How to Write a Business Proposal With ChatGPT, Step by Step

The discovery call went well, the client said "send me a proposal," and now you have a blank document and a deadline. If you are working out how to write a business proposal with ChatGPT, the tempting move is a single prompt: "Write a proposal for an onboarding redesign for a mid-sized company." What comes back has every section a proposal should have and nothing in it that this client said.

We use AI for proposals in a different order. The model interviews us first, then drafts from a brief we have approved, then reads the result as the buyer. Price, scope boundaries, and proof stay in human hands. Below are the three prompts and what to check after each one.

What a proposal is for, and where AI fits

A proposal is a decision document. The person approving it wants to see that you understood the problem, what exactly they get, what it costs, and what happens next. Most of that information sits in your head and your call notes, where the model can't reach it. That is why a one-line prompt returns a brochure.

Here is how we split the work.

Section	What the model does well	What only you can supply
Their situation	Organizes call notes, keeps the client's own phrases	What was said, and what was only implied
Approach	Turns your method into plain steps	The method, and why it fits this client
Deliverables	Formats a table, flags vague items	What is in scope and what is not
Timeline	Lays out milestones and dependencies	Your real capacity
Price	Presents options clearly	Every number
Proof	Trims it and places it	Results you can document
Next step	Reduces it to one action	The date

Step 1: Let the model interview you

Paste your raw call notes, then run this before anything gets drafted.

I need to write a proposal for a prospective client. Before you write
anything, interview me.

Ask one question at a time and wait for my answer. Cover, in order:
1. The client's situation in their own words (my call notes are below)
2. What happens if they do nothing for six months
3. Who signs off, and what that person cares about
4. What we will deliver, and what is explicitly out of scope
5. Timeline, what we need from their side, and how the price is built
6. Proof we can cite truthfully
7. What might make them say no

If an answer is vague, push back once with a sharper question.
Stop after 10 questions at most. Then give me a one-page brief with
these headings: Situation, Cost of inaction, Decision-maker, Scope,
Out of scope, Timeline, Price, Proof, Risks, Open questions.
Mark anything I did not actually tell you as OPEN. Do not fill gaps
with plausible guesses.

CALL NOTES:
[paste notes]

One question at a time matters. A list of ten questions gets ten thin answers typed in a hurry. A single question, followed by a sharper one when you dodge, gets the detail you would have left out.

Take a typical task for a two-person HR consultancy. After a call with the operations director of a 120-person company, the notes say: new hires take about three months before they stop asking basic questions, every manager onboards in their own way, and two people left within their first 60 days this year. The interview tends to produce questions like these:

• "Did the director put a cost or a consequence on the slow start, or is that your reading of it?"
• "Who approves the budget: the director, or someone in finance who was not on the call?"
• "You wrote 'redesign onboarding.' Does that include training the managers, or only the materials and the schedule?"

The third question is worth the whole exercise. Scope disputes begin exactly there, and most of us would not have written "manager training is out of scope" without being asked.

Step 2: Draft from the brief

Read the brief, correct it, and answer or accept each OPEN item. Then draft.

Using only the brief above, draft the proposal.

Structure:
1. Your situation: their problem in their words. Quote phrases from
   the call notes where they fit.
2. What we propose: the approach in 3 to 5 plain steps.
3. Deliverables: a table with columns deliverable, what it includes,
   how you'll know it's done.
4. Timeline: a table with columns week, milestone, what we need
   from you.
5. Investment: leave every figure as [PRICE]. I will fill them in.
6. Why us: only the proof listed in the brief.
7. Next step: one action and one date.

Rules:
- Two pages at most. Short paragraphs. No adjectives about us.
- Do not invent clients, results, statistics, or credentials.
- Where the brief says OPEN, write [OPEN: the question] instead of
  guessing.
- Write a summary of 5 sentences at most after everything else is
  drafted, and place it at the top.

The difference shows in the first section. From the one-line prompt, "Your situation" opens like this:

In a competitive talent market, effective onboarding is essential to employee retention and productivity.

From the brief, it opens like this:

New hires take about three months before they stop asking basic questions, each manager onboards in their own way, and two people left within their first 60 days this year. You told us the goal is to shorten that first stretch and make it the same in every team.

The second version could only have been written for this client, and that is what a buyer checks first.

What to fix by hand at this point: compare every figure with your notes. Models tidy numbers, so "about three months" may come back as "90 days on average," which nobody said. Fill in each [PRICE] yourself. Resolve every [OPEN] marker by asking the client, or by stating your assumption in the proposal. If a section misses, a targeted follow-up prompt repairs it faster than redrafting the whole document.

Step 3: Read it as the person who signs

The reader who matters often wasn't on the call. Ask for that reading in a new chat, with only the proposal pasted in, so the model can't lean on context the real reader won't have.

Read this proposal as the person who approves the budget and has two
other proposals on the desk. You were not on the discovery call.

1. After the first paragraph, do you know what you are buying and
   what it costs you to wait? If not, what is missing?
2. List every sentence that could appear in any vendor's proposal.
3. Where is the scope ambiguous enough to cause a dispute later?
4. What would you ask before signing?

Do not rewrite anything. Give me the list, ordered by how likely
each point is to cost us the deal.

PROPOSAL:
[paste proposal]

We ask for a list because we want to decide what changes. Point 2 usually returns the "Why us" section almost whole, which tells you where to add a documented result or cut. If the proposal has to be presented rather than sent, the same structure-first approach applies to slides: How to Make a Presentation With AI: Structure First, Slides Last.

Before you send: a ten-minute check

• Every number traces back to your notes or your price sheet, and the totals are recalculated by hand.
• Every claim about you is one you can document if asked.
• "Out of scope" is written down, in plain words.
• The first paragraph uses the client's words for the problem.
• There is one next step with a date.
• Stock phrases are gone. Our list of words that sound like AI speeds up this pass.

Where this breaks

• Invented proof. Ask for a "Why us" section without supplying facts and you get case studies that never happened. The "only the proof listed in the brief" rule exists for this reason, and it still needs a human check.
• Arithmetic. Models slip on totals, discounts, and taxes inside tables. Treat every calculation as a draft and recompute it.
• Confidential material. Call notes often contain names, internal figures, or details covered by an NDA. As of October 2026, OpenAI may use conversations on the consumer plans (Free, Plus, Pro) to improve its models unless you switch off "Improve the model for everyone" under Settings, Data Controls, while Business, Enterprise, and Edu workspaces are excluded from training by default. Redact what you can, and use a workspace covered by your company's agreement for the rest.
• Formal tenders. An RFP with a mandatory structure and a compliance matrix is a different job. The model can extract the requirements into a list, and you still have to verify each one against the source document.
• Terms and conditions. The model can explain a clause but can't judge whether it protects you. If the proposal carries terms, read our piece on reviewing a contract with AI, and involve a lawyer for anything that binds you.

Where we go deeper

Commercial proposals sit in the second session of our two-week practicum, next to the technique this article is built on: making AI pull the context out of you instead of guessing it. In the session we practice that interview on real work tasks, and the assignment afterwards is a template for your most frequent correspondence. If proposals follow your outreach, the same session covers the earlier step too, which we wrote up in cold email prompts. For how the program maps to client work, see the page for consultants.