Writing Instructions AI Models Understand
Latent Library Editorial
© 2026 Latent Library. All rights reserved.
This book is licensed for the personal use of the buyer. Please do not share or resell the file. Short quotations with attribution are welcome.
The information in this book is provided for education. It is not legal, financial or other professional advice. Product and company names are used only to describe them and belong to their owners.
This is a free sample: the introduction and first chapter.
Ask two people to get the same job done with the same AI assistant, and you will often get two very different results. One receives a vague, generic paragraph. The other receives a precise, well-structured answer that can be used as it is. The model was the same. The difference was the prompt.
A prompt is everything you give a language model before it responds: your instructions, the background information, any examples, and the conversation so far. Writing prompts well is not a bag of magic phrases. It is a skill closely related to something you already know: briefing a capable colleague who knows a lot about the world but nothing about your situation.
This book teaches that skill systematically. It begins with how a model reads a prompt, because good technique follows from understanding the reader. It then sets out the building blocks of a strong prompt, the use of examples and formats, and methods for getting careful reasoning. Later chapters give tested patterns for common tasks, a method for fixing prompts that fail, and advice on building reusable templates for yourself or your team.
The techniques here are deliberately model-agnostic. Different assistants have different strengths, and new versions appear constantly, but the principles of clear communication change slowly. Where behaviour varies between models, the book says so and shows you how to test it.
The chapters fall into four parts, and each part builds on the one before.
Part one: the fundamentals (Chapters 1 to 4). How a model reads your prompt, the six building blocks of a strong prompt, the use of examples, formats and structure, and techniques for careful reasoning. If you read nothing else, read these four chapters; everything later applies them.
Part two: real work (Chapters 5 to 8). Tested patterns for writing, summarising, extracting, analysing, brainstorming, coding and learning; then the special demands of long documents, multi-turn conversations, and images, tables and files.
Part three: reliability (Chapters 9 to 11). A step-by-step method for fixing prompts that fail, a simple way to test prompts on a set of real examples, and the precautions that keep your prompts safe: protecting private data, inviting honest uncertainty and handling untrusted text.
Part four: scale and habit (Chapters 12 to 16). Reusable templates and team libraries, system prompts for custom assistants, a casebook of ten before-and-after rewrites, patterns for eight professions, and advice on building a personal prompting practice that lasts.
Two appendices close the book: a one-page checklist to keep beside you, and practice exercises with suggested answers.
You can read straight through, but you do not have to. If you already use AI tools daily, skim Chapter 1, read Chapters 2 to 4 carefully, then jump to whichever later chapter matches your work. If you are responsible for AI tools that other people use, give special attention to Chapters 10, 11 and 13.
Most of all, try things as you go. Many chapters contain prompts you can adapt, marked with Try it boxes. Use your own tasks rather than invented ones: you will learn faster when you can judge the output against what you know good work looks like. Keep a note of the prompts that work; by the end of the book you will have the start of a personal library.
A word on examples: names of companies, products and people in the examples are invented. The prompts are written in plain English and work with any capable assistant; where a feature exists only in some tools, such as file upload or code execution, the text says so.
A language model generates its reply one small piece of text (a token) at a time, each time choosing a likely continuation given everything that came before. Your prompt is the start of that sequence. Everything in it shapes what is likely to come next.
This has several practical consequences:
Most assistants distinguish between several kinds of input:
Because the whole conversation is visible to the model, a long chat can drift. Earlier misunderstandings or abandoned drafts keep influencing new answers. When a conversation goes off track, it is often faster to start a fresh one with a better first prompt than to keep correcting.
Models can only consider a limited amount of text at once, called the context window. Modern windows are large, often enough for long reports or many documents, but two cautions apply. First, when very long contexts are used, models may pay less reliable attention to details in the middle than to the beginning and end. Second, everything you include costs time and, with paid APIs, money. Include what is relevant, not everything you have.
You have reached the end of the free sample.
The full book continues with 15 more chapters, plus everything else in its 80 pages.