How to write a freelance proposal with AI (without sounding like AI)

Most freelance proposals that get ignored fail for the same reason: they talk about the freelancer instead of the client. AI makes this problem worse if you use it lazily, because the default output is polite, generic and full of phrases every client has already read ten times this month.

Used well, AI does the opposite. It helps you organise what the client told you, structure a clear offer, and check your draft for weak spots. This guide shows you a process that keeps the thinking yours and uses AI for the parts that are slow and mechanical.

Why AI proposals sound like AI

Before fixing the problem, it helps to know what causes it.

• No real input. If you ask for "a proposal for a website redesign", the model has nothing specific to work with, so it fills the gap with generic phrasing.
• No voice sample. Without examples of how you write, you get the model's default voice: formal, balanced, slightly overexcited.
• Features before problems. Models tend to list services and deliverables first, because that is what most proposals online look like.
• Padding. Restating points, three-adjective lists and closing lines like "We look forward to partnering with you on this exciting journey".

Every step below is designed to remove one of these causes.

Step 1: Start from the discovery call, not a blank page

The quality of a proposal is set before you write it. If your discovery call did not uncover the client's problem, goals, constraints and decision process, no prompt will fix that.

If you want a structure for the call itself, use the discovery call question plan. It groups questions by situation, problem, past attempts, success and constraints, and includes follow-ups for vague answers.

After the call, paste your notes or transcript into the call notes to recap email prompt. You get two things:

1. An internal brief with the client's own words in quotes.
2. A recap email you can send the same day, asking the client to correct anything you misunderstood.

That recap email is quietly one of the most useful sales tools you have. The client sees you listened, and any correction they send makes your proposal more accurate.

If the lead came from a marketplace

On Upwork and similar platforms there is often no call before the first proposal. In that case the job post is your brief. The Upwork proposal prompt first asks the model to list what the client actually cares about, then writes a short cover letter whose first two lines answer their problem, because those lines are what shows in the preview.

Step 2: Give the model your voice

Pick two or three things you wrote without overthinking: an email to a client, a LinkedIn post, a message to a colleague. Paste one of them into the proposal prompt as a voice sample.

This single change does more for the result than any instruction about tone. The model copies sentence length, formality and word choice from the sample far more reliably than it follows adjectives like "friendly but professional".

Step 3: Draft the proposal in the client's order

Use the freelance proposal draft prompt with your internal brief, your approach, deliverables, exclusions, timeline, price and voice sample. The prompt forces this order:

1. Where you are now: their situation and problem, in their words.
2. Where you want to be: goals and how success will be measured.
3. How we get there: your approach in phases.
4. Timeline.
5. Investment.
6. Not included.
7. Why me.
8. Next steps.

Notice that price comes after the goals and the plan. The client reads the cost after they have agreed with your description of their problem, not before.

Offer options instead of one number

A single price gives the client two choices: yes or no. Three options give them a way to say yes at a level that fits. The three pricing options prompt builds a small, recommended and complete version of the project, checks each against your minimum rate, and writes an intro that explains the difference in outcomes rather than features.

Make the middle option the one you actually want to deliver.

Step 4: Remove the AI sound

Even with a voice sample, a first draft will have some stock phrasing. Run it through the de-AI edit prompt. It does three things:

1. Lists traits of your real writing from your samples.
2. Finds and quotes the patterns that make text sound generated: stock openers, restated sentences, three-adjective lists, "it is not just X, it is Y" constructions, paragraphs of identical length.
3. Rewrites the draft in your style, cutting at least 20% of the words, while keeping every fact, number and commitment.

Then do your own pass. Read the proposal out loud. Any sentence you would not say to the client on a call, rewrite or delete.

A quick manual checklist

• Does the first paragraph mention the client's problem, not you?
• Is there at least one phrase the client actually said, in quotes or close to it?
• Could this proposal be sent to a different client with only the name changed? If yes, it is not specific enough.
• Is every number real? Did the model add any result, client name or statistic you did not give it?
• Is the "Not included" section clear enough to point to later?

Step 5: Prepare for the reply

Some clients will say yes. Many will reply with a question, a concern or silence.

• For "too expensive", "we need to think about it" or "we got a cheaper quote", use the proposal objection replies prompt. It identifies the real objection, suggests a clarifying question, and writes a reply that offers options without lowering your price unless you remove scope.
• For silence, the follow-up sequence prompt writes three short messages that each add something, and ends with a polite close-the-loop email.

What not to hand over to AI

Some parts of the proposal should stay fully yours:

• The price. AI can check your math and structure options. It does not know your costs, your pipeline or how much you want this project.
• Promises. Timelines and results must come from you. Never let a model write a commitment you did not decide.
• Client data. Check your AI tool's data settings before you paste client documents, and remove anything confidential you do not need.
• The final read. You are the one who will deliver the work. Read the proposal as if you were the client, then send it.

A simple workflow to reuse

1. Run the discovery call with a question plan.
2. Turn notes into an internal brief and a recap email.
3. Draft the proposal from the brief, with a voice sample.
4. Build three pricing options.
5. Run the de-AI edit, then read it out loud.
6. Send, and prepare your objection and follow-up replies.

Save your best proposals and the notes that produced them. After a few projects you will have your own examples to paste in, and the drafts will need less editing each time.