Tech Tips
[Not yet joined the group? Join now]
AI Classics for CAs – Essential Habits & Time-Savers
| Shortcut / Action | What it does | Classic |
|---|---|---|
| Give AI context, not just instructions | Tell AI who you are, what you need, and how you want the answer | #01 |
| Interrogate, don’t just summarise | Ask about obligations, assumptions, risks and missing points | #02 |
| Make AI attack your conclusion | Ask it to challenge weak assumptions and alternative interpretations | #03 |
| Ask AI to expose uncertainty | List assumptions, uncertainties and points that need verification | #04 |
| Use AI for structure, keep the judgment | AI drafts & outlines; you own the final interpretation and advice | #05 |
| Verify convincing answers | Check source, time, facts and applicability before relying on it | #06 |
| Use AI as a research assistant | Ask it to research, compare sources and prepare structured reports | #07 |
| Know what the AI can access | Check permissions, connected apps and data the tool can see | #08 |
| Don’t blindly copy ready-made prompts | Use them as ideas; frame the request yourself with your context | #09 |
| Be aware of AI’s context about you | AI may remember role, preferences or past instructions | #10 |
Give AI context, not just instructions.
Instead of:
“Draft a reply.”
Try:
“I am a practising CA specialising in GST. Draft a professional reply declining a client's request for a deadline extension. Keep it polite and firm.”
Notice the difference:
You are giving AI your context, rather than simply telling AI to act as a CA or GST expert.
A useful AI request usually tells it:
• Who you are / the relevant context
• What you want it to do
• How you want the answer presented
Don't make AI guess what you mean.
Don't just ask AI to summarise.
Instead of:
“Summarise this notification.”
Ask:
• What obligations arise from it?
• What assumptions does it make?
• What could create financial or compliance risk?
• What is missing?
Don't just ask AI to compress information. Ask it to interrogate the information.
Make AI your second reviewer.
Don't use AI only to support your conclusion. Ask it to attack it.
Try:
“Act as a strict reviewer. Challenge my conclusion. Identify weak assumptions, missing risks and alternative interpretations.”
Useful for:
• GST classification
• Legal interpretations
• Notice replies
• Internal memos
• Professional opinions
AI can be more useful when you ask it to disagree with you.
Ask AI to expose uncertainty.
AI can sound certain even when the answer is uncertain.
Ask:
“List the assumptions, uncertainties and points in this answer that need verification.”
Then verify them yourself.
For professional work, check:
• Section / provision
• Date / AY / FY
• Thresholds / limits
• Applicability to your facts
• Primary source
Confidence is not evidence.
Use AI for structure. Keep the judgment.
Use AI for:
• Breaking down complex concepts
• Creating checklists
• Drafting an outline
• Analysing source material
• Improving wording
But keep these with you:
• Final interpretation
• Compliance position
• Client advice
• Professional judgment
A useful rule:
If you wouldn't sign it directly, don't send it directly.
AI can accelerate the work. It cannot take responsibility for it.
AI can be wrong in a very convincing way.
An answer can look perfect and still be wrong.
Before relying on it, check:
Source → Can you trace the answer to a reliable source?
Time → Is it current for the relevant FY/AY?
Facts → Do the facts actually match?
Applicability → Does the provision actually apply?
The more convincing the answer looks, the more important the verification becomes.
Don't use AI as just a search box.
AI can now do much more than answer a question.
Give it a problem and ask it to:
• Research the issue
• Compare multiple sources
• Analyse documents
• Identify differences
• Prepare a structured report
For example:
“Compare the current position under these three notifications. Identify the changes, dates, applicability and points that need verification.”
AI can be a research assistant — not just a question-answering machine.
Know what your AI tool can actually access.
AI tools can increasingly work with:
• Uploaded files
• Web pages
• Connected apps
• Emails / documents
• Other tools and services
But these capabilities depend on the tool, its settings and the permissions you give it.
So before using AI for professional work, ask:
What data can this tool access?
Where can that data go?
What have I actually given the tool permission to see?
AI safety isn't only about what you type into the prompt. It's also about what the AI can access.
Don't blindly use ready-made prompts.
You will find plenty of:
“10 prompts every CA must use”
or
“Copy this exact prompt for better AI answers.”
Use them as ideas, not as magic formulas.
A prompt that works for someone else's:
• Facts
• Objective
• AI tool
• Expected output
may not work for yours.
Understand the task. Give AI the relevant context. Then frame the request yourself.
The skill isn't finding the perfect prompt. It's knowing what to ask AI — and why.
The same prompt can give different answers to different people.
Why?
AI may have context about you beyond what you typed in that particular chat.
Context is the background information AI has about you or your task that helps it interpret your prompt.
For example, it may know your:
• Role / profession
• Preferences
• Previous instructions
• Past conversations or saved information
So two people can type exactly the same prompt and still receive different responses.
Don't assume AI sees only the words in your latest prompt.
Know what context your AI has about you — and use or correct it when necessary.
