Working with AI assistants: prompting that gets usable output

Write requests that get usable answers the first or second time, check what comes back, and keep a prompt library you reuse.

Most people use an AI assistant the way they use a search box: a few words in, a generic answer out, then they decide AI is overrated or spend twenty minutes fixing the result. Usually the request was the weak link, since an assistant knows nothing about your situation, your reader or your standards unless you tell it. This course teaches you to brief it the way you would brief a capable new colleague, to steer it when the first answer misses, and to check what it gives you before your name goes on it. You keep the prompts that work in a library, so each good request only has to be written once.

What you'll be able to do

• Write a request with context, task, constraints and format that gets a usable first draft
• Use examples and style samples to get output that matches a standard you set
• Break a large job into a sequence of smaller requests and review each step
• Give feedback that improves a draft instead of regenerating and hoping
• Get accurate, quoted answers from long documents and files you upload
• Set up custom instructions, projects and memory so you stop repeating yourself
• Check an answer for errors with a routine matched to the stakes
• Build and maintain a personal library of tested prompt templates

Syllabus

Module 1: Write a request an assistant can act on

Learn the four parts of a request that gets a usable answer, context, task, constraints and format, and practise rewriting vague prompts into complete briefs.

• 1.1 The assistant only knows what you put in front of it
• 1.2 Give context the way you would brief a colleague
• 1.3 Set constraints and ask for the format you need
• 1.4 Rewrite five weak prompts into strong ones

Module 2: Show it what good looks like

Use examples, style samples and counter-examples to get output that matches a standard, and understand what assigning a role does and does not change.

• 2.1 One good example beats a paragraph of description
• 2.2 Style samples and examples of what to avoid
• 2.3 What telling it to act as an expert really does
• 2.4 Build an example set for a task you repeat

Module 3: Break a big job into steps

Get better results on large tasks by splitting them into a sequence of smaller requests, asking for a plan first, letting the assistant ask you questions, and using reasoning modes well.

• 3.1 Why one giant prompt gives you a mediocre result
• 3.2 Ask for the plan before the work
• 3.3 Let the assistant ask you questions first
• 3.4 When to use a reasoning mode
• 3.5 Run a five-step chain on a real task

Module 4: Improve a draft instead of starting over

Treat the first answer as a draft and steer it with specific feedback, knowing when to edit, when to regenerate and when to start a fresh chat.

• 4.1 The first answer is a draft, so treat it like one
• 4.2 Feedback that gets a better second draft
• 4.3 Edit, regenerate, or start a fresh chat
• 4.4 Take a draft through three rounds

Module 5: Get accurate answers from files and long documents

Understand what happens when you upload a file, ask for quoted and located answers, and compare or extract information across several documents without losing accuracy.

• 5.1 What happens when you upload a file
• 5.2 Make it quote and point to the source
• 5.3 Compare and extract across several documents
• 5.4 Build a checked summary of a long report

Module 6: Set it up once with projects, memory and custom instructions

Stop repeating yourself by putting standing context into custom instructions, projects and memory, and keep control of what the assistant remembers about you.

• 6.1 Custom instructions: what to say once
• 6.2 Projects: a workspace for each recurring job
• 6.3 Memory: what it keeps and how to manage it
• 6.4 Set up your assistant for your working week

Module 7: Check the output before your name goes on it

Match how hard you check to what is at stake, use the assistant to critique its own work without trusting it blindly, and check numbers, quotes and code the right way.

• 7.1 Match the check to what is at stake
• 7.2 Ask for a critique, and know its limits
• 7.3 Checking numbers, quotes, sources and code
• 7.4 Audit three answers you were about to use

Module 8: Choose the right assistant for the job

Compare assistants on the things that matter to you, check their data settings and your company's rules, and run a fair side-by-side test instead of relying on reviews.

• 8.1 What actually differs between assistants
• 8.2 Data settings and the rules at your workplace
• 8.3 Run a fair side-by-side test

Module 9: Build a prompt library you will actually reuse

Turn prompts that worked into templates with blanks to fill, store them where you will find them, and keep the library tested and current.

• 9.1 Turn a prompt that worked into a template
• 9.2 Store and organise it where you work
• 9.3 Build and test your starter library

Frequently asked questions

How long does the course take?

About nine hours across nine modules, including the exercises. Most people finish in three to four weeks at two or three hours a week.

Which assistant should I use for the course?

Any mainstream assistant, such as ChatGPT, Claude, Gemini or Copilot. Free plans cover most exercises, though some features like projects or large file uploads may need a paid plan. Check the current plan details on each product's website.

Do I need to take AI fundamentals first?

It helps but is not required. If you already understand that assistants predict text, have a limited context window and can invent facts, you can start here.

How is this different from AI at work?

This course teaches the general skill of directing any assistant. AI at work: writing, research, spreadsheets and meetings applies that skill to specific jobs such as email, research, spreadsheets, slides and meetings. Take this one first.

Will these techniques stop working when the models change?

The core skills, clear briefs, examples, steps, feedback and checking, have held up across many model versions because they are how you would brief a person too. Product features will change, which is why the course teaches you to check the help pages rather than memorise menus.

Explore more

• All Academy courses