Advanced techniques

Once you have the basics down, these techniques let you handle more complex tasks and get consistently better results.

Chain-of-thought prompting

Adding “Think through this step by step” to your prompt forces the AI to show its reasoning instead of jumping to a conclusion. This is especially useful for analysis, calculations, and decision-making.

Analyze whether I should hire a full-time bookkeeper or continue outsourcing. My revenue is $1.4M CAD, I spend $2,000 CAD/month on outsourced bookkeeping, and I process about 300 transactions per month. Think through this step by step before giving a recommendation.

The AI will lay out the logic - cost comparison, workload analysis, hidden costs - before reaching a conclusion. You can spot where it went wrong and correct it.

Few-shot examples

Instead of describing what you want, show 2-3 examples of the finished product. The AI will match the pattern.

Write product descriptions for my items. Here are two examples of the style I want:

“Oak Dining Table - Solid oak, seats 6. Built to last 20 years. $2,000 CAD.”

“Leather Armchair - Full-grain leather, walnut frame. Gets better with age. $1,450 CAD.”

Now write descriptions for: Standing Desk, Bookshelf, Coffee Table.

This is faster and more reliable than trying to explain your style in words.

System prompts

A system prompt is a persistent instruction that applies to an entire conversation. Think of it as a standing brief. Many AI tools let you set these in settings or at the start of a chat.

You are a business advisor for Canadian SMEs. Always consider Canadian tax law and regulations. Use plain English. When you reference numbers, use CAD. Keep all responses under 300 words unless I ask for more detail.

Set this once, and every message in that conversation follows these rules.

Structured output

When you need data you can actually use, ask for a specific format.

List my top 5 expenses from this bank statement as a table with columns: Category, Monthly Amount, Percentage of Total Revenue, Suggested Action.

You can also request JSON, CSV, or bullet-point formats. The key is being explicit about the structure you need.

Multi-step prompting

For complex tasks, break the work into sequential steps rather than asking for everything at once.

• Step 1: “Summarize this 40-page contract in 10 bullet points”
• Step 2: “Now identify the 3 clauses that carry the most financial risk”
• Step 3: “Draft an email to my solicitor asking about those 3 clauses”

Each step builds on the last. You review and correct between steps, which keeps errors from compounding.

When to use each technique

Check your understanding

Next steps

Techniques are tool-agnostic, but different AI tools have different strengths. Learn which tool to reach for in Tools and Models.