Prompt Writing Is the New SOP Writing: How I’m Learning to Work With AI
- 19 hours ago
- 4 min read

As I work to build my AI skills, I keep thinking back to what happened when email and the internet first started becoming part of everyday business.
Yes, I am old enough to remember both.
At the time, I was one of the younger people in the office. I approached these new tools with excitement. I picked them up quickly and often found myself showing supervisors and managers how to use them and, more importantly, how to use them better.
For about 15 years, I made much of my living by learning how to use the internet and related technology while also working in public accounting.
Today, AI feels very similar to me.
This Is the Worst AI Is Going to Be
Think about the internet in the 1990s compared with what it became.
That is how I am approaching AI.
The technology we are using today is probably the worst version of AI we will ever use. It will get better, faster and easier.
But I believe the people who embrace it now have an opportunity to get ahead.
The question is: How do you get ahead of everyone else who has access to the same AI tools?
I think part of the answer is surprisingly old fashioned.
You need to become very good at explaining exactly what you want done.
Treat AI Like a Smart Intern
I like to think of AI as an intern who knows a little about your profession but doesn't necessarily know your company, your processes or exactly how you want the work performed.
Would you tell an intern:
"Prepare this report."
And walk away?
Probably not.
You would explain the objective, provide the necessary information, walk through the steps, explain what the finished product should look like and review the work afterward.
AI isn't that different.
The quality of the output depends heavily on the quality of the instructions.
And there is another step I think is extremely important:
Ask AI to quality check its own work against your original instructions.
Don't just give it the process.
Give it the process AND the review process.
Prompt Writing Looks a Lot Like SOP Writing
This reminds me of something I have believed in for years: good Standard Operating
Procedures.
I have always been a big believer in documenting processes step by step so someone brand new could pick up the instructions and perform the task.
One of my tests for a good SOP is simple:
Could a reasonably capable high school student follow these instructions and successfully complete the process?
Could that student walk into your warehouse and pick an order correctly?
Could they enter an order into your accounting system?
Could they ship an order?
Could they update a web page using your optimization guidelines?
If they can't, where are the gaps in your instructions?
Have someone new test the process.
Watch where they get confused.
Update the documentation.
Repeat.
That same thinking applies to AI prompts.
Clear Beats Pretty
Being an accountant, my operating manuals were never going to win many design awards.
But they were clear.
They contained the information someone needed to complete the task.
You can always score bonus points by having someone make the final document look better.
I saw this firsthand while working with an RSM client implementing SAP FI/CO. I helped write process documentation and capture screenshots showing users exactly what to do.
And a quick shoutout to Snagit. I have used it extensively for process documentation and still love it.
But the important part wasn't making the documentation beautiful.
It was making it usable.
My AI Prompting Lesson
I recently learned this lesson again while building a detailed golf strategy and yardage book using ChatGPT.
Needless to say, it has been frustrating at times.
I would refine the instructions, get one part working correctly and then discover that something we had previously established disappeared from the next version.
That led me to make two changes.
First, I added a quality control feedback loop to the prompt.
After completing the work, AI needs to go back through the instructions and verify that each requirement was actually followed.
Second, I started managing my important master prompts outside of ChatGPT.
Instead of expecting AI to remember every detail indefinitely, I have AI help me build and improve the prompt. Then I maintain the master version myself.
That creates something very similar to a controlled SOP.
The prompt becomes the operating procedure.
AI becomes the person executing it.
And the quality control loop becomes the review process.
My Simple AI Prompting Framework
When I am building a reusable prompt today, I increasingly think about five things:
Objective: What exactly am I asking AI to accomplish?
Inputs: What information does AI need to complete the task?
Process: What steps should it follow and in what order?
Output: What exactly should the finished product look like?
Quality Control: Before delivering the answer, did AI check its work against every requirement?
Then I ask myself one final question:
Are these instructions clear enough that a high school student could understand what I want accomplished?
If they aren't, the prompt probably isn't finished.
AI may be revolutionary technology.
But one of the skills that may help us get the most out of it isn't revolutionary at all.
It's learning how to create really good instructions.
And accountants, operations professionals and anyone who has spent years building good processes may have more of a head start than they realize.
What are you seeing as you use AI? Do you agree that prompt building is starting to look a lot like SOP development?
If this was useful, save it for the next time you're building a reusable AI prompt, share it with someone experimenting with AI and comment with what you're learning. I would especially like to hear what you've found works and what continues to frustrate you.
AI disclosure: I provided the ideas, experiences, examples and point of view for this article. I used AI to help organize, edit and refine my original draft. The final content was reviewed and approved by me.



Comments