How to Write a Proposal with AI: Prompt and Review Checklist

AI can prepare a useful proposal draft when you give it a complete brief and make it stop at missing information. You still need to set the scope, price, dates, and terms, then check every line before the client sees it.

A reliable workflow has four stages: collect the facts, ask the AI to identify gaps, generate the draft, and review the result against the source notes. Asking for a “winning proposal” without those inputs usually produces confident but generic copy.

Give the AI a structured brief

Start with information that can be checked:

1. Client and project: correct names, company, project title, and contact details.
2. Goal: the problem or outcome discussed with the client.
3. Deliverables: concrete outputs, quantities, formats, and limits.
4. Exclusions and assumptions: work outside the fee and facts the schedule depends on.
5. Timeline: proposed dates, milestones, client inputs, and review time.
6. Price: currency, line items, tax treatment, deposit, balance, and optional work.
7. Terms: revision rounds, change requests, validity, and approval method.

The structure in the freelance proposal guide is a practical source checklist. AI should organise your facts, not fill gaps with plausible details.

Do not paste client secrets, personal data, credentials, or confidential documents into an AI service until you have checked the service's data terms and your commitments to the client. Remove information the draft does not need.

Use a prompt that exposes missing facts

This prompt creates a draft and makes uncertainty visible. Replace the bracketed section with your notes.

Draft a client proposal from the notes below.

Before drafting:
1. List any missing facts that affect scope, price, dates, tax, or approval.
2. Do not invent names, deliverables, numbers, results, legal terms, or deadlines.
3. Mark unresolved information as [confirm] and keep it out of firm promises.

Use these sections:
- project goal
- deliverables
- exclusions and assumptions
- timeline and client dependencies
- price and payment schedule
- revisions and change requests
- approval and next step

Write in plain English. Keep claims specific to the notes. After the draft,
return a review list containing every name, date, amount, percentage, and link.

[PASTE REDACTED NOTES]

The final review list matters. It lets you compare the details with the brief instead of rereading the proposal only for tone.

AI drafts, you approve

Draft inside the proposal workflow

Use Proposa's assisted draft or connect Codex, Claude Code, Gemini CLI, and other MCP clients. Create a private draft, review totals and terms, then publish only after explicit approval.

Start freeRead the MCP guide

Separate extraction from writing

For a long call transcript or brief, do not ask for the finished proposal in the first instruction. Use two passes.

Pass 1: extract the facts. Ask for a table with the source statement, proposed field, and confidence. The useful fields are goal, deliverables, exclusions, dependencies, dates, price, and open questions.

Pass 2: draft from the approved table. Correct the extracted facts, answer the open questions, and then ask the AI to write only from that version.

This separation makes an error easier to find. If the draft contains “two revision rounds,” you can trace whether the number came from the brief or appeared during generation.

Keep five decisions with the freelancer

AI can format and rephrase. It should not make these commercial decisions for you:

• Scope: only you know what you can deliver and what must stay outside the fee.
• Price: a model does not know your costs, capacity, risk, or negotiation position unless you provide and approve them.
• Schedule: a fluent date is still wrong if it ignores your calendar or a client dependency.
• Terms: copied clauses may be unsuitable for the jurisdiction, client, or type of work.
• Publication: a draft should remain private until you have checked the operative version.

If information is missing, ask the client or write a qualified assumption. Do not turn uncertainty into a precise promise because the sentence sounds complete.

Remove generic AI wording

Generated proposals often describe the freelancer rather than the work. Replace broad claims with details already supported by the brief.

Generic:

I am excited to bring my extensive expertise to this transformative project and deliver a seamless, high-quality solution tailored to your unique needs.

Specific:

I will design and build the five agreed pages, connect the existing contact form to HubSpot, and provide one recorded handover after launch.

The second version names the work. It does not claim enthusiasm, quality, or uniqueness that the proposal cannot prove.

Check for these patterns:

• an opening that repeats the client's request without adding a decision;
• long paragraphs about your commitment or passion;
• deliverables described as activities rather than outputs;
• three adjectives where one fact would work better;
• a confident business result you cannot control;
• a closing that asks for several different next steps.

Keep the vocabulary simple. Repeat the clear name for a deliverable rather than rotating through “solution,” “offering,” and “package.”

Check numbers outside the prose

Do not trust a total because it appears in a polished paragraph. Recalculate line items, discounts, tax, recurring charges, deposit percentages, and balances in the system that will create the final proposal.

For example, a draft may correctly repeat “40% deposit” but pair it with the wrong amount after the project fee changes. Verify both the percentage and the currency value. Apply the same check to dates: a four-week schedule may conflict with the milestones the AI listed.

Keep price and payment terms in one section so the client does not have to reconcile several versions of the total.

Use a controlled publish workflow

If your AI client can work with proposal software, give it an explicit boundary:

1. read the workspace context and find the right client;
2. create a draft only;
3. show totals, recurring fees, warnings, and revision;
4. wait for your approval of the latest version;
5. publish and return the secure URL only after that approval.

Proposa's MCP connection guide documents this flow for Codex, Claude Code, Gemini CLI, OpenCode, and compatible clients. Creating a draft does not create a public link. Publishing requires an explicit confirmation, and Proposa returns the client URL without sending a message on your behalf.

That last detail keeps two decisions separate: making the proposal accessible and choosing how to contact the client.

Review checklist before sending

• Do all names, companies, and contact details match the source?
• Can every deliverable be checked as complete or incomplete?
• Are exclusions and assumptions visible?
• Do dates reflect your calendar and client dependencies?
• Have line items, totals, percentages, tax, and recurring fees been recalculated?
• Are revision limits and change requests clear?
• Have unsupported results and invented proof been removed?
• Does the proposal contain only data the client should receive?
• Is there one approval route?
• Am I reviewing the latest revision?

With Proposa AI proposal software, you can prepare the first draft, edit every field, preview the client view, and publish a secure proposal link after review. The draft stays part of the same workflow that later records opens, messages, and the client's decision.

A draft you can inspect

Use AI without giving up control of the proposal

Bring in the brief, review scope and totals, and publish only the version you approve.

Try Proposa for freeExplore proposal software