Five Ways to Prompt AI, With Examples You Can Copy Today
Tags: prompt engineering, AI for small business, AI tools, productivity, AI workflows
Most people prompt AI the way they would type into a search bar. They ask for the output and hope it lands. That works for simple requests and falls apart on anything that depends on your business, your voice, or your numbers.
The fix is rarely a longer prompt. It is a better matched one. Each prompt type below solves a specific problem, and knowing why it works makes it much easier to pick the right one. Every example can be pasted into any major AI assistant and adapted to your own work.
1. Zero-Shot: Instruction Only
Example prompt:
Write a follow-up email to a customer who asked about delivery times.
Why it works: AI models learned from an enormous amount of text, so common tasks like drafting an email are already familiar territory. A plain instruction is enough to trigger that pattern.
The weakness is guessing. With nothing about your customer or your brand, the model fills each gap with the most average version of the task. That is why zero-shot output tends to read as generic.
Best for: quick, low-stakes drafts you plan to edit anyway.
2. Role, Context, and Constraints
Example prompt:
You are an experienced instructional designer. I am building a 20-minute onboarding module for new warehouse supervisors at a 40-person logistics company. Most of them have never managed people before. Write the learning objectives for the module. Make each objective measurable, avoid corporate jargon, and limit the list to five objectives.
Why it works: The model builds its response from everything in the prompt. Each detail you add narrows the range of likely answers. The role sets the expertise and vocabulary. The context supplies the situation the model cannot know on its own. The constraints mark the edges of the output, so you spend less time cutting what you never wanted.
This is the same briefing you would give a new team member on day one. Nobody expects a new hire to do good work without knowing the audience and the goal.
Best for: work where the audience and purpose change what good looks like.
3. One-Shot and Few-Shot: Show an Example
Example prompt:
Here is an example of how I reply to customer questions:
"Hi Dana, thanks for asking. Your order ships Thursday and should reach you by Monday. Let me know if the timing is a problem and I will see what I can do. Sam"
Write a reply in the same voice to the customer message below.
[paste the customer message here]
Why it works: AI models are strong pattern matchers. A single example shows tone, length, and structure in a way that is hard to describe with words alone. One example is called one-shot. Two or more is few-shot, which gives the model a clearer sense of what stays constant from one sample to the next.
The example you choose matters as much as having one. A short, casual email gives the model a clean signal of your voice. A formal template carries headers, rigid sign-offs, and legal phrasing, and the model tends to copy that scaffolding along with the style you wanted.
Match the example to the job. A casual sample is ideal for a casual reply. When the final output needs a formal register, a casual example will pull the result too loose, so use a formal sample for that task.
Best for: matching your voice or a format you already use.
4. Examples Plus a Fact Checklist
Example prompt:
Use the sample reply above as your guide for tone. Use only the facts below. Standard shipping takes 3 to 5 business days. Returns are accepted within 30 days. We do not ship to PO boxes. Do not mention anything that is not on this list.
Why it works: A sample reply teaches the model how to sound. It does not teach the model what is true. AI generates plausible text, not verified text, so it can sound completely confident while getting a price or a policy wrong.
A fact checklist puts the correct details directly in front of the model, which removes the need to guess. The two pieces do different jobs, so it is worth testing them separately to see which one moves your output more. Reading the draft before it goes out still belongs in the process.
Best for: customer-facing writing where accuracy matters as much as tone.
5. Chain of Thought: Ask for the Reasoning
Example prompt:
I have $6,000 in savings and a credit card balance of $3,500 at 24 percent interest. I can put $500 a month toward either goal. Should I pay down the card first or build savings first? Work through this step by step, show your math, and state your assumptions before you give a recommendation.
Why it works: On problems with several steps, an answer that jumps straight to a conclusion is less reliable. Asking for the steps gives the model room to work through the problem, and each step builds on the one before it.
There is a second benefit that matters to you as the reader. Visible steps let you find the specific assumption that is off instead of trusting a final number. Many people also find that reading the reasoning teaches them the method, not only the answer.
One caution applies. Reasoning can look tidy and still be wrong, so check the assumptions the model states before acting on the result.
Best for: math, comparisons, decisions with trade-offs, and any task where you want to audit the logic.
Choosing the Right One
| What you need | Prompt type |
|---|---|
| A fast draft you will edit | Zero-shot |
| Output shaped to a specific audience | Role, context, and constraints |
| Your voice or a set format | One-shot or few-shot |
| Accurate details in your voice | Example plus fact checklist |
| A decision or calculation you can verify | Chain of thought |
These types also combine well. A strong customer reply prompt might set a role, include one casual sample, and attach a fact list in the same message.
Try This Week
Pick one prompt you already use every week and rebuild it with the type that fits best. Run both versions side by side and compare how much editing each draft needs. That comparison usually shows more than any explanation can.
When you are ready for help building prompts your team will actually use, a free 30-minute discovery call is a no-pressure place to start. You can book one at www.toddstoneconsulting.com.
