How to Write a Client Proposal With AI - and Get the Client's Comments Before It's Final (2026)

By Marcin, founder of Markloop · Updated September 2026

What AI is good at in a proposal, and what stays yours

A proposal is part writing, part judgment. AI is very good at the writing part and has no way to do the judgment part for you.

The biggest risk with an AI-written proposal is that it states things you never told it, and states them confidently: a case study that didn't happen, a "typical 30% uplift", a delivery date that ignores your calendar. The steps below are built to stop that.

Step 1 · Give it a real brief

Paste your call notes or the client's email, and add what only you know:

• The client's problem, in their words - quote the email or your notes from the call.
• Scope - what you'll do, and just as important, what you won't.
• Timeline - real weeks, based on your availability.
• Price - line items, total, payment terms, and whether there are options.
• Proof - one or two real past projects you're happy to mention, or "none".

Then set the rules in the prompt:

Write a client proposal from these notes. Use only facts from my notes - if something is missing, add a clearly marked [TODO] instead of making it up. Sections: summary, what we heard, scope (in and out), timeline, price, next steps. Plain, direct language, no buzzwords.

Step 2 · Ask for one polished HTML page, not a chat answer

A proposal pasted into an email looks like a chat reply. The same content as a designed page looks like a proposal. Both ChatGPT and Claude write good HTML, and one HTML file is easy to share, opens in any browser, and keeps the layout exactly as you made it:

Turn this proposal into one standalone HTML page with inline CSS. Make it look like a polished client document: a clear title, short sections, the scope and price as a table with a total, and our name and date at the top. No external images or scripts.

Why not Google Docs or Word? You can, but pasting AI output into Docs flattens the layout, and a .docx attachment starts the email ping-pong you're about to avoid. HTML is also what AI tools now produce by default when they build something visual - Claude calls it an artifact, ChatGPT can host it as a Site.

Step 3 · Check it yourself before anyone else sees it

This takes about five minutes. Ask the model to help, then read it yourself:

• Numbers: Recalculate every total and subtotal in this proposal and list any that don't add up.
• Invented claims: List every factual claim in this proposal that isn't in my notes. Delete or fix each one.
• TODOs: search for [TODO] and fill them in.
• Names and dates: the client's name, company and dates are the classic AI slip.
• Tone: read the summary out loud. If you wouldn't say it on a call, rewrite it.

Step 4 · Send one link the client can comment on

The draft took ten minutes. Feedback by email can take two weeks, and the client replies "looks good, but can we change the second part?" while you guess which second part. How you share decides what kind of feedback you get back:

With Markloop there are two ways in. No setup: download the HTML file your AI made and upload it to Markloop. Or connect your AI once (Claude, ChatGPT, Cursor or Grok; setup is a one-time step, shown in the app after signup) and ask it:

Publish this proposal to my Markloop project as a new document.

Turn on comments for the share link and send it. The client opens it in any browser, types their name, and comments right on the line they mean: "Can we skip the blog for now?" on the blog row, not somewhere in an email. Colleagues on their side can join the thread. You can also leave your own question as a comment on the timeline - "Which start date works for you: 6 or 20 October?" - and they reply in the thread. Not sure what the client will see? Watch the 2-minute walkthrough.

A proposal carries your pricing, so you can give the share link an expiry date when you create it, or revoke it at any time. Who can see the link, and where the data lives: see the FAQ.

Step 5 · Let your AI apply the comments and publish the next version

When the comments are in, go back to the same chat:

Pull the open comments on the Northwind proposal, apply the changes, recalculate the totals, and publish the next version.

If you'd rather decide first, ask Summarize the client's comments and suggest what to change and decide in chat. Either way, version 2 goes live at the same link the client already has: no new attachment, no new email thread. They refresh and see the new total. Markloop doesn't email them about new versions, so a one-line "v2 is up" note helps. Repeat until you both agree on scope and price. When it's final, "Print / save as PDF" in Markloop gives you a clean PDF of that version for your contract or invoice.

Five more prompts that work well for proposals

• Give me three pricing options for this scope: minimal, recommended and full. Put them side by side in one table.
• Rewrite the summary in three sentences a busy CEO would read.
• Add an "Assumptions" section listing everything the price depends on.
• Add a short "What's not included" list based on the scope.
• Read this as the client's finance lead. What would they push back on?

FAQ

Related: Claude artifacts explained, with 10 examples · How to write a client report with AI · HTML is the new Markdown - reviewing it is the missing half · ChatGPT Sites explained · AI agent artifact sharing compared