Most people use an SOP generator to write down a process they already understand and want to hand off. That is fine, and it works. But the far more valuable use of the tool — which I stumbled into entirely by accident on a slow Thursday — is to write down the process you only THINK you understand, and let the structure quietly show you where it falls apart. The decision branches are where all the bodies are buried, and a good SOP prompt is essentially a metal detector for them.
Why messy input beats clean input
The first time I ran the SOP Generator from a Rough Process Description prompt, I made the rookie mistake of tidying up my description before pasting it in. I smoothed the edges, removed the contradictions, and made it sound like we had everything figured out. The SOP that came back was clean, professional, and completely useless, because all it did was reflect my own tidy fiction back at me in a nicer font. It documented the process I wished we had, not the one we actually ran.
The second time, I pasted the genuinely messy version, contradictions and all: 'then someone checks the contract, usually legal but sometimes the AE if legal is slammed, and if it is over fifty grand it goes to the VP except last quarter we skipped that twice because of the deadline.' Read that sentence again. That is not a process. That is a confession. And Claude Opus 4.8 took that confession and did something I did not expect — instead of smoothing it over, it surfaced the contradiction directly.
The audit that wrote itself
In the Procedure section, Opus produced an honest decision branch — 'If contract value exceeds the approval threshold, route to VP for sign-off; if the VP is unavailable, escalate per the documented fallback policy.' Then, because my description was genuinely self-contradictory, the prompt's clarifying-questions behaviour kicked in and asked the two questions I had no answer for: who exactly is the fallback approver when legal is slammed, and what is the real dollar threshold, given that we apparently skipped it twice? Those were the precise two things our team had never actually agreed on. The model did not paper over the gap. It pointed a spotlight at it and waited.
- Feed it the real, contradictory description — not the sanitised version you would put in a board deck
- Read the Decision Branch lines first, before anything else; that is where ambiguity hides in plain sight
- Treat every clarifying question it returns as a confirmed process gap, not as a failure of the model
- Run it deliberately on any process where two reasonable people might answer 'how do we do X here' differently
What we actually found
Across about a dozen processes that I suspected were inconsistent, the pattern repeated almost every time. The model would surface a clean decision branch, the branch would have a hole in it, and the hole turned out to be the exact thing two people on the team did differently without either of them ever realising they disagreed. We found four genuine landmines: an approval threshold nobody could agree on, a customer hand-off that two teams each assumed the other one owned, a refund step that quietly skipped a verification check about half the time, and a reporting cadence that existed only inside one person's personal calendar and would have vanished the day they left.
Takeaway
One practical warning before you try this: be ready for it to be a little uncomfortable. Running the messy version in front of the team that owns the process means publicly surfacing the fact that the process has holes, and people can get defensive about work they thought was solid. I learned to frame it gently — 'the tool found a branch we never decided' lands a lot better than 'turns out you have been doing this wrong.' The model is not accusing anyone; it is just refusing to guess. Keep that framing and the room stays collaborative instead of going quiet.
A good SOP tool does not just document your reality — it interrogates it, politely but relentlessly, until the contradictions surface. So use the messy input on purpose, the uglier the better. When the gaps come up, I take them straight into my Decision Doc: Options, Risk, and a Clear Recommendation prompt to actually resolve them instead of just admiring the problem. Grab the SOP Generator from a Rough Process Description prompt on Prompt Dock and run it on the one process you are quietly a little worried about. It will find the worry for you, and naming the worry is most of the fix.