Don't use AI just to get answers. Use it to change your behaviour.
AI assistants can explain complicated topics, generate alternatives, examine your assumptions, review documents, help you practise conversations, plan projects and create working code. But producing an impressive plan is not the same as improving your life.
The useful loop is: think → test → act → observe → adjust. AI can accelerate every stage. It cannot live your life for you.
Start with a real problem, not a fashionable tool. Your situation, constraints and values are the inputs. Small actions in the real world are the outputs.
“Motivate me to get fit.”
You receive a polished pep talk. Tomorrow, nothing changes.
“Design a 10-minute first step.”
You receive a doable plan, identify obstacles, try it, then report what actually happened.
The LEVEL system
A simple framework for using AI constructively—not merely consuming more advice.
Locate the actual problem
Describe the situation in concrete terms: what is happening, what you have tried, and what outcome would count as progress.
Examine assumptions
Ask for blind spots, competing explanations, hidden constraints and the strongest case against your favourite idea.
Visualise options
Request two or three viable paths, with cost, effort, uncertainty and likely trade-offs. Avoid false precision.
Execute one small step
Translate the answer into one calendar entry, message, experiment, practice session or deliverable.
Learn from reality
Return with evidence: what happened, what did not, what surprised you. Revise the plan rather than requesting endless new ones.
“Help me improve [area]. First ask me up to five questions about my situation. Then identify the likely bottleneck, give me three options with trade-offs, challenge your own recommendation, and finish with one action I can complete in 15 minutes.”
Prompting is not magic wording. It's giving useful context.
You do not need a collection of secret commands. You need to explain the task, offer relevant details, state constraints and define a useful output.
Context
“I have 20 minutes a day, a limited budget, and I am starting from zero.”
Task
“Build a beginner practice plan that fits those limits.”
Constraints
“No paid software. Do not assume I can work weekends.”
Output
“One-page checklist, the first action, and common failure points.”
Five requests that improve almost any answer
- “What assumptions are you making?”
- “Which parts of this are uncertain or need checking?”
- “Give me the best argument against this.”
- “Make it simpler without losing the important details.”
- “What is the next physical action I should take?”
15 areas where AI can be useful
Clearer thinking
Challenge assumptions, untangle confusing problems and compare arguments.
Learning
Private tutoring, practice questions, examples and feedback.
Health preparation
Understand general information and organise questions for professionals.
Money decisions
Understand bills, model budgets and evaluate spending choices.
Habits & routines
Design manageable systems and troubleshoot why they break.
Career transitions
Map skills, rehearse interviews and compare pathways.
Writing
Generate alternatives, critique drafts and clarify structure.
Coding & building
Prototype websites, debug issues and learn software concepts.
Relationships
Practise conversations and explore different perspectives.
Household admin
Organise maintenance, paperwork, shopping and schedules.
Research
Frame questions, find sources, assess evidence and track unknowns.
Creativity
Experiment with new formats, constraints and creative directions.
Communication
Rewrite unclear messages while preserving your meaning.
Planning
Compare scenarios, schedules and realistic resource limits.
Accessibility
Convert complex instructions to plain language and alternative formats.
A thinking partner, not an agreement machine.
AI assistants can be excessively agreeable. If you ask, “Is my idea brilliant?”, the response may flatter you. Request a rigorous critique instead.
Use a structured debate
- State a belief in one sentence.
- Ask for the strongest evidence for and against it.
- Identify what evidence would change your mind.
- Distinguish facts, interpretations and values.
- Write a more carefully qualified conclusion.
When you are stuck in a loop
Ask for a short summary of the decision, the uncertainty you can tolerate, and one bounded next action. Repeated reassurance-seeking can make anxiety and rumination worse; don't use an AI to rehearse the same fear indefinitely.
AI can help structure reflection. It is not a reliable judge of other people's secret intentions, nor proof that a feared scenario is real. When thoughts feel threatening or difficult to reality-test, involve a trusted person or qualified clinician rather than repeatedly interrogating a chatbot.
Build capability, not dependence.
