Beginner

Using AI Properly

AI is a powerful tool when used with the right habits. These are the workflows and practices that separate people who get reliable value from AI from those who get unreliable output they cannot trust.

The Verification Loop

The single most important AI habit: before you use or share AI output, verify its factual claims. The verification step differs by output type.

Output type	How to verify	Common failure
Code	Run it — does it execute? Does it handle edge cases? Any security issues?	Code looks correct but fails on real inputs or introduces vulnerabilities
Factual claims	Look up the specific claim in a primary source (official website, published study, authoritative reference)	Statistics that sound plausible but are fabricated; numbers that are close but wrong
Citations and references	Search for the source; check it exists; check it says what AI claims it says	AI fabricates plausible-looking citations with correct journal name but wrong author, year, or content
Maths and calculations	Recalculate independently using a calculator or spreadsheet	Multi-step arithmetic errors; correct-looking formula applied incorrectly
Summaries	Check the summary against the source material; look for omissions and distortions	AI drops important caveats, emphasises minor points, or adds detail not in the source

Reference-First Habits

Before you cite any source that AI mentions, verify it yourself. This is non-negotiable for published work, professional advice, and anything presented as factual to others.

AI citation traps

Fabricated papers with real-sounding titles and plausible authors
Real papers with wrong page numbers, publication years, or conclusions
Real statistics from a misidentified source (wrong study, wrong year)
Paraphrased quotes that subtly change the meaning of what was said

Reference-first workflow

Ask AI to help identify what sources are relevant to your topic
Search for those sources yourself (Google Scholar, official sites)
Read the source — confirm it exists and says what you need
Use the source directly; do not rely on AI's paraphrase of it

Drafting vs Deciding

Use AI as a first-draft tool, not a decision-maker. The distinction matters most for consequential outputs.

AI's role: drafting

Generate options and alternatives for your consideration
Create a first draft that you review, edit, and take ownership of
Structure raw information you provide
Suggest frameworks and approaches for you to evaluate

Your role: deciding

Final judgement on correctness, tone, and appropriateness
All consequential decisions (personnel, financial, legal, medical)
Any output you put your name on or present as your work
Accountability for outcomes — you cannot delegate this to AI

Work-Type Checklists

What "done and checked" means depends on what you made. Use these before publishing, shipping, or sending AI-assisted work.

Code

Does it actually run without errors?
Have you tested it with real inputs, including edge cases?
No security anti-patterns (SQL injection, exposed secrets, unvalidated input)?
Is it readable and maintainable by someone else?
Does it handle the error cases it is supposed to handle?

Documents & reports

Are all factual claims verified against a source?
Are all statistics correctly attributed with the right date and context?
Is the tone appropriate for the audience?
No AI voice tells (see red flags section below)?
Does it actually answer the question it was supposed to answer?

Analysis & research

Are the underlying sources real and correctly cited?
Are claims supported by the sources AI referenced?
Is the methodology (if any) valid for the question being asked?
Are the limitations and caveats stated, not just the conclusions?

Presentations & decks

Is every data point and statistic verified?
Does the narrative actually reflect your position and argument?
No generic filler content that adds length without meaning?
Would you be comfortable defending every claim if questioned?

Effective Prompting Habits

Context first: give AI the context it needs before asking your question — who you are, what you are working on, what you already know
Specify format and length: "in 3 bullet points" or "as a table" or "in 200 words max" gets you a usable output instead of an essay
State constraints explicitly: "do not include X" / "assume the audience already knows Y" / "match this tone: [example]"
Iterate, don't accept: treat the first response as a starting point; follow up to refine, cut, or redirect
Ask for alternatives: "give me three different approaches" reveals options you might not have considered

Knowing When to Stop Using AI

Some situations call for stopping AI-assisted work and using a different approach.

Current information is required: news, live prices, today's status — use real-time sources
Professional expertise is needed: legal advice for your specific situation, medical diagnosis, licensed financial planning — consult a qualified human professional
Real-world action is required: AI cannot make a phone call, submit a form, or take a physical action — do it yourself
Genuine human judgment is essential: empathy, ethics, interpersonal nuance, organisational politics — AI is a poor substitute
You have checked twice and it is still wrong: some tasks are not suitable for current AI; do it yourself rather than fighting the model

Checklist: Do You Understand This?

For each output type (code, factual claims, citations, maths), what is the correct verification step?
What is the reference-first workflow — and why is reading the source yourself essential?
What is AI's role in the drafting-vs-deciding model, and what is your role?
Before shipping AI-assisted code, what five things should you check?
Name three situations where you should stop using AI and take a different approach.
Why is iterating on a first response usually better than accepting it?