Prompt Patterns & Techniques
Move past one-off prompts and learn the named patterns pros reach for — worked examples, chain-of-thought reasoning, strict output shapes, and delimiter guardrails.
Asking for JSON, tables, and lists
Make the output ready to use If a human has to reformat every answer, the model has only done half the job. When you specify a strict output shape, the result drops straight into a spreadsheet, a script, or a document — no cleanup. Ask for JSON with named keys To make output machine-readable, name […]
Chaining: when one prompt is too many jobs
The prompt that does four things does none of them well A request like “read these fifty reviews, find the themes, rank them by severity, and draft a response to each” looks efficient. It is the most reliable way to get mediocre output. Quality degrades on every sub-task, and when the result is wrong you […]
Diagnosing a pattern that stopped working
Every pattern in this course has a failure mode Patterns are not magic words. Each one works by changing what the model attends to, and each fails in a way you can recognise once you have seen it. Few-shot drifting into copying Symptom: the output reuses details from your examples — the same company name, […]
Choosing the cheapest pattern that works
Every pattern costs something Patterns are usually taught as pure upside. In production they are trade-offs: examples cost input tokens on every single call, reasoning costs output tokens and latency, chains cost round trips, self-checking multiplies everything by two or three. The ladder Start at the bottom and stop as soon as the output is […]
Working with Claude
Get practical results from Claude — what it is good at, how to feed it long documents, how to ask for structured output, and how to use it responsibly.
Structured outputs & step-by-step
Ask for the exact shape you need Left to its own devices, Claude answers in prose. But you can ask for almost any shape — and getting the shape right the first time saves you cleaning it up afterwards. Name the format Bullets — “Give me five bullet points, one line each.” A table — […]
Working across many documents at once
A large window changes what you can ask Claude’s long context is usually described as “you can paste more”. The more interesting consequence is that you can ask questions no single document answers — comparisons, contradictions, and gaps across a whole set of material. Questions that need several documents at once Comparison — “Here are […]
Projects, memory and reusable setups
Stop re-explaining yourself every morning If your first three messages each day set up the same context — who you are, what the company does, how you like output formatted — that setup belongs somewhere persistent rather than being retyped. What to put in a persistent setup The things that are true every time: your […]
ChatGPT: Reliable Results
The intermediate rung: give ChatGPT structure and examples, get output in the shape you need, and verify it before you use it.