Don't request only an explanation; make the assistant test what you understand. Retrieval practice, examples, corrective feedback and applied exercises are more valuable than passively reading a fluent answer.
Diagnose
“Ask five questions to find my actual skill level.”
Explain
“Teach one concept with a concrete example, then a contrasting example.”
Practise
“Give me a problem. Wait for my attempt before showing the solution.”
Transfer
“Give me a realistic project where I must use this without hints.”
Useful for languages, maths, job skills, philosophy, coding, writing, history, public speaking and anything that improves through repeated feedback.
Prepare better questions. Don't outsource medical decisions.
AI can explain medical terminology in plain English, organise a symptom timeline, generate questions for an appointment, summarise publicly available clinical guidance and help understand the difference between established evidence and hypotheses.
Good uses
- “Help me turn these notes into a concise doctor-appointment summary.”
- “What questions should I ask about the known benefits, risks and alternatives of this treatment?”
- “Explain this research result and distinguish absolute risk from relative risk.”
- “Help me prepare a realistic sleep routine I can discuss with my clinician.”
Do not rely on an AI to diagnose an illness, interpret alarming symptoms, change psychiatric medication, calculate a personal dose or replace urgent care. Outputs can be inaccurate. For urgent symptoms or danger, seek immediate human assistance. In Australia, call 000 for emergencies.
Make the numbers visible.
AI is useful for organising financial choices, but its arithmetic, tax claims, rates and regulatory advice still need verification. Provide fictionalised or redacted data when possible.
Understand spending
Classify a redacted expense list, distinguish recurring from one-off costs and find areas worth reviewing.
Compare decisions
Calculate the total cost of a subscription, loan or purchase under different assumptions. Check figures independently.
Prepare questions
Understand terminology before speaking to an accountant, lender or financial adviser.
Create a plan
Build a simple budget with essential expenses, buffers and review dates.
For Australian information, cross-check with ASIC's Moneysmart and the ATO, depending on the question.
Stop building perfect plans you won't follow.
The biggest trap is turning AI into an endless planning machine. A sophisticated spreadsheet is not evidence of a changed habit.
Build a minimum viable habit
- Pick one behaviour you want to repeat.
- Set the minimum version so small you can do it on a difficult day.
- Attach it to an existing cue, time or location.
- Anticipate a specific obstacle.
- Review actual completion after one week rather than starting over daily.
Make the assistant reduce the scope until the plan fits your life. Then close the chat and do it.
Use AI to produce work, not just discuss it.
Draft a proposal, rehearse an interview, analyse a job description, improve a CV without fabricating qualifications, create a website outline, compare business ideas, or write a clear client handover.
A useful job-search loop
Find a real job posting → extract selection criteria → match to demonstrated experience → identify gaps → rehearse examples → revise the application → verify facts and submit.
A useful business loop
Describe a customer problem → ask the AI to challenge market assumptions → conduct real customer interviews → sketch the smallest paid or usable offer → test demand before expanding.
Use public or authorised material. Never paste confidential employer or client information into a consumer AI account without permission and appropriate data safeguards.
Let AI expand your possibilities without replacing your voice.
Generate alternative structures, ask for line edits, prototype a web page, question plot logic, build a simple tool or get a technical explanation. Keep decisions about taste, meaning and originality with the human creator.
Writing
“Identify what feels generic, vague or unsupported in this draft. Do not rewrite it unless I ask.”
Design
“Give me three visual directions, with typography, layout rules and accessibility concerns.”
Coding
“Explain the bug, propose a minimal patch, and include tests and rollback steps.”
Original ideas
“Show me where this concept resembles existing work and suggest more distinct approaches.”
Check generated code, images, claims and rights before publishing. Models can hallucinate APIs, introduce security flaws and produce material that needs copyright review.
Rehearse better communication.
Use AI to practise a difficult conversation, identify inflammatory wording, distinguish observation from accusation, and think through what a reasonable boundary might sound like. Ask for multiple plausible interpretations rather than a verdict on someone's character.
“I want to discuss [issue] with someone. Help me distinguish what I observed from what I assumed. Write a short, calm opening that explains my needs without assigning motives. Then role-play a few possible responses.”
