Playbook 2
A stack of SOPs
AI Tools Field Guide
Not affiliated with OpenAI, Google, xAI or Anthropic; your organization’s rules come first.
One shape, many documents: fix the instructions once, then run every SOP through them.

Set it up once
- Pick the target shape. Numbered steps, one action per step, a warning before the step it applies to, references in brackets.
- Put the shape in a helper. A Gemini agent, a custom GPT or a Grok project holds it, so SOP forty gets what SOP one got.
- Attach one finished example. The model copies a good example far better than it follows a description of one.
- Want a Word file back? Grok with a skill, and expect minutes. Otherwise paste the text into your own template.
Run each one
- One SOP per chat. Attach it, name the shape, and forbid adding, dropping or reordering a step.
- Ask for a change list. Original step, rewritten step, what changed; you review the list, not two documents.
- Break up a long one. Ask for the section headings, then take one section at a time.
- Count the steps in and out. A dropped step is the dangerous error; warnings stay before their step.
- Numbers carry over unchanged. Torques, times and limits must match the original exactly.
Say it like this
Rewrite the attached SOP in the shape of the attached example: numbered steps, one action per step, warnings before the step they apply to.
Add, drop or reorder nothing. List anything ambiguous. Return a change list: original step, rewritten step, what changed.
Where the original is ambiguous, quote it and write CONFIRM. Keep every warning, number and limit exactly as written.

Take away
Shape once, run many, check by change list. The helper makes them consistent; the department that owns the SOP still reviews it and keeps it correct.