How to use AI to write a simple scope of work document before you start a client project so disputes don't happen later

PMI's Pulse of the Profession{target="_blank"} data puts scope creep at 52% of all projects — and most of those failures trace back to a vague agreement at the start, not bad work during execution. This post walks you through a repeatable AI workflow for drafting a scope of work document from a client email or brief, covering every section that actually matters. If you run this process once and save the prompts, you'll spend 15–20 minutes on documentation that previously cost you 1–3 hours — or much more when a client dispute hit mid-project.

What You Need Before You Start

Claude{target="_blank"} or ChatGPT{target="_blank"} — either drafts professional business documents from raw text inputs; both handle long email threads without hallucinating scope items, provided you review the output. Pricing: Claude's free tier handles this use case, though Claude Pro{target="_blank"} at $20/month (as of early 2026) gives you access to Claude 3.7 Sonnet, which handles longer, messier email threads better. ChatGPT Plus{target="_blank"} at $20/month (as of early 2026) unlocks GPT-4o, which performs comparably. The free tiers of both tools are sufficient for shorter client briefs.

Time required: 15–20 minutes for a single SOW once you have the workflow set up. First-time setup — reading this post, saving your prompts, building your review checklist — adds another 30–45 minutes up front.

Skill level: No technical background required. You need to be able to copy and paste text and edit a document. If you can write an email, you can run this workflow.

---

How to Prepare Your Inputs Before Asking AI to Write Anything

The quality of an AI-generated scope of work document is directly proportional to the quality of what you feed it. Vague inputs produce vague outputs — which defeats the purpose entirely.

Before you open Claude or ChatGPT, collect the following into a single document or text block:

1. Copy the full client email thread or brief — paste the entire thing, including any back-and-forth. Don't summarize it yourself; the AI will do that, and you want it working from the raw language the client used, not your interpretation of it.
2. Add any verbal agreements — if you discussed anything on a call that didn't make it into writing, type a bullet-point summary: "Client also mentioned they want X" or "We agreed on a call that Y was out of scope."
3. Note your standard terms — revision limits, payment schedule, any clauses you use for every project. These are inputs, not things you'll ask the AI to invent.
4. Flag anything you already know is ambiguous — if you read the brief and thought "I'm not sure what they mean by X," write that down. You'll ask the AI to surface more, but your own flags matter too.

That's it. You now have everything the AI needs. Don't clean it up further — messy, real client language is actually more useful here than a polished summary.

---

How to Write a Scope of Work for a Client Project: Step-by-Step

Step 1: Open Claude or ChatGPT in a new conversation.

Step 2: Paste the following prompt, filling in the bracketed fields with your specifics:

You are helping a [type of service business — e.g., freelance graphic designer / web developer / marketing consultant] draft a scope of work document for a new client project.

Here is the raw client brief and/or email thread:

[PASTE CLIENT EMAIL OR BRIEF HERE]

Here are any additional verbal agreements or notes:

[PASTE YOUR NOTES HERE]

First, identify any ambiguous or missing information in this brief that could cause a dispute later — list each item clearly.

Then, draft a complete scope of work document with the following sections:

1. Project Overview
2. Project Objectives
3. Specific Deliverables — include a clear acceptance criterion for each deliverable
4. What Is NOT Included (Exclusions) — be thorough; derive this from ambiguous language in the brief
5. Timeline and Milestones
6. Payment Schedule
7. Revision Policy — limit client revisions to two rounds per deliverable unless otherwise noted
8. Client Responsibilities

Use the client's own language where possible. Flag any section where you've made an assumption rather than relying on stated information.

Step 3: Read the ambiguities list the AI returns before you look at the draft. This is the most valuable output. You'll see gaps you missed — unclear deliverable formats, undefined "final" states, missing approval processes. Answer each one in writing.

Step 4: Paste your answers back into the same conversation:

Here are my answers to the ambiguities you identified: [YOUR ANSWERS]

Now revise the scope of work document incorporating these answers. Keep the same structure.

Step 5: Copy the revised SOW into a Google Doc or Word document. Do not send it yet — go to the review section below first.

The reason the workflow splits into two passes — ambiguities first, then draft — matters more than it might seem. If you skip straight to the draft, the AI fills gaps with plausible-sounding assumptions that look reasonable on first read but won't hold up when the client's definition of "final deliverable" turns out to be different from yours.

---

What to Review Before You Send It — The Four Things AI Consistently Gets Wrong

1. Vague deliverable descriptions. The AI will often write something like "design of homepage." That's not a deliverable — it's a category. Revise to: "One homepage design, delivered as a Figma file with desktop and mobile breakpoints, incorporating up to two rounds of client feedback." Each deliverable needs a format, a quantity, and a completion state.

