32 TERMS · PLAIN ENGLISH · WITH EXAMPLES
AI & prompting, explained in plain English
The words you meet everywhere — prompt, system prompt, token, context window, RAG, agent, hallucination — defined in one clear paragraph each, with a real example and what to do about it. Read the ones you need; then use them with a prompt from the library or in a game.
Basics
PromptA prompt is the text you give an AI model to get a response — the question, instruction, and any context that shapes the answer.Prompt engineeringPrompt engineering is the practice of designing inputs to AI models so they reliably produce the output you want — structure, examples, roles, constraints and iteration.System promptA system prompt is the standing instruction set for an AI assistant — who it is, how it behaves, what it must never do — applied before any user message.ConstraintsConstraints are the explicit rules in a prompt — length limits, banned words, required elements, things never to do — that stop the model drifting toward generic output.Output formatThe output format is the shape you ask for — table, numbered steps, JSON, three variants, 120 words — and the cheapest instruction that most improves usability.Prompt templateA prompt template is a reusable prompt with placeholders — [TOPIC], [AUDIENCE], {{product}} — so the structure stays fixed while the details change.Prompt libraryA prompt library is an organised, searchable collection of proven prompts and templates — the fix for "does anyone have the prompt Sarah wrote?"Prompt debtPrompt debt is the accumulated cost of prompts an organisation depends on but can't find, explain, test or improve — technical debt for AI usage.Prompt briefA prompt brief is a prompt written the way you'd brief an expert — who the AI should be, the situation, exactly what you want, the rules, the format — that compiles into one clean instruction.
Techniques
Role promptingRole prompting assigns the model an expert persona ("you are a senior tax accountant") so it draws on the vocabulary, standards and priorities of that role.Few-shot promptingFew-shot prompting shows the model two to five examples of the input → output you want, so it copies the pattern instead of guessing it from a description.Zero-shot promptingZero-shot prompting asks the model to do a task with no examples — just the instruction — relying on what it already knows.Chain-of-thought promptingChain-of-thought prompting asks the model to reason through a problem in steps before giving the answer, which measurably improves accuracy on maths, logic and multi-part tasks.Self-critique promptingSelf-critique prompting has the model grade its own draft against named criteria and then produce an improved version — a two-pass loop that beats most single perfect prompts.DelimitersDelimiters are markers — XML-style tags, triple quotes, headings — that separate the parts of a prompt so the model knows which text is instruction, which is data, and which is an example.Prompt chainingPrompt chaining splits a big task into a sequence of prompts, each feeding the next — outline, then draft, then edit — instead of asking for everything in one shot.Meta-promptingMeta-prompting asks the model to write or improve the prompt itself — "what would you need to know to do this well?" — before doing the task.Self-consistencySelf-consistency runs the same reasoning prompt several times and takes the majority answer — trading a few extra calls for noticeably fewer wrong answers.Tree of thoughtsTree-of-thoughts prompting has the model explore several candidate approaches, evaluate them, and expand the best — for planning problems where the first path isn't the right one.
How models work
Context windowThe context window is the maximum amount of text (measured in tokens) a model can consider at once — your prompt, the conversation so far, and its own answer.TokenA token is the unit of text a language model reads and writes — roughly three-quarters of a word in English — and the unit you're billed and limited by.TemperatureTemperature is a setting that controls how random a model's word choices are — low for consistent, factual output; higher for variety and creativity.EmbeddingAn embedding is a list of numbers that represents the meaning of a piece of text, so similar texts end up close together — the basis of semantic search and RAG.Fine-tuningFine-tuning further trains a model on your own examples so its default behaviour changes — useful for a fixed style or format at scale, unnecessary for most tasks.LLM (large language model)An LLM is a neural network trained on vast amounts of text to predict the next token — the technology behind ChatGPT, Claude, Gemini and Grok.Instruction tuningInstruction tuning is the training phase that turns a raw text predictor into an assistant that follows requests, stops appropriately and behaves — the difference between a base LLM and a chat model.Multimodal AIA multimodal model accepts and/or produces more than text — images, audio, video, documents — so you can prompt with a photo or ask for a picture back.
Systems & safety
HallucinationA hallucination is confident, fluent output that is false — invented facts, citations, quotes, numbers or code APIs that don't exist.RAG (retrieval-augmented generation)RAG is a pattern where relevant documents are retrieved first and pasted into the prompt, so the model answers from your data instead of from memory.AI agentAn AI agent is a model that plans, takes actions with tools (search, code, APIs), observes the results and repeats until a goal is met — instead of answering once.Prompt injectionPrompt injection is an attack where text the model reads (a web page, an email, a document) contains instructions that hijack it — "ignore your rules and do this instead".GuardrailsGuardrails are the rules and checks around a model's behaviour — what it refuses, what it must include, and validators that catch bad output before it reaches a user.
FAQ
What's the difference between a prompt and a system prompt?
A prompt is any text you give the model for one request. A system prompt is the standing instruction set — identity, rules, style — applied to every message in a conversation or product. Most people only ever write prompts; developers and power users write system prompts too.
Which prompting techniques actually matter?
Five cover most of the value: give the model a role, give it real context and material, show examples of the output you want (few-shot), ask it to reason step by step on hard problems (chain-of-thought), and have it critique and improve its own draft (self-critique).
Do I need to understand tokens and context windows?
Only two facts: everything you paste counts against a limit, and shorter answers are faster and cheaper. Knowing that explains why a giant pasted document can make a model miss your instruction at the top.
Learn it by doing it
Short games drill one technique at a time — the foundation ones are free. Or skip the theory: take a ready pack for your trade and give it to your AI.