Skip to content

Command Palette

Search for a command to run...

Start here

Start Here: A Seven-Day Path to Using AI Well

One short session a day. By the end you will write reliable prompts, choose tools sensibly and have a small recurring workflow with safety limits in place.
Seven sessions, about 1 to 2 hours each (estimate)Beginner

Last reviewed . AI products change quickly; check the linked sources before relying on details.

  • 7short sessions, one a day
  • 1 to 2 hoursper session (estimate)
  • Beginnerno prior experience assumed
Your seven-day pathA path of 7 steps: Day 1, Prompting fundamentals; Day 2, Choosing a model; Day 3, A personal assistant workflow with mock data; Day 4, A content engine with fact-checking and originality; Day 5, Sales and lead generation with privacy and anti-spam care; Day 6, Everyday life, including health admin limits; Day 7, Build a recurring workflow.1Day 1Promptingfundamentals2Day 2Choosing amodel3Day 3A personalassistantworkflow withmock data4Day 4A contentengine withfact-checkingandoriginality5Day 5Sales and leadgenerationwith privacyand anti-spamcare6Day 6Everyday life,includinghealth adminlimits7Day 7Build arecurringworkflow
One short session a day, each building on the last.
Production warning

Check facts before you rely on them, keep sensitive data out of unvetted tools, and keep a person in charge of anything that sends, posts, buys or deletes. Business and financial results are never guaranteed, and AI output is not professional medical, legal or financial advice.

Day 1: Prompting fundamentals

Estimated time: About 45 to 60 minutes (estimate).

Objective

Write a prompt that gets a usable first answer.

What you will understand

  • Why context, task, constraints and format matter
  • How to ask the model to list its assumptions

Exercise

Pick a real email or short document. Fill in the universal template twice, once vague and once specific, and compare the results.

Universal prompt structure

A reusable skeleton: role and context, task, constraints, format, then checks. Fill in the brackets and paste it into any assistant.

Model-agnostic templates
Use it for
Turning a vague request into one that gets a usable first answer
Who it is for
Anyone starting out with an AI assistant
You will need
  • Your goal in one sentence
  • Any background the assistant cannot guess
  • The format you want back
What you get back
A focused answer in the format you asked for, plus a list of assumptions and items to verify.
Prompt
ROLE AND CONTEXT
You are helping [WHO YOU ARE / YOUR SITUATION]. Background: [2-4 SENTENCES OF CONTEXT].

TASK
[ONE CLEAR TASK, e.g. "Draft", "Summarise", "Compare", "Plan"].

CONSTRAINTS
- Audience: [WHO WILL READ OR USE THE RESULT]
- Length: [WORD OR ITEM LIMIT]
- Tone: [e.g. plain, friendly, formal]
- Use only the information I have given you. If something is missing, ask me before guessing.

FORMAT
Return [BULLETS / TABLE / NUMBERED STEPS / SHORT EMAIL].

CHECKS
Before you answer, list any assumptions you made. After your answer, list the three claims I should verify myself.

Check the output, safety and limits

Verify

  • Check names, numbers, dates and quotes against the original source before using them.
  • Read the assumptions list first; wrong assumptions are the most common failure.

Safety and privacy

  • Do not paste passwords, card numbers, government IDs, health records or other people's private details into a chat tool.

Limitations

  • Models can still invent details. The checks section reduces this but does not remove it.

Learn more: Seven-day start here path

Expected result

A noticeably better second result, and a saved template you reuse.

Common mistakes

  • Asking several things at once
  • Leaving out the audience
  • Accepting the first answer without checking facts

Safety and privacy

Do not paste passwords, ID numbers or other people's private data.

Next: Day 2, Choosing a model

Day 2: Choosing a model

Estimated time: About 60 minutes (estimate).

Objective

Understand how assistants differ and test two on the same task.

What you will understand

  • Model families differ in strengths, file support, web access, privacy terms and price
  • Names and versions change quickly, so check current official pages

Exercise

Take one prompt from Day 1. Run it in two assistants and note quality, speed, cost and what data settings each offers. Read the comparison guide first.

Read first: Compare AI models.

