How I Write My Best SOPs with AI (Prompts Included)

Generative AI is a total game changer. In many ways, it felt like OpenAI was Prometheus stealing fire from the gods and giving it to man… in the form of a simple chat bot.

Now, AI is omnipresent. Everyone claims to integrate AI into their products, even when it doesn’t really seem to belong. Still, I run into dozens of mid- to late-career professionals who still don’t understand the full potential of these systems. They still use it like an advanced search engine.

The next level professional understands the power of a generative process to help create original documents but stop at the first drafting in one chat. Here, I’m going to reveal how I use multiple chats to draft, revise, improve, and finalize a process document.

Step 1: Define Success

I like to start with my favorite model, depending on the process I’m defining. ChatGPT, Gemini, Claude, and Grok are solid all-purpose models, but each offer strengths in different domains. In this step, the goal is to iteratively define the end state of the SOP. We’re just looking for an outline at this point. Here’s a solid prompt to get started:

You’re a business analyst with 15 years of experience developing standard operating procedures (SOPs). I want to outline an SOP for [the process you’re working on]. I will feed the output of our chat into a new chat to produce a rough draft of the SOP. Include relevant metadata with the outline so the next chat has the right context to produce a complete document. Do not produce the outline until we are completely aligned on what should be included. What do you need to know from me to get started?

This will lead you down a path to drafting the outline. I like to end each Q&A response with another prompt to encourage more questions, like, “what did I miss that would make this a complete specification?”

Once you’re satisfied that you’re aligned on the objective, you can instruct the chat to produce the outline. If you don’t add the boundaries, the chat will jump to a generic answer that may feel complete, but it doesn’t incorporate your unique conditions. This process will also help you think deeply about optimizing the process before documenting it.

Step 2: Produce a Zero Draft

With a complete outline in hand, you can start a new chat with the drafting agent. This is a slightly different skill set so you need to engage a different agent. For this, I like to use a good copywriting model, like ChatGPT, Claude, or Jasper. Here’s the prompt I would use:

You’re a business copywriter with 15 years of experience writing standard operating procedures (SOPs). I pasted an outline below with the most important elements of the SOP I want to draft in this session. Review it and ask any clarifying questions before producing the first draft. [Paste the output from Step 1].

This should get you a solid draft in one pass. You may run through a few Q&A iterations, but most of the heavy lifting should have been done in Step 1.

Step 3: Review and Update the Zero Draft

After you have a good zero draft, you can run through it by yourself to fix anything that was left hanging or jump right in with another agent. I like more analytical models for this step, like Grok or Gemini. With a good prompt, they’ll be more critical than what you may get from the copywriter. Try this one on for size:

You’re a process documentation editor with 15 years of experience deeply analyzing standard operating procedures (SOPs). I’m going to paste my zero draft of an SOP for [the process you’re working on], and I want you to give me a robust analysis on how to improve it to get to a complete first draft. Provide your response in bullet pointed format, citing each paragraph clearly with your recommendation (s). Ask any clarifying questions before giving me your complete analysis.

Often, I’ll be in full agreement with the critique and ask the agent to implement the changes. If needed, you can go point by point to select what works for you. With the full set of notes in hand, you have a few options from here.

Step 4: Produce Draft 1.0

Creating version 1.0 starts with combining the outputs from the prior two steps. You can take multiple paths. By far, the easiest approach is to ask the agent in Step 3 to implement the recommended changes. In this case, I like to instruct the agent to make the changes without fundamentally altering the style and unaffected copy.

The next easiest approach is to take the recommendations from Step 3 as a response to the draft from Step 2. In this case, I like to give a prompt like this:

I ran this by an experienced process documentation editor, who gave the following recommendations. Implement all the recommendations. Only modify the unaffected copy to maintain the appropriate flow and style of the existing document.

Finally, the most complex approach is to start a new chat. I typically only do this if I want to test the output of a different model or to start with no historical context in the same model. In that case, I like to start with a very similar prompt as in Step 2:

You’re a business copywriter with 15 years of experience writing standard operating procedures (SOPs). I pasted an initial draft of an SOP for [the process you’re working on] below with some recommendations from an experienced process documentation editor. Review both texts and ask any clarifying questions before producing the first draft. [Paste the output from Steps 2 and 3].

Optional Step 5: Review and Tighten Draft 1.0

If you still want another set of eyes on the document before finalizing it, repeat the same process from Step 3. You can use the same prompt and iterate accordingly.

I rarely go this deep. Often, the reason for going through multiple steps is to ensure you’re capturing everything in a field that may be a little beyond your expertise. In that case, I’ll also open a new chat to understand the subject matter better while I review the SOP.

Step 6: Finalize Version 1.0

As with any SOP, it’s easy for the document to become worthless without an implementation process. This last step is all about operationalization. For this, I like the following prompt in a new chat:

You’re a business process implementation expert with 15 years of experience deploying processes in [discipline of the new SOP] departments. I’m going to give you an SOP we just produced. I want you to help me design an implementation plan for it. Our team consists of [number of people and their experience levels]. I want you to provide potential risks and mitigants to ensure the process gets adopted with fidelity. Ask me any clarifying questions before you draft the plan. [Paste the final version 1.0]

This step may not be necessary if you already have a good deployment process. I find that the additional support often brings up ideas and risks that you don’t always consider.

Final Thoughts

I like to think of AI chat bots as highly educated analysts across any defined level of experience. With that in mind, it’s important to remember that it’s all “book smarts.” The experience is strictly a persona, and it isn’t actual experience. While this is so obvious it seems to go without saying, I found that this framing helps you craft better prompts.

Each prompt defines the quality of the output. You can instruct it to be an Ivy League professor and get a completely different result than an experienced professional. You also can see that iterating with different agent definitions gives a better result than one long string of chats using the same agent.

This is the core of modern agentic prompting. The next level here is bringing this into a programmatic loop in a tool like n8n.

What SOP will you try first?