Turn the knowledge stuck in your head into procedures people can follow.
Every operations person knows the feeling. There is a process only you understand. Onboarding a vendor, closing the month, handling a refund the right way. You keep meaning to document it. You never do, because writing a standard operating procedure from scratch is tedious, and you are busy doing the actual work.
Then someone goes on leave, or leaves entirely, and the knowledge walks out with them. Suddenly the team is reverse-engineering a process from old emails.
This is not a soft problem. Research into workplace knowledge found that 42 percent of the institutional knowledge people rely on to do their jobs sits with one individual and is shared with no colleague, and that knowledge workers lose roughly 5.3 hours a week either waiting on information or recreating something someone else already knew. [1] The person who holds the process pays for being the only source, too: the average employee is now interrupted about every two minutes during core hours, often to answer the same question for the third time. [2]
Here is what AI changes: it removes the blank page. You no longer have to compose a clean, numbered procedure in your head and type it out. You describe how you do the thing, messily, and the AI does the structuring. That single shift is why a task you have avoided for a year can be done before lunch.
A repeatable loop for turning tacit knowledge into a written procedure. The human owns steps 1, 4 and the final sign-off.
Do not try to write a tidy SOP. Try to explain the task to a smart new colleague, out loud or in rough notes. Ramble. Include the “oh, and watch out for” bits. Those asides are usually the most valuable part of any procedure.
If you are on a phone or laptop, use voice. Talk through the process as if you were screen-sharing with someone. Most people explain a process far more completely when they speak it than when they try to write it. Paste the transcript straight in.
A real example of a good brain-dump opening: “Okay so when a refund request comes in, first I check if it’s within 30 days, which I find in Shopify under the order date. If it’s over 30 days I have to loop in Hina. If it’s under, I check whether the item shipped, because if it did we deduct the postage…” Keep going like that until you run out. The mess is fine. The mess is the point.
Once you have your rambling explanation, hand it to ChatGPT with clear instructions. Here is a prompt that works across almost any process:
“Below is me explaining how I do a task, in rough notes. Turn it into a clear standard operating procedure. Use numbered steps in the order they happen. Write each step as an instruction starting with a verb. Where I was vague or skipped something, add a note saying what is unclear so I can fill it in. Keep the language plain enough for a brand-new team member. Do not invent any steps I did not mention. Here are my notes: [paste].”
Two phrases in that prompt are doing the heavy lifting. “Do not invent any steps I did not mention” keeps the AI honest and stops it adding generic filler. “Add a note saying what is unclear” turns the model into a second pair of eyes instead of a stenographer. This pairs well with a broader AI workflow approach if you are documenting several processes at once.
This is the step people skip, and it’s the one that makes the SOP actually good. After you get the first draft, ask the model to challenge it:
“Read this SOP as if you were a new hire on day one with no context. List every point where you would get stuck, confused, or have to ask someone. Be specific.”
You will be surprised. The model flags the assumptions you did not know you were making: a tool the new person does not have access to, a decision rule you never stated, a step that only makes sense if you already know the history. This is where tacit knowledge becomes explicit.
It’s the same instinct behind using ChatGPT for project management: the value is not the AI doing the work, it’s the AI forcing you to make the implicit parts visible.
An SOP is only real if someone other than the author can follow it. So test it before you publish it to the team.
Do not skip this because the draft looks finished. A polished-looking SOP that nobody can actually follow is worse than no SOP, because people trust it and then get stuck.
A procedure that lives in a forgotten doc is useless. Build in the boring metadata that keeps it alive.
Ask ChatGPT to add a header block with those fields and to suggest a short, searchable title. Then paste the finished thing into wherever your team already looks, not a new system nobody will check.
Let’s be honest about the limits, because pretending AI does this perfectly is how teams get burned.
Pick the one process that scares you most if the person who knows it were unavailable tomorrow. That is your highest-value SOP, and it is almost always sitting undocumented in someone’s head.
Block 45 minutes. Brain-dump it using voice, run the structuring prompt, run the gap-finding prompt, and hand the draft to one colleague to test. You will have a real, usable procedure by the end of the session, and you will wonder why you put it off for so long.
Do that once a week and within two months your most fragile knowledge is written down. That is not a documentation project. It’s just a Friday afternoon habit with a very good return.
Start by brain-dumping how you actually do the task, in rough notes or by talking it through with voice input. Then paste it into ChatGPT with a prompt like: “Turn these notes into a numbered standard operating procedure, start each step with a verb, flag anything unclear, and do not invent steps I did not mention.” Review the draft, ask the AI to find gaps a new hire would hit, and pressure-test it with someone who has never done the task.
Only as accurate as what you give it. AI is excellent at structuring and clarifying your input, but it does not know your tools, your edge cases, or your judgment calls unless you tell it. It can also fill gaps with plausible but wrong steps, so always include “do not invent steps” in your prompt and have a human who knows the process verify the final version. For regulated or safety-critical procedures, a qualified person must sign off.
A reliable one: “Below are my rough notes on how I do a task. Turn them into a clear SOP with numbered steps, each starting with a verb, written for a brand-new team member. Where I was vague, add a note flagging what is unclear instead of guessing. Do not add any steps I did not mention.” Then follow up with: “Now read it as a new hire and list every point you would get stuck.”
For a single process you understand well, plan on around 45 minutes to an afternoon, including brain-dumping, structuring, gap-checking, and one round of testing with a colleague. The slow part used to be writing from a blank page; AI removes that, so most of your time goes into the valuable steps: filling gaps and pressure-testing rather than composing prose.
ChatGPT, Claude, and Gemini all handle SOPs well, and the method matters more than the tool. ChatGPT is a strong default and pairs nicely with Custom Instructions for consistent formatting. Claude tends to handle long, detailed brain-dumps cleanly. Gemini is convenient if your procedures live in Google Docs. Use whichever is already in your workflow, because the gains come from the brain-dump-and-pressure-test method, not the brand.
This guide draws on hands-on operations and learning-design work helping non-technical teams document the knowledge stuck in people’s heads, plus Panopto’s Workplace Knowledge and Productivity Report and Microsoft’s 2025 research on workplace interruptions. It is a method piece: the prompts and steps are tested, the statistics are sourced, and the limitations are stated honestly.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.