You can edit Word documents with AI in three fundamentally different ways: upload the file directly to ChatGPT or Claude, install a Word add-in like Spellbook or Microsoft 365 Copilot, or connect an MCP tool like Scaffold to the AI you already use. Each approach has a different cost, setup requirement, and output format — and only one of them produces real Word tracked changes without an admin install or desktop subscription.
This article breaks down each option so you can choose the right tool for your workflow.
What Does "Editing a Word Document with AI" Actually Mean?
It helps to be precise. There are three distinct things people mean when they say they want AI to "edit" a Word document:
- Reading and summarizing — the AI reads the document, flags issues, or answers questions. The document itself is not modified.
- Suggesting changes in text — the AI proposes rewrites or deletions, but the output is plain text you copy-paste back into Word yourself.
- Producing a redlined .docx — the AI's proposed changes are embedded in the document as tracked changes (Word's native insertion/deletion markup), so a reviewer can accept or reject each change in the Review pane.
The first two are widely available. The third — real tracked changes in a .docx file — is the one that matters for professional document workflows, and it requires the right tooling.
Approach 1: Native File Upload to ChatGPT or Claude
Both ChatGPT and Claude let you upload a .docx file directly to the chat window. The AI reads the document and can answer questions, summarize it, identify risky clauses, or propose revised language. This works well and requires no setup.
What it supports: .docx, .pdf, and plain text files. Most standard Word documents upload and parse correctly.
What it does not do: It does not produce a new .docx with tracked changes. The AI's output is always text — a revised paragraph, a list of suggested changes, or a rewritten clause shown in the chat window. To apply those changes, you open Word and make them by hand.
Cost: Free on ChatGPT and Claude's free tiers, with usage limits. More reliable on paid plans (ChatGPT Plus, Claude Pro).
Best for: One-off tasks where you need to read, summarize, or get AI feedback on a document. Not suited for professional redline workflows where tracked changes and an audit trail are required.
Approach 2: Word Add-Ins (Spellbook, Microsoft 365 Copilot, Others)
Word add-ins run inside the Word desktop application and give AI capabilities a direct connection to the document's content. Spellbook is the best-known legal AI add-in; Microsoft 365 Copilot is the enterprise option built into the M365 suite.
What they support: Because add-ins run inside Word, they can read the full document structure — tables, styles, headers — and write changes back as tracked changes. This is the most reliable way to get AI-generated tracked changes, because the add-in has direct API access to the Word document object model.
What they require: A Word desktop installation (not Word Online), and typically an IT-managed add-in deployment or admin permission to install from the Office Add-ins store. Copilot requires a Microsoft 365 subscription with the Copilot add-on ($30/user/month on top of M365). Spellbook starts at over $100/month and is designed specifically for legal teams.
Cost: $100–$200+/month depending on the product. M365 Copilot requires an existing M365 plan plus the Copilot license.
Best for: Teams already inside the Microsoft 365 ecosystem who have IT support for add-in deployment, and legal organizations with budget for specialized legal AI tools. Not practical for individual professionals or small teams without M365 infrastructure.
Approach 3: MCP Connectors Like Scaffold
The Model Context Protocol (MCP) is an open standard that lets AI models call external tools during a conversation. Instead of adding a feature to Word, an MCP connector adds Word document capabilities to the AI. This is a meaningful difference: the AI you already use (Claude, ChatGPT, Copilot, Gemini) gains the ability to read, redline, and template Word documents without any changes to Word itself.
The Scaffold MCP connector is built specifically for this. You connect it to your AI client once, and from that point on you can ask your AI to read a document, apply tracked-change redlines, fill in templates, and download a finished .docx — all within the same conversation window.
What it supports: .docx files. Scaffold reads the document's full content, applies changes as proper Word tracked changes (<w:ins> and <w:del> markup), and returns a downloadable file that opens in Word's Review mode.
What it requires: A Scaffold account and a one-time MCP connector setup in your AI client. No Word desktop required. No IT admin. Works in any browser.
Cost: Free 7-day trial, then $29/month (Pro) or $29/user/month (Team).
Best for: Individual professionals — attorneys, HR specialists, consultants, ops managers — who want AI-assisted redlining without a desktop install, IT deployment, or $100+/month subscription.
Side-by-Side Comparison
AI Word Editing Options at a Glance
Native file upload to ChatGPT or Claude costs nothing extra, requires no setup, and accepts .docx files. However, it produces text output only — no tracked changes, no downloadable .docx. Best for reading, summarizing, and getting feedback when you do not need an audit trail. Word add-ins like Spellbook or Copilot produce real tracked changes inside Word, but require Word desktop, IT admin deployment, and budgets starting at $100/month — designed for enterprise legal and M365 teams. MCP connectors like Scaffold produce real tracked changes as a downloadable .docx with no desktop install, work inside the AI client you already use, and start at $29/month. They are the practical middle option for individual professionals who need professional-grade output without enterprise overhead.
Which Approach Is Right for You?
Use native upload when you need a quick read on a contract or document and will apply any changes manually. It is free and requires nothing extra.
Use a Word add-in if your organization already runs Microsoft 365 with Copilot licenses, or you are part of a legal team with an existing Spellbook subscription. The add-in path makes sense when IT infrastructure is already in place.
Use Scaffold MCP if you need real tracked changes without a $100+/month add-in or a desktop Word installation — especially if you already work in Claude or ChatGPT. Setup takes a few minutes, and the output is a properly formatted .docx redline you can send to clients or counterparties exactly as you would any manually-marked document.
Frequently Asked Questions
Does Scaffold work with both Claude and ChatGPT?
Yes. The Scaffold MCP connector is compatible with any AI client that supports the Model Context Protocol — currently Claude (claude.ai web and Claude Desktop) and ChatGPT Plus and Team. You connect Scaffold once in your AI client's settings and it is available in every subsequent conversation.
Do I need Word installed to use Scaffold?
No. Scaffold is fully web-based. You upload your document through the AI conversation, Scaffold processes it server-side, and you download the finished .docx to open in Word, Google Docs, or LibreOffice.
Can Scaffold fill in document templates, not just redline existing documents?
Yes. The Scaffold MCP connector also supports document templates — define variable fields in a Word template and ask the AI to populate them from your inputs. Useful for generating first drafts of contracts, offer letters, or any document with a standard structure and variable facts.
If native upload is not enough for your workflow, start a free 7-day Scaffold trial and see how AI-assisted redlining works inside the AI client you already use.