Using models well · Part 1 — The AI Dictionary

The instructions and context you give a model — the single biggest lever on the quality you get back.

A prompt is everything the model reads before it answers: your question, pasted material, examples, and any instructions about format or tone. Small changes — adding who the audience is, showing an example, asking for a table — can transform the result.

Behind-the-scenes instructions that set a model's role, rules and style for the whole conversation.

Apps put a system prompt in front of every chat to define behavior: "You are a support agent for Acme. Be concise. Never discuss pricing." You control your own version through features like ChatGPT's custom instructions and Projects, Claude's Projects and styles, and Gemini's Gems.

A setting that controls randomness — low for precise, repeatable answers; high for creative variety.

When a model picks its next token it has a list of candidates with probabilities. At low temperature it almost always takes the top choice (consistent, predictable). At higher temperature it samples more adventurously (varied, creative, occasionally weird).

Showing the model a few examples of what you want before asking it to do the real task.

Instead of describing a format in words, you demonstrate it: two or three input → output pairs, then the new input. Models are excellent pattern-matchers, so examples often beat paragraphs of instructions — especially for tone, labels and structure.

Example: Two examples of product descriptions in your brand voice, then: "Now write one for the blue kettle."

Having the model work through a problem step by step before giving its answer.

Asking a model to reason step by step — or using a reasoning model that does it automatically — improves accuracy on multi-step problems like math, logic, planning and debugging. The intermediate steps give the model room to catch mistakes.