Write a Business Proposal with AI: From a Brief, Not a Template
How to write a business proposal with AI without the template feel: a five-question brief, a Claude Project with your tone, a sample prompt, and a fact check.
A templated proposal gets spotted in about ten seconds. "Industry-leading solutions," "tailored approach," "win-win partnership." Those emails go straight to trash, because they miss the one thing that matters: proof that you understand who you are writing to and which problem you are solving.
AI can cut both ways here. Ask it to "write a business proposal for a client" and you get the same template, just generated in a minute. Give it a real brief and your own context, and you get a draft worth refining instead of rewriting. The difference is not the model, it is what you feed into it.
Why templated proposals fail instantly
A template gives itself away with three tells. Generic claims instead of specifics: "we boost efficiency" could be written by anyone, to anyone. The client is missing from the text: five paragraphs about you, zero about their problem. And a structure the decision maker has already seen twenty times this week.
The reader scans your proposal with one question in mind: "do these people actually get what I need?" If the first two lines say nothing about their situation, their numbers, or their pain, they stop reading. That is why a good proposal starts with a brief, not with writing.
The brief: five questions before the first prompt
Before you open Claude, answer five questions in writing.
- Who is the client, and who will read it. The company, the decision maker's role, what they care about: price, speed, reliability, reporting.
- What hurts. What is on fire for them right now, and why they would open your email at all. If you do not know, find out before sending, not after.
- What you offer. Exactly what you propose, in what scope, with what outcome.
- What numbers you have. Real ones only, from your reports and past projects: timelines, costs, results.
- What objections to expect. What the client will push back with and how you answer: "too expensive," "we tried this already," "no time."
This brief is your proposal in raw form. AI will turn it into clean, structured text, but it cannot invent the substance for you. More precisely, it can, and that is the worst outcome: one made-up number kills a deal faster than sending no proposal at all.
Writing a business proposal with AI: five steps
Step 1. Set up a Claude Project. Create a project called "Proposals" and upload two or three of your best past proposals, a service list with current prices, and a short note on how your company writes: greeting, length, what you avoid. How Projects work and why context beats clever prompt wording is covered in a separate post.
Step 2. Write the brief straight into the chat. No polish needed, just the five questions above. The numbers below are placeholders, use your own from real reports.
``` Draft a business proposal from this brief.
Client: a six-location coffee chain, the reader is their marketing director. Pain: their previous social media agency missed two launches, posts keep going out late. Offer: full social media management, a month of content planned ahead, posts on schedule. Numbers: take them from the attached report on a similar client, add nothing on your own. Objections: "we already got burned by an agency," "too expensive." Format: one page, no warm-up paragraphs, pricing at the end with two packages. Tone: match the past proposals in this project. ```
Step 3. Read the draft as the client. Two questions: can you tell from the first lines that this was written for this specific client, and would you defend every claim in a live meeting?
Step 4. Refine with targeted asks. Not "make it better" but specifics: "rewrite the opening around the client's pain," "answer the price objection with the case from the report." Two or three rounds are usually enough.
Step 5. Verify every number, then send. Each figure in the text must have a source. More on that below, because this is where AI slips most often.
Where AI gets it wrong and what to check by hand
Invented numbers. If a figure was not in your brief but shows up in the proposal, it is a hallucination. "We will grow your sales by 40%" sounds convincing right up until the client asks where the number came from. Simple rule: a number either comes from your file, or it does not belong in the text.
Promises you never made. The model loves to sweeten the offer: "launch within a week," "guaranteed results." Sounds great, but you are the one answering for it in the meeting. Cut everything you would not sign your name under.
Corporate speak creeping back. Even with a good brief, the text sometimes slides into "leveraging synergies" and "end-to-end solutions." The fix is real past proposals in the project plus a direct instruction: "write like a human, drop the corporate speak."
Facts about the client. Ask it to "add a paragraph about their company," and the model may fill gaps with things it does not know: wrong city, wrong scale, outdated products. Everything about the client comes from your own sources only.
Stale prices. If the proposals you uploaded carry last year's prices, they will leak into the new text. Keep a current price list in the project and state explicitly that it is the single source of truth for pricing.
What not to upload
One rule: do not upload anything you are not prepared to see outside your company.
- Clients' personal data. IDs, personal phone numbers, home addresses. A proposal does not need them anyway.
- Documents under NDA. Contract details a client shared with you under a non-disclosure agreement.
- Credentials and account details. Bank accounts, passwords, API keys.
- Internal costs and margins, if your company keeps them private. Final prices are enough for a draft.
Case studies can be anonymized with no loss: "a six-location coffee chain" works as well as the real name. And check the privacy settings of your Claude plan: how your data is stored and used depends on the plan and on what you have enabled.
The short version
The working setup: a brief built on five questions, a project with your tone and past proposals, a draft, targeted editing, and a manual check of every number. Yes, AI can generate a template in a minute, but the client will spot it just as fast. The context only you have is the reason your proposal gets read all the way to the price. For a wider look at where else Claude fits into everyday work, see the overview of Claude for business.
If you want to build the skill hands-on, AGINE Academy teaches through game missions inside the real Claude, with an AI mentor along the way, and every lesson ends with a working result instead of notes. The starter block of four lessons is free and open without signup, so you can test the brief mechanics on your very next proposal.
Questions
It can draft one, but not finish it. The model is good at structure and wording, but the substance comes from your data: the offer, the numbers, the answers to objections. Without a brief you get a template the client spots in the first paragraph. The working scheme: brief, draft, targeted edits, manual fact check.
Two or three of your best past proposals, a service list with current prices, and a short note on your company's tone: how you address clients, how long you write, what you avoid. Every new draft then sounds like you rather than like averaged internet text.
In the prompt, explicitly forbid the model to add numbers of its own and point to the source: an attached report or price list. After the draft, check every figure against the file. If a number appears that was not in your materials, it is a hallucination, so cut it without hesitation.
Clients' personal data, documents under NDA, bank details, passwords, and internal costs if your company keeps them private. Anonymize case studies: business type and scale work as well as a real name. Also review the privacy settings of your Claude plan.