A five-step path to writing prompts that reliably get you the output you actually wanted, instead of a
close-but-wrong answer you have to fight with. This is a skill you'll reuse in every other path on this
site — it's worth doing properly once.
Who this is for: you already use an AI chatbot or tool regularly, but you're getting
inconsistent or mediocre results and suspect the problem is how you're asking, not the model itself.
Who this is not for: if you haven't used an AI chatbot at all yet, start with the
AI Fundamentals path first — step 2 there is a gentler on-ramp to prompting basics than this path assumes. If your goal is
wiring prompts into an automated workflow rather than getting better one-off chatbot answers, the
No-Code AI Automation path covers that once you finish here.
The sequence
1
Learn the structure behind a good prompt
Goal: understand why vague prompts get vague answers, and what specificity, context, and constraints each actually do.
Learn Prompting is the most complete free curriculum on this specific skill — it covers the core
techniques (role prompting, few-shot examples, chain-of-thought) in a structured order rather than as
scattered tips. Work through the basics section fully before jumping to advanced techniques.
Watch each technique applied to a real, messy prompt
Goal: see the gap between a first-draft prompt and a rewritten one, side by side, so you recognize the pattern in your own prompts.
Reading a technique and recognizing when to use it are different skills. This channel walks through
actual before/after prompt rewrites rather than abstract advice, which reinforces step 1 with
concrete examples.
Learn how one specific model actually wants to be prompted
Goal: recognize that "good prompting" isn't fully universal — each model family has documented quirks and preferences worth knowing.
Anthropic publishes its own prompting guidance for Claude directly, written by the people who built
the model — more reliable than third-party guesswork about what a specific model responds well to.
Learn the same idea from the other major model family
Goal: confirm which prompting techniques are genuinely universal versus specific to one vendor's models, and pick up patterns like function calling and system prompts.
OpenAI's own docs and cookbook cover prompting patterns for GPT models directly, including practical
ones — system prompts, few-shot formatting, structured output — that go beyond what step 1 covered
in general terms.
See prompting used inside a real AI agent, not just a chat window
Goal: understand how prompts function differently once they're embedded in code — as system instructions, tool descriptions, and context injection — instead of typed by hand each time.
This is the bridge from "I write better chatbot messages" to "I can write the prompts that make an
AI agent behave correctly," which is where prompting actually gets used once you're building
anything beyond a chat window.
Pick a real prompt you type regularly and are lukewarm on the results of — a work email draft, a code
explanation, a research summary, anything you repeat often enough to have a "usual" version. Write it
exactly as you normally would. Then rewrite it twice more, applying a different technique from steps
1–2 each time (add explicit constraints, add a role, add one worked example). Compare all three outputs
side by side.
You're done when you can show:
Your original prompt and its output, plus two rewritten versions and their outputs
A one-line note on which technique made the biggest difference and why
A reusable prompt template you'll actually keep using for that task going forward
Where to go next
If step 5 was the most interesting part, the No-Code AI Automation path picks up right
where this leaves off — wiring the prompts you just practiced into actual automated workflows. If steps
3–4 were the most interesting part, the rest of the Hands-On Coding category on the
dashboard goes deeper into building with these models directly. An AI Application Developer path is
planned — check back or use the dashboard directly in the meantime.
This path was last reviewed and confirmed against live resource listings on 2026-07-21.