Don't treat a chatbot's interpretation of a partner, friend or family member as established fact. It has only the account you provided.
Take friction out of ordinary tasks.
- Meals: create a week's realistic meals based on food already in the cupboard, then check food safety and allergies.
- Household: make a rotating cleaning plan sized to available energy and time.
- Travel: compare routes and itineraries, but verify timetables, prices and closures.
- Documents: summarise complicated letters, identify deadlines and draft questions.
- Technology: troubleshoot a device step-by-step, one reversible change at a time.
- Family: generate low-cost activities, study games and age-appropriate explanations.
- Consumer decisions: identify meaningful product specifications and warning signs, then check independent sources.
ChatGPT, Claude or Gemini?
All three are capable general-purpose assistants. Their relative performance varies by task, model version, account tier and personal workflow. There is no defensible permanent “best” model across every use case.
| Assistant | Useful capabilities to explore | Relevant practical workflow | Things to check |
|---|---|---|---|
| ChatGPT Open ChatGPT | Web search, file and data analysis, image creation, voice, projects, customisation and research tools (depending on plan). | Turn a CSV into a chart, practise a conversation by voice, work iteratively on an ongoing project. | Available models, usage limits, privacy settings and tool access. |
| Claude Open Claude | Writing, analysis, coding, project context, shareable Artifacts and research features (plan-dependent). | Critique a long draft, build a small interactive prototype, work through a document. | Artifact behaviour, model limits, connectors and paid research access. |
| Gemini Open Gemini | Multimodal work with documents and other uploaded files; Google-connected workflows and research features where enabled. | Analyse a document, organise information associated with a Google workflow, explore a research question. | Google account permissions, region, plan, supported integrations and privacy controls. |
How to compare them fairly
- Write one actual task with specific success criteria.
- Give the identical input and constraints to each assistant.
- Check accuracy against primary sources and your own data.
- Compare revision effort, usability, privacy and cost—not just first-response polish.
- Repeat with two more different tasks before deciding.
Pricing: Free tiers and paid plans exist, but inclusions and limits change frequently. Confirm actual Australian-dollar prices, taxes, eligibility and renewal terms on ChatGPT pricing, Claude pricing and Google AI plans. Don't subscribe to all three before you have a real use for them.
You can use a single assistant for most everyday tasks. Adding another model makes sense when you need a second independent draft, a different feature or a meaningful second opinion—not because three subscriptions automatically make you more productive.
45 prompts worth keeping
Choose a category, copy a prompt, replace the brackets and add the context you are comfortable sharing. These prompts work across ChatGPT, Claude and Gemini, though available tools vary.
The copy buttons copy only the prompt text. Prompts and filters run locally in your browser; this page does not send your text anywhere.
AI can make you more capable—or more dependent.
Confident errors
Models can invent studies, sources, statistics, legal rules and software functions. Confirm consequential claims through trustworthy primary sources.
False agreement
A chatbot may echo your assumptions. Request counterarguments and independent evidence; do not mistake affirmation for validation.
Outsourced judgment
AI should support decisions, not become the authority on what you should value, believe or do.
Overplanning
Chatting about productivity feels productive. Track finished actions, not prompt counts.
Privacy & confidentiality
Remove identifying details where possible; review data settings before sharing health, financial or client information.
Automation mistakes
Connected tools may have permission to act. Review drafts and consequences before allowing messages, purchases, edits or deletions.
Unhelpful reassurance loops
Repeatedly asking about the same fear can intensify rumination. Set limits and seek human support if concerns persist.
Skill atrophy
Sometimes attempt the task first. Ask for hints or feedback instead of a finished solution.
Privacy checklist
- Check model-training and chat-history controls for each platform. Product policies differ.
- Check whether third-party connectors can access files, email or calendars and what actions they may take.
- Redact names, addresses, account details and personal identifiers whenever possible.
- Do not assume a paid account automatically provides business-grade confidentiality.
- For sensitive work, check retention, access controls, data processing terms and organisational policy.
OpenAI documents user-facing ChatGPT data controls. Review the current Anthropic privacy information and Gemini privacy information separately.
