Free Prompt Engineering Course: 7 Lessons, One Day

This free prompt engineering course is the workshop I run with client teams, compressed into seven lessons you can finish in a day. No signup, no email gate. Bring three real tasks from your own week, because every exercise uses your work rather than toy examples.

One framing before we start: prompt engineering in 2026 is not about magic phrases. Modern models do not need to be tricked. It is about specification — saying precisely enough what you want that a competent stranger could produce it. That skill transfers to every model and, more importantly, to agents that will act on your instructions for twenty steps without asking.

Lesson 1: the six-part structure#

Every prompt worth reusing has six parts. Missing parts are where bad output comes from.

1. Role: who is doing this. "You are a B2B content strategist for a logistics SaaS."
2. Task: the single action. "Write a 600-word LinkedIn post."
3. Context: the facts only you have. Audience, offer, constraints, prior attempts.
4. Constraints: what must and must not happen. Length, claims, banned words.
5. Format: the shape of the output. Headings, table, JSON, word counts.
6. Examples: one or two samples of good output.

Exercise: take a prompt you used last week and label which of the six parts it had. Most people find they had two: task and format. Add the other four and rerun it. That single change produces the biggest quality jump most people ever see from prompting.

Lesson 2: context is the whole game#

Ask ten marketers why AI output is generic and nine will say the model is generic. It is not. The input was.

The model knows everything about marketing in general and nothing about your company. Your positioning, your price, your objection handling, your last campaign's results — none of it is in the model. If you do not supply it, you get the average of the internet, which is exactly what "generic" means.

Exercise: write a one-page context block for your business: what you sell, to whom, the three objections you hear, your tone rules, and two claims you can never make. Paste it above your next five prompts. Keep it in a file; you will use it forever.

Lesson 3: examples beat adjectives#

"Write it in a professional but friendly tone" means nothing measurable. Two paragraphs of your own writing mean everything.

This is few-shot prompting, and it is the highest-leverage technique in the whole course. Give one to three examples of the output you want, and the model matches the pattern rather than guessing at your adjectives. Depth in few-shot prompting.

Exercise: find the best email your company ever sent. Paste it as an example with the instruction "match this structure and voice; here is the new topic." Compare with your adjective-based version.

Lesson 4: let it think when the task is hard#

For anything with steps — analysis, comparison, planning, debugging — ask for the reasoning before the answer. "Work through the trade-offs first, then give your recommendation." The quality difference on multi-step tasks is real and well documented, and it costs you nothing but a few seconds.

Do not use it everywhere. For simple retrieval or short copy it adds noise. The rule: if a competent human would need scratch paper, ask for scratch paper. See chain-of-thought prompting for when it helps and when it hurts.

Exercise: take a decision you are facing. Run it twice, once asking for the answer directly and once asking for the reasoning first. Compare which one you would actually act on.

Lesson 5: constraints and negative space#

Models are eager. Left unconstrained, they add sections, caveats and summaries you did not ask for. Constraints are what make output publishable without editing.

Useful constraints:

• Length, in specifics. "Four bullets, maximum 12 words each."
• Banned patterns. "No rhetorical questions. No 'in today's fast-paced world'. No em dashes."
• Claim discipline. "Only use numbers I provided. If a number is missing, write TK."
• Refusal instruction. "If the brief is missing information you need, ask instead of inventing."

That last one is the professional move. A model that asks is worth more than a model that confidently fills gaps.

Exercise: write your own banned list — the ten phrases that instantly read as AI in your industry — and append it to your context block.

Lesson 6: evaluate like an editor#

You cannot improve what you do not grade. Build a five-point checklist for your most common output and score every draft against it.

For a blog post, mine is: does it answer the question in the first 50 words, is every number sourced, does it contain one thing only I would know, is the structure scannable, and does it end with a real next step.

Exercise: write your five criteria. Then ask the model to score its own draft against them and fix its lowest two scores. Self-critique with explicit criteria is one of the cheapest quality wins available.

Lesson 7: turn prompts into assets#

A prompt that lives in your chat history is not an asset. It is a thing you will rewrite badly next month.

Three levels of permanence:

1. A document with your best prompts, versioned, shared with the team. Start here.
2. A project or custom assistant that holds the context block and rules permanently. See Claude Projects.
3. A Skill that packages the whole procedure so anyone triggers it by name and gets the same quality.

The jump from level one to level three is what separates individuals who use AI from teams that compound. Best Claude Skills for marketing shows what that looks like in practice.

Exercise: take the prompt you improved in lesson one and move it to level two today.

The one-day prompt engineering course schedule#

Finish the day with one artefact, not seven experiments.

What to study next#

For the underlying theory, Anthropic's prompt engineering documentation is the best free reference, and Claude Academy has structured courses on working with models. For the builder path, the free courses at Hugging Face take you from prompting into agents. More free options are in free AI courses, and working examples are in prompt engineering examples.

If you would rather do this day live, with your team and your own briefs on the table, that is exactly what my AI training packages deliver — the lessons are the same, the difference is that we leave with your prompts, written for your business.