Rewrite a prompt for a different model

Ask one assistant to adapt a prompt you already use so it works well on another tool.

Model-agnostic templates
Use it for
Moving a working prompt between ChatGPT, Claude, Gemini or an open-source model
Who it is for
People using more than one assistant
You will need
  • The prompt that works today
  • The name of the target assistant
  • Any file or tool differences you know of
What you get back
A revised prompt plus a short list of features to check in the target tool.
Prompt
Here is a prompt that works well for me in [CURRENT TOOL]:

"""
[PASTE PROMPT]
"""

I want to use it in [TARGET TOOL]. Rewrite it so it still gets the same kind of result. Point out anything that depends on features the target tool may not have (file upload, web search, memory, long documents). Keep the placeholders in [SQUARE BRACKETS] so I can reuse it. Do not add new requirements.

Check the output, safety and limits

Verify

  • Run the old and new prompts on the same sample input and compare the results side by side.

Limitations

  • The assistant may not know the target tool's latest features. Check the tool's own help pages.

Learn more: Compare AI models

Expected result

A short note on which tool suits which task for you.

Common mistakes

  • Choosing on brand or hype alone
  • Ignoring data and training settings
  • Assuming a free tier matches the paid one

Safety and privacy

Review each provider's data-use and retention settings before pasting work content.

Next: Day 3, A personal assistant workflow with mock data

Day 3: A personal assistant workflow with mock data

Estimated time: About 60 to 90 minutes (estimate).

Objective

Design an inbox-and-calendar helper and define what it may and may not touch.

What you will understand

  • Propose-then-approve is safer than act-automatically
  • Permission boundaries: read-only, one label, no send, no delete
  • Email text can contain instructions; treat it as data, not commands

Exercise

Write five invented emails and a mock calendar. Run the prompt. Then write a one-paragraph permission statement for the real version.

Personal assistant practice with mock data

Rehearse an inbox-and-calendar assistant using invented data before connecting anything real.

Productivity
Use it for
Testing what an assistant produces without exposing real messages
Who it is for
Beginners building a personal assistant workflow
You will need
  • Five invented emails you write yourself
  • An invented calendar for the week
What you get back
Summaries, proposed tasks, clashes and a list of open questions.
Prompt
You are my planning assistant. Below are five EXAMPLE emails and my EXAMPLE calendar. They are invented for practice.

EMAILS:
[PASTE 5 INVENTED EMAILS]

CALENDAR:
[PASTE INVENTED CALENDAR]

Produce: (1) a two-line summary per email, (2) proposed tasks with a suggested day, (3) any clash with the calendar, (4) a list of things you could NOT decide without asking me. You may only propose; do not claim to have sent, booked or changed anything.

Check the output, safety and limits

Verify

  • Compare each proposed task with the invented email it came from.

Safety and privacy

  • Use invented data only on this day.
  • Do not connect your real mailbox until you have defined permissions.

Expected result

A digest of proposed tasks and a written permission boundary.

Common mistakes

  • Connecting your real mailbox first
  • Granting full-account access when one label would do
  • Letting the assistant send or accept invitations

Safety and privacy

Mock data only today. Real mail may contain third-party and sensitive information.

Next: Day 4, A content engine with fact-checking and originality

Day 4: A content engine with fact-checking and originality

Estimated time: About 90 minutes (estimate).

Objective

Produce a draft from your own ideas and verify every claim.

What you will understand

  • AI drafts reflect common patterns; originality comes from your experience
  • Every fact needs a source you opened
  • Disclosure and platform rules can apply to AI-assisted content

Exercise

Choose a topic you know. Supply your angle and examples, generate the outline and claims checklist, verify each claim, then write the final in your own words.

Article outline with a fact-check list

Plan a piece of content and produce a list of claims that need a source before you publish.

Writing & Content
Use it for
Drafting original content that is checked before it goes live
Who it is for
Bloggers, marketers, newsletter writers
You will need
  • Topic
  • Your own experience or angle
  • Audience
  • Sources you already trust
What you get back
An outline and a checklist of claims, with no invented numbers or citations.
Prompt
I am writing about [TOPIC] for [AUDIENCE]. My angle, based on my own experience: [YOUR ANGLE AND EXAMPLES].

