Protection for business users. - **Ending (~38%–44%)**: Shifts to the cognitive foundation of prompt engineering—understanding that Copilot is a probabilistic machine predicting word sequences, not a knowledge base. Introduces the C.R.A.F.T. framework as a repeatable structure for pruning the "probability tree" and forcing high-value, specific responses. ## 【Key Takeaways】 - **The "Copilot" philosophy is your ethical compass** (Early): Unlike an autopilot that takes over, a copilot assists while you remain in command—you set the destination, make judgment calls, and own quality control. This framing prevents the common mistake of blindly trusting AI output. - **Copilot is three technologies fused into one** (Early): GPT-4 provides reasoning, Bing provides live grounding in current reality, and DALL-E provides visual creation. Understanding this trinity explains both its power (real-time, multimodal answers) and its limits (it's a statistical engine, not a knowledge base). - **The "New Topic" button is your most critical control** (Late): LLMs have a finite context window; as conversations drag on, the AI conflates old instructions with new requests. Always reset when switching tasks—this flushes short-term memory and prevents "noise" from contaminating fresh work. - **Mode selection determines output quality** (Late): Precise mode grounds answers in search results and is least likely to hallucinate—use it for coding, math, science, and factual research. Creative mode is for ideation and writing. Never mix them: no code in Creative, no poetry in Precise. - **Authentication tier = data safety tier** (Late): Anonymous use limits turns and features; personal accounts unlock full tools but may use de-identified chat data for training; work/school accounts with Entra ID activate Commercial Data Protection (the green shield)—no eyes-on access, no training on your data, no leakage into search results. - **Detailed prompts prune the probability tree** (Ending): Vague prompts like "Write a contract" force the AI to guess among generic templates. Specific prompts ("Act as a California real estate attorney... favoring the landlord... prohibiting subletting") aggressively cut off wrong branches, reducing guessing to near zero. - **Ask three questions before every prompt** (Late): What is my goal (list, paragraph, table, code)? Who is the persona (creative writer, harsh critic, tutor)? What is the context (does the AI know, or must I paste background)? These three calibrations determine output quality more than any technical setting. ## 【Reading Tips】 - **Deep-read the opening philosophy section** (~0%–9%): The "Copilot vs. Autopilot" distinction and the three-pillar architecture (LLM + Search + DALL-E) frame everything that follows. Skim if you're already familiar with AI basics, but don't skip the philosophy—it shapes responsible usage. - **Treat Chapter 1 as a reference, not a narrative** (~19%–38%): The interface anatomy, mode selection, and authentication tiers are practical checklists. Bookmark the green shield explanation and the mode-selection guidance—you'll return to them constantly. - **Pay special attention to the "New Topic" golden rule** (~28%): This single habit prevents most user frustration with AI confusion. It's a small detail with outsized impact on output quality. - **The C.R.A.F.T. framework section is the book's core value** (~44%): The probabilistic machine explanation and the framework for pruning probability space are the most transferable skills. Read this slowly and practice with real prompts. - **Skim the interface anatomy if you're already a Copilot user** (~28%–34%): The microphone, image icon, and feedback loop descriptions are useful but self-evident once you've used the tool. Focus instead on the mode selection and authentication sections. ## 【Coverage Limits】 The excerpts cover the book's opening philosophy, interface setup, mode selection, authentication, and the beginning of the prompt framework (C.R.A.F.T.). They do not cover the full C.R.A.F.T. method details, multimodal workflows (image uploads, PDF analysis), Microsoft 365 integration specifics, or the ethical guardrails chapter—these are promised but not detailed in the sampled material. ##