For years, I was the bottleneck for every new client proposal at my agency.

The process lived in my head. Kickoff call, scope summary, Figma mockup, follow-up email. All of it scattered across Google Docs, meeting notes, and Slack threads. Every new project felt like I was rebuilding the wheel. And every time I tried to write the whole thing down, I lost two hours and ended up with half a document.

Then I tried something different. I stopped trying to write the SOP and started talking through it instead. I narrated the process into a voice recorder, pasted the transcript into Claude, and asked it to clean it up.

Twenty minutes later, I had the SOP I’d been meaning to write for six months.

That’s what this guide is about. A simple method, narrate and structure, for turning a process you already know into a finished SOP without burning a whole afternoon. You bring the knowledge. AI handles the structure.

Why you probably don’t have SOPs yet (and why that’s about to change)

Here’s the truth. No one skips SOPs because they’re lazy. They’re skipped because writing them is boring work that doesn’t feel urgent.

Documenting a process means remembering every step, putting them in order, explaining the why behind each one, and formatting the whole thing for a future teammate who doesn’t exist yet. That’s a lot of thinking for a task that produces zero revenue today. So it slides. Week after week.

But the knowledge isn’t the problem. You have the knowledge; you run the process every week. The problem is the writing. And that’s the part AI is weirdly good at.

You do the part only you can do, which is explain what you do and why. AI does the part you hate, which is turning your words into clean, structured prose. That’s the deal, and it’s a good one.

The “narrate and structure” method

Four steps. Nothing clever.

Pick one process. Not your whole business. One process. “How we onboard a new client.” “How we process an invoice.” “How we prep a proposal.” Start small. You can always write more tomorrow.
Narrate it. Speak it out loud into your phone, use a tool like Granola that transcribes as you talk, or even directly into Claude. If speaking feels weird, type rough notes instead. Whatever is fastest. Don’t edit; just capture what you do and why you do it.
Paste it into AI. Drop your notes or transcript into ChatGPT or Claude with the prompt below. Ask for a structured SOP.
Review and refine. Check for missing steps, wrong assumptions, and unclear language. Fix them. Save the final version.

That’s it. The method works because you skip the hardest part of writing, which isn’t knowing the content. It’s turning thoughts into structure.

The SOP prompt that gets it right the first time

Here’s the prompt. Copy it, fill in the brackets, paste your notes at the bottom, and hit enter.

My goal is to write a Standard Operating Procedure for my business.

Process name: [e.g., "Onboard a new client"]
Audience: [e.g., "A new team member with 0-3 months experience"]
Detail level: [choose: overview / standard / detailed]
Format: numbered steps with a short "why" under each step, plus a "things to watch for" section at the end

Here are my rough notes on how I do this process today:

[paste your notes or transcript]

Turn this into a clean SOP. Use my language and examples. Don't invent steps I didn't describe. If something is unclear or missing, ask me before making it up.

Three things make this prompt work.

The audience line tells AI how much to explain; a new hire needs more context than a senior teammate, and “standard” gives you middle ground. The detail level dials the tone up or down. And those last two sentences, “don’t invent steps” and “ask me before making it up,” are the guardrails. Without them, AI will fill gaps with fiction. With them, it flags the gaps and asks you instead.

Small prompt. Big difference.

Real example: a client proposal SOP

Back to my proposal problem.

My rough notes, typed in about two minutes, looked like this:

Kickoff call happens in Granola. I grab the transcript afterward. Summarize scope of work from the transcript, send it to the client for review. Once scope is approved, I build the proposal in Figma using the template. Draft a follow-up email with the proposal link. Send it.

Seven sentences. That’s all I gave it.

I pasted that into Claude with the prompt above. Here’s what came back:

SOP: Prepare a Client Proposal Audience: Any team member onboarding a new client Detail level: standard

Run the kickoff call in Granola. Why: Granola auto-records and transcribes, so you don’t have to take notes during the call.

Export the transcript after the call. Why: The transcript is the source of truth for scope.

Summarize the scope of work from the transcript. Pull out deliverables, timeline, and budget. Why: The client needs to confirm scope before you spend time on a proposal.

Send the scope summary to the client. Ask them to confirm or correct it. Why: Catching misalignment here saves hours later.

Build the proposal in Figma using the client-proposal template. Why: Template keeps branding and structure consistent.

Draft a follow-up email with the proposal link. Keep it under 150 words. Why: Short emails get replies faster.

Send the email.

Things to watch for:

If the client doesn’t reply to the scope summary within 48 hours, follow up before building the proposal.

If the scope changes after Figma work starts, update the summary first, then the proposal.

Fifteen minutes, start to finish. The thing I’d been meaning to write for six months was done before my coffee got cold. No joke.

How to review your AI-generated SOP

Don’t publish the first output. It’s a draft. A good draft, but a draft.

Check four things.

Missing steps. Did AI skip anything you do but forgot to mention? Steps that feel obvious to you don’t feel obvious to a new hire.
Assumed knowledge. Does the SOP use terms a new hire wouldn’t know? Replace them with plain language, or define them the first time they appear.
Edge cases. What happens if the client doesn’t reply? If the file is missing? If the software is down? Add a “things to watch for” section and list the big ones.
Ownership. Who runs this? Add the owner’s name or role at the top. An orphan SOP goes stale fast.

Then, and this is the step most people skip, run the SOP once yourself, following it exactly as written. You’ll find gaps in five minutes that you’d miss in an hour of review. Real feet, real steps, real gaps.

From SOP to skill to workflow

Here’s where this gets really fun.

A written SOP is a great start. But a written SOP still needs a human to read it and do the work. The next rung up the ladder is a skill. An SOP that AI can run for you.

Back to the proposal story. After I wrote that SOP, I turned each step into a small AI skill:

A skill for pulling the Granola transcript and summarizing the scope of work
A skill for prepping the Figma proposal from the approved scope
A skill for drafting the follow-up email

Then I wrote a meta skill, /prep-client-proposal, that runs all three in the right order.

The SOP told me what to do. The skill does it for me. The workflow does the whole thing while I’m on a different call.

That’s the ladder I want you to see: simple prompt → reusable skill → multi-skill workflow. Start with the SOP. Find one step a computer could do. Turn that step into a skill. Stack skills into workflows.

But don’t skip the SOP. The workflow is only as good as the process underneath it. Garbage in, garbage out still applies, and AI doesn’t change that.

FAQ

How detailed should an SOP be? Detailed enough for a new hire to run the process without asking you questions. No more than that. If you’re writing an SOP just for yourself, keep it short; you already know the context.

How often should I update SOPs? Review every 90 days. Update immediately when the process changes. Stale SOPs are worse than no SOPs because they teach the wrong thing.

Can AI write SOPs for complex processes? Yes, if you narrate the process in enough detail. Break complex processes into smaller SOPs; one SOP per decision point is a good rule.

What if my process has a lot of judgment calls? Add a “things to watch for” section. List the judgment calls and what to consider for each one. AI can’t make the call, but it can flag where a call needs to be made.

Do I need a dedicated SOP tool? No. ChatGPT or Claude plus Google Docs works for most small businesses. Add a dedicated tool when you hit ten SOPs and need search, versioning, or screen recording. Before that, you’re buying features you won’t use.

How to Use AI to Write SOPs for Your Business