1. Propose an outline with 5-7 headings.
2. Under each heading, give two bullet points that use MY angle and examples, not generic advice.
3. List every factual claim the finished piece would need (statistics, dates, product features, legal points) as a checklist titled "Claims to verify". Do not supply statistics or citations yourself; I will find them.
4. Suggest what kind of source would settle each claim (e.g. official documentation, regulator, peer-reviewed paper).

Check the output, safety and limits

Verify

  • Resolve every item on the checklist with a real source you have opened.
  • Run a plagiarism or similarity check if the piece will be published under your name.

Safety and privacy

  • Disclose AI assistance where your platform, employer or publisher requires it.

Limitations

  • The model may still slip a statistic into the outline. Remove any figure you did not source yourself.

Expected result

An outline, a resolved checklist and a draft you stand behind.

Common mistakes

  • Publishing unchecked statistics
  • Copying generated text wholesale
  • Using someone else's work as input without rights

Safety and privacy

Respect copyright and licences, and disclose AI assistance where required.

Next: Day 5, Sales and lead generation with privacy and anti-spam care

Day 5: Sales and lead generation with privacy and anti-spam care

Estimated time: About 90 minutes (estimate).

Objective

Prepare a respectful, accurate first contact using public business information.

What you will understand

  • Unsolicited commercial messages are regulated; examples include the Spam Act in Australia, CAN-SPAM in the US and GDPR/ePrivacy in the EU and UK
  • Consent, honest identification and a working opt-out are common requirements
  • Rules vary by country and recipient type, so seek local advice

Exercise

Choose three real companies. Gather public text yourself, run the research brief, then write one email by hand using the draft prompt as a reference. Do not send anything today.

Prospect research brief from public information

Prepare for an outreach call using only information you have collected from public pages.

Business & Sales
Use it for
Preparing a relevant, respectful first contact
Who it is for
Small business owners, sales and partnerships staff
You will need
  • Company name
  • Public text you copied (website, press release)
  • What you offer
What you get back
A short brief with evidence for each suggested need.
Prompt
I sell [YOUR OFFER] to [TYPE OF BUSINESS]. Here is public information about [COMPANY]:

"""
[PASTE PUBLIC TEXT]
"""

Using only this text:
1. Summarise what the company does in 3 sentences.
2. List 3 possible needs related to my offer, each tied to a quote from the text.
3. Suggest 2 questions to ask on a first call.
4. Say what you could NOT tell from the text.
Do not guess about individual people's personal circumstances.

Check the output, safety and limits

Verify

  • Confirm key facts on the company's own site before you mention them.

Safety and privacy

  • Use only business information. Check consent and local rules (for example Spam Act, CAN-SPAM, GDPR) before contacting anyone.
  • Do not assemble profiles of private individuals.

Expected result

Three sourced briefs and one reviewed draft.

Common mistakes

  • Scraping or buying contact lists without checking consent
  • Automating bulk sending
  • Making claims you cannot support

Safety and privacy

Business information only. This page is not legal advice.

Next: Day 6, Everyday life, including health admin limits

Day 6: Everyday life, including health admin limits

Estimated time: About 60 minutes (estimate).

Objective

Use AI for household planning and appointment preparation within safe limits.

What you will understand

  • AI is useful for organising and drafting
  • AI is not a substitute for medical professionals and must not be used to diagnose or choose treatment
  • Keep identifying health details out of tools you have not vetted

Exercise

Plan a week of dinners with the meal prompt from the library, then prepare questions for a real or imaginary appointment with the prompt below.

Prepare questions for a medical appointment

Organise your own notes into questions for your doctor. It does not diagnose or recommend treatment.

Everyday Life
Use it for
Health admin: structuring questions and a timeline
Who it is for
Anyone preparing for an appointment
You will need
  • Your own short notes on symptoms or concerns, without identifying details
  • Appointment length
What you get back
A timeline, prioritised questions and a bring-with-you list.
Prompt
I have an appointment with my doctor in [DAYS] and about [MINUTES] minutes. Here are my notes (no names or ID numbers):
[NOTES]

Organise them into: a short timeline, the three most important questions to ask, and a list of information to bring (for example medication list, dates). Do not diagnose, suggest conditions or recommend treatments. If anything sounds urgent, tell me to contact emergency services or my doctor now.