Seven days. Seven real outputs.
This is not a test of how many things AI can do. It's a controlled introduction to using it well. Aim to produce one completed action each day.
Day 1 · Identify friction
List five recurring annoyances. Ask AI to sort them by impact and ease of improvement. Choose one.
Day 2 · Learn something
Spend 20 minutes in tutor mode. Finish with five closed-book questions.
Day 3 · Finish a task
Take one delayed email, application, household task or document from vague intention to completion.
Day 4 · Challenge a belief
Choose an idea you care about. Ask for counterevidence and update your view accordingly.
Day 5 · Make something
Create a one-page document, basic website, meal plan, checklist or learning resource and actually use it.
Day 6 · Run an experiment
Test a modest change to your routine. Record what happened rather than what you hoped would happen.
Day 7 · Review without hype
Identify what saved time, what was wrong, where AI made work harder, and which single workflow deserves repeating.
Not messages sent. Not hours spent chatting. Useful work completed, skills developed, decisions improved and time returned to your life.
Five higher-value AI workflows
1. Create a reusable project brief
Write a clear document describing a project, its purpose, audience, constraints, style and definition of done. Use a project or workspace feature if supported, and regularly correct outdated context.
2. Research with a verification log
Ask for primary sources, publication dates, supporting quotations and a separate column for claims not verified. Open the actual sources and confirm that the cited passage supports the statement.
3. Work on real data
Provide a suitably anonymised spreadsheet or CSV. Ask the assistant to explain the cleaning steps and calculations. Verify totals and sample rows yourself.
4. Make something you can test
Request a simple prototype and a testing checklist. Check edge cases, security and accessibility. Use real users and real-world feedback.
5. Build an iterative review cycle
Keep a simple record of goals, action taken, outcome and next adjustment. Feed back observed results once a week. You can explore reminders and scheduled follow-ups where the product supports them, but verify availability and permissions.
Frequently asked questions
Do I need to pay for AI to benefit?
No. Free versions can handle many writing, learning, planning and explanation tasks. Pay only when a specific feature or usage limit demonstrably blocks useful work.
Which is best: ChatGPT, Claude or Gemini?
There is no universal answer. Evaluate them on a few real tasks with the same inputs, and consider cost, features, reliability, privacy and the amount of correction needed.
Will AI make me lazy or less intelligent?
It can reduce productive effort if you let it do all your thinking. Used as a tutor, critic or practice partner, it can support active learning. The difference is whether you test yourself, verify answers and perform the work.
Can AI replace a therapist, doctor, lawyer or accountant?
No. It can help you prepare questions and understand general concepts. It cannot reliably perform professional assessment or assume responsibility for high-stakes decisions.
What if the answer sounds convincing but is wrong?
Ask for checkable sources and calculations; independently verify them. Confidence and polished prose do not establish accuracy.
How much personal information should I share?
Share the minimum required. Use summaries or redacted examples for sensitive matters, and examine the product's current privacy and retention settings.
What is the best first prompt?
“Here is a recurring problem in my life: [problem]. Ask me up to five useful questions, challenge my assumptions, and help me choose one realistic action I can take today.”
Can I use these prompts with any assistant?
Most are tool-agnostic. Prompts requiring web research, uploaded data, voice, connectors or scheduled actions depend on each product's available capabilities and plan.
Check the tools, not just the claims.
This guide focuses on practical methods, not claims that an AI product will transform your life. Feature descriptions were cross-checked against official product documentation. Plans, prices and capabilities can change, and the guide is not a head-to-head benchmark.
- OpenAI — ChatGPT capabilities overview
- OpenAI — Data controls in ChatGPT
- OpenAI — Plans and pricing
- Anthropic — Claude plans and pricing
- Anthropic — Claude Artifacts
- Anthropic — Projects
- Anthropic — Research feature
- Google — Upload and analyse files in Gemini Apps
- Google — Gemini Apps privacy information
- Google — Google AI plans
- ASIC — Moneysmart
Updated 17 November 2026. This page provides general education, not individual medical, financial or legal advice. Products are independently described; their inclusion is not an endorsement or affiliation.