2. Missing or weak exclusions. The exclusions section is the single most important and most commonly omitted part of a freelance SOW, according to and.co's analysis of freelance disputes{target="_blank"}. Read every deliverable and ask: what adjacent work could the client reasonably assume is included? That adjacent work goes in the exclusions. The AI will generate a list, but it will be conservative — add to it manually.

3. Absent revision limits. Even when you explicitly prompt for a two-round revision limit, the AI sometimes buries it in general language rather than attaching it to each deliverable. Check that the revision policy is specific: "Each deliverable includes two rounds of revisions. Additional revision rounds are billed at $X/hour."

4. Payment terms that don't match your actual policy. The AI will generate plausible-sounding payment terms based on your brief, but it doesn't know your actual payment schedule, your late fee policy, or whether you require a deposit. These must be edited manually. Never send an AI-generated SOW with payment terms you haven't read line by line.

---

When Something Goes Wrong

The AI invents deliverables that weren't in the brief. This is rare with Claude 3.7 Sonnet and GPT-4o when given a full brief, but it can happen with very short inputs. The fix: re-read every deliverable in the draft against your source material. If you can't point to something in the brief or your notes that supports a deliverable, delete it or flag it for client confirmation before sending.

The exclusions section is generic and unhelpful. If the AI produces exclusions like "any work not explicitly mentioned above" without specifics, your input brief was probably too short. Go back to your notes, pull out every adjacent service you can think of, and manually add them. The prompt can also be sharpened: ask explicitly to "derive exclusions from ambiguous phrases in the client's email, citing the specific phrase for each."

The client pushes back that the SOW doesn't match what they understood. This usually means the AI drafted from your notes rather than the client's language. Re-run the prompt with the actual client email pasted in — not a summary — and check that the deliverable language mirrors what the client wrote. Contract Simply's SOW guidance{target="_blank"} makes the point that matching the client's own words reduces disputes about intent more reliably than formal legal language.

---

How to Turn This Into a Repeatable Workflow for Every New Project

Save your prompt as a template in a Google Doc, Notion page, or wherever you keep operating procedures. The only fields that change per project are the client brief, your verbal notes, and your service type context at the top.

After your first three uses, you'll also have a library of exclusions that commonly apply to your type of work — add those as a standing input to the prompt so the AI starts with your standard exclusion list and adds project-specific ones on top.

If you're attaching the SOW to a service agreement, have a lawyer review the template once for your business type and jurisdiction. The AI-generated document gives you strong protection as a practical matter, but a one-time legal review of your standard language is worth the cost. After that, the AI handles the project-specific customization and you handle the final review — which is how this saves 1–3 hours per project without sacrificing accuracy.

Link your SOW template to your client onboarding checklist so it's a required step, not an optional one. For a full walkthrough of building that checklist, see our guide on how to build a client onboarding workflow with AI.

---

FAQ

How is a scope of work different from a contract? A scope of work defines what you're delivering, when, and under what conditions — it describes the work itself. A contract governs the legal relationship: payment enforcement, intellectual property, liability, termination. A SOW doesn't replace a contract, but attaching a well-drafted SOW to even a simple service agreement significantly strengthens your legal position if a dispute goes to small claims or mediation. Most freelancers need both; the SOW is the document the client actually reads and agrees to before work starts.

Can I use the free version of Claude or ChatGPT to write a scope of work? Yes, for most client briefs. The free tiers of both Claude and ChatGPT handle inputs up to several thousand words without issue. Where you'll notice a difference is with very long email threads (10+ messages) or complex multi-phase projects — in those cases, Claude Pro or ChatGPT Plus at $20/month gives you a longer context window and a more capable model. For most small service businesses, the free tier is sufficient.

What if I don't have a client email — just a phone call summary? Type your call notes into a bullet-point list as accurately as you can, including anything the client said that struck you as potentially ambiguous. Paste that in place of the email. The workflow runs the same way; the AI will flag missing information, and you fill the gaps. The output will be slightly less precise than when you have written client language to work from, so pay extra attention to the exclusions and deliverable descriptions in your review.

Will an AI-generated scope of work hold up legally? It carries meaningful weight as a written record of what was agreed — courts and arbitrators take written project documentation seriously. What it isn't is a substitute for a lawyer-drafted contract in high-stakes engagements. For projects under a few thousand dollars, a solid AI-drafted SOW attached to a simple service agreement gives you practical protection. For larger projects, have an attorney review your standard template once; you'll use it repeatedly without needing them again for routine projects.

How much time does writing a scope of work with AI actually save? The honest answer: the research points to 1–3 hours saved per project compared to starting from a generic template or writing from scratch. The floor of that range applies when you have a clean client brief and light deliverables. The ceiling applies when the project is complex and you'd otherwise spend significant time chasing down what was and wasn't agreed to. The more useful number may be this: one avoided mid-project dispute — even a small one — almost certainly exceeds the value of the time saved across a dozen SOWs.