Check the output, safety and limits

Safety and privacy

  • AI is not a substitute for medical professionals.
  • In an emergency contact local emergency services, not a chatbot.
  • Leave out names, date of birth and ID numbers.

Expected result

A meal plan and an appointment question list.

Common mistakes

  • Treating answers as medical advice
  • Pasting records with names and numbers
  • Relying on generated dosages or interactions

Safety and privacy

For anything urgent or clinical, contact a doctor or emergency services.

Next: Day 7, Build a recurring workflow

Day 7: Build a recurring workflow

Estimated time: About 2 hours (estimate).

Objective

Specify and build a small scheduled workflow in n8n, Make or Zapier.

What you will understand

  • A workflow has a trigger, actions, error handling and monitoring
  • Platforms have execution limits and schedule intervals that depend on plan; read the current documentation
  • A workflow runs continuously only if the platform and plan support it, and you can verify that in its run history

Exercise

Write the specification, then build only the first two steps in a free or trial plan using test data. Add an error notification and check the execution log after a manual run.

Read first: AI agents and workflow platforms.

Write a workflow specification before building

Define trigger, steps, permissions, approvals and failure handling in plain language first.

Productivity
Use it for
Planning a workflow in n8n, Make or Zapier
Who it is for
Anyone building a first recurring workflow
You will need
  • The task you repeat
  • How often it happens
  • Apps involved
What you get back
A one-page workflow specification you can follow in a builder.
Prompt
I repeat this task: [TASK] about [FREQUENCY]. Apps involved: [APPS].
Write a specification for an automation with these headings: Trigger; Steps (mark each as fixed rule or AI step); Permissions needed per app (minimum possible, read-only where possible); Human approval point; What happens on error; How I will monitor it; What it could cost per run (describe the drivers only, no figures); Things that must never be automated. Keep it under one page.

Check the output, safety and limits

Verify

  • Check the permission list against what each app actually offers.

Safety and privacy

  • Ask the assistant to flag any step where it could send, post, buy or delete.

Expected result

A working test run, an error alert and a one-page specification.

Common mistakes

  • Building before specifying
  • No error alerts
  • Granting broad permissions for convenience
  • Assuming a workflow ran because no error appeared

Safety and privacy

Use test data, least-privilege access and a human approval step before any outbound action.

Next: learn how agents differ from workflows

Frequently asked questions

How do I start learning to use AI step by step?

Follow the seven-day path, with one short session a day. It runs from prompting fundamentals and choosing a model through a mock-data assistant, a fact-checked content draft, sales research, everyday life and health admin limits, to building a small recurring workflow. Session times are estimates, from about 45 minutes to about 2 hours.

How long does the seven-day AI course take?

Each day is estimated at roughly 45 minutes to 2 hours, depending on the day. Day 1 is about 45 to 60 minutes, and Day 7, building a recurring workflow, is about 2 hours. These are estimates, so your time will vary with your tasks and how carefully you check the results.

Why does the course use mock data for the personal assistant exercise?

Real mail can contain third-party and sensitive information. On Day 3 you write five invented emails and a mock calendar, then define what an assistant may and may not touch. Propose-then-approve is safer than acting automatically, and you should not connect a real mailbox until permissions are defined.

Can AI give me medical advice?

No. The course states that AI is not a substitute for medical professionals and must not be used to diagnose or choose treatment. The Day 6 exercise only organises your own notes into questions for your doctor. Leave out names and ID numbers, and contact emergency services for anything urgent.

Can I use AI to send sales emails?

Be careful. Unsolicited commercial messages are regulated, with examples including the Spam Act in Australia, CAN-SPAM in the US and GDPR and ePrivacy in the EU and UK. Consent, honest identification and a working opt-out are common requirements. Rules vary by country, so seek local advice. This is not legal advice.

Where to go next