How to use AI to write proposals that actually win.

Set up the proposal context (one-time, 30 minutes)

Create a dedicated Claude Project titled "Proposals — [Your Company]." Upload three things to the knowledge base:

1. Your standard proposal structure. The actual section headings you use. Executive summary → understanding → approach → scope → pricing → timeline → team → terms. Or whatever your version looks like.
2. Three to five anonymized winning proposals. The wins, not the misses. The AI learns your voice and structure from examples.
3. Your brand voice guide. If you don't have one, write 5 bullet points: tone (formal/casual), perspective (first person/brand voice), things to avoid (jargon, AI-cliche openers, etc.), things to favor (specifics, numbers, named outcomes).

Then write a system prompt for the Project. Use this template:

You are the proposal-drafting assistant for [Company]. We sell [what we sell] to [who we sell to]. Our value proposition is [one sentence].

When drafting a proposal:
- Use the standard structure in the knowledge base
- Match the voice of the example proposals — not generic consulting speak
- Cite specifics from the discovery notes; do not invent details
- If a detail is missing, write [TK: ask sales team about X] instead of guessing
- Pricing: always reference our standard rate card in the knowledge base
- Executive summary: 3 paragraphs maximum, focused on outcomes not features

Capture the deal-specific discovery

For each new proposal, gather three inputs:

— The RFP language, statement of work, or whatever the client sent over
— Your discovery call notes (raw is fine — the AI extracts structure)
— The Gong/Chorus transcript if you have one

Paste them all into a new chat inside your Proposal Project. Tell Claude: "Here's the discovery for the [Client Name] proposal. Read it carefully and ask me up to 5 clarifying questions before drafting."

Generate the structured draft

Once you've answered the clarifying questions, prompt:

Now draft the proposal using our standard structure. Match the voice and depth of the example proposals in the knowledge base. Flag any section where you're guessing with [TK: ...]. Don't write the pricing yet — that's a separate pass.

The output should be roughly 80% of a finished proposal. The 20% gap is what you fill in step 4.

Edit for voice and accuracy (the actual work)

Two passes:

Pass 1 — Accuracy. Read every [TK: ...] and resolve it. Read every claim about your capabilities and verify it. The AI will occasionally claim you've done something you haven't. Find those and fix them. Time required: ~20 minutes.
Pass 2 — Voice. Read aloud. Anywhere it sounds like a marketing brochure, rewrite. Anywhere it uses an AI-cliche phrase ("In today's rapidly evolving landscape...", "It's important to note that..."), kill it. Time required: ~15 minutes.

Iterate on the close

Spend 80% of your remaining time on two sections: the executive summary and the pricing/scope language. These are the only parts of the proposal most readers actually read carefully. The rest is reference material.

For the executive summary, ask Claude to draft 3 alternative versions — different angles, different lead-ins. Pick the one that hits hardest on the client's actual stated priorities, and refine.

For pricing language, ensure the value framing precedes the dollar figure. Never let the dollar figure be the first thing on the page.