What prompts are — and what prompting can and cannot do
Explain what a prompt is and name three things prompting can improve and three it cannot.
Welcome
This free course teaches one skill in depth: communicating with AI systems so that they produce useful, checkable results. You will write prompts, test them, look critically at what comes back, and improve them step by step. By day 22 you will have built a small, documented prompt toolkit for a real task of your choice.
The course takes about 22 days at roughly one hour a day (around 22 hours in total). That is a suggested pace, not a deadline: your progress is saved and you can pause at any time. You can read every lesson without an account; register (free) if you want to save drafts, track progress and take the assessments.
What a prompt is
A prompt is the input an AI model receives before it responds. In a chat app it looks like your message, but the model may also receive things you do not see — the provider's own instructions, earlier messages in the conversation, or files you attached. For this course we treat the prompt as everything you control: your instructions, the context you add, any material you paste in, and how you ask for the answer to be shaped.
A useful way to think about it: the model only knows what is in its training and what is in the prompt. It does not know your job, your audience, your deadline, your organisation's rules or what "good" means to you — unless you say so.
What good prompting changes
Clear prompts reliably improve:
- Relevance — the answer addresses your situation rather than a generic one.
- Shape — length, structure, tone and format match what you need (a table, a checklist, three options, 120 words).
- Effort saved — fewer rounds of "no, I meant…".
- Consistency — a well-written prompt gives similar quality each time, so you can reuse it.
What prompting cannot fix
Prompting is powerful, but it is not magic. No wording can:
- Guarantee facts. Models generate plausible text; they do not check it against reality. A polite, detailed prompt can still receive a confident, wrong answer.
- Add knowledge the model does not have. If the information is recent, private, or highly specialised, you must supply it or use a tool that can look it up — and check the result.
- Remove your responsibility. You decide what to send, publish or act on. "The AI said so" is not a defence for an error.
- Make an unsuitable task suitable. Some tasks (legal, medical or financial decisions about real people; anything requiring verified data) need qualified humans and reliable sources, whatever the prompt says.
Keep both lists in mind. This course is about getting the first list right without forgetting the second.
Prompts, outputs and iterations
You will hear three words throughout the course:
| Term | Meaning |
|---|---|
| Prompt | What you send to the model |
| Output (or response) | What the model returns |
| Iteration | A revised prompt, written because the last output fell short |
Professional prompt writers rarely get the best result on the first attempt. They write a reasonable first prompt, look carefully at the output, decide what is wrong, change one or two things, and try again. Module 6 turns this into a repeatable method.
Your prompt log
Create a prompt log today — a document, spreadsheet or notebook page. For every prompt experiment in the course, record:
- Date and tool (which assistant or model you used, if you know it).
- Goal — what you wanted.
- Prompt — the exact text you sent.
- Output summary — what came back, and anything wrong with it.
- Next change — what you will try next and why.
The workbook in this lesson's resources contains a ready-made log page. You will use the log for every exercise and it becomes evidence for your final project.
Which AI tool should I use?
Any general-purpose AI assistant works for this course, including free tiers. Tools differ in quality, features (web search, file upload), data policies and limits, so note which one you use in your log. Do not paste anything confidential into any tool until you have read Module 10 — and check your school's or employer's policy first. If you have no access to an AI assistant, you can still complete the course: write the prompts, predict the outputs and explain your reasoning.
Key ideas
- A prompt is everything you give the model; the model only knows its training plus your prompt.
- Good prompts improve relevance, shape, effort and consistency.
- Prompts cannot guarantee truth, supply missing knowledge or remove your responsibility.
- Improvement comes from iteration, recorded in a log.
Practice exercise
Set up your prompt log and make your first entry:
- Pick a small, low-stakes task — for example "plan a 30-minute study session" or "suggest names for a book club".
- Write the shortest prompt you can for it and, if you have an assistant, run it.
- In your log, note the goal, the prompt, a one-line summary of the output, and one thing you would change to make the answer more useful to you.