ChatGPT for Sales Proposals

Outline and retone. Price and promises stay in the CRM

Sales proposals are a bad place to discover that ChatGPT completes patterns. A three-year software proposal will attract a fluent scope, a helpful discount, and implementation weeks that match some other company's blog. None of that is your deal. Your deal lives in the CRM, the price list, the statement of work, and the signature block. Using ChatGPT for sales proposals means asking for an outline and a clearer tone after those commercial facts are already chosen. You still put every number back from the system of record. You still decide what you are promising. The model does not attend the negotiation. You do. A generated proposal is a draft until a named person signs.

The typical failure is write a proposal from these call notes. The notes mention a budget signal and a wish for a fast start. The draft turns the signal into a price and the wish into six weeks. A buyer highlights the six weeks. You now own a timeline nobody in delivery accepted. Another failure is a discount the model added to sound commercial. Your job is to keep price, scope, and promises under human control, and to use the model for structure, headings, and a retone that does not add commitments. If the outline cannot be filled from the CRM without new numbers, the outline is wrong. Change the outline. Do not invent numbers so the pages look complete.

This essay is for people who already sell, not for a demo that produces a fake proposal in thirty seconds. Thirty-second proposals are how fiction becomes a PDF. Take the extra hour. Fill the commercial core from the system. Then let ChatGPT tighten the story around that core. If legal or deal-desk must approve wording, send them the sourced version, not the chat. A named signer still sits at the end. If you cannot say who that is, you are not ready to generate. You are still in discovery, and discovery notes are not a proposal.

What the model may touch, and what the CRM owns

Let AI outline sections, retone a blunt paragraph for a senior buyer, and produce a shorter executive summary from facts you paste. Let it list questions a procurement reader will ask. Do not let it set list price, discount, payment terms, liability language, or go-live dates. Those have owners: commercial, legal, delivery. Give the model the buyer, the problem you are allowed to claim, and the products already on the opportunity. Ask for [missing] where a section needs a figure. An empty cell in a pricing appendix is more honest than a plausible number. Honest is how you stay in the deal after due diligence, when someone checks the PDF against the quote line.

Competitive language is a special risk. ChatGPT will happily slander a rival or invent a benchmark. You do not need that. You need your differentiator as your company states it, and numbers you can show. Paste the approved talk track, not a request to beat vendor X. If you use an approved copilot on a proposal already in Microsoft 365, the same split applies. The tool can rearrange. It cannot authorise a discount. Deal-desk still authorises. You still sign. Speed that skips deal-desk is not a win. It is a future credit note.

A proposal brief that cannot invent the deal

Export or paste only the opportunity fields you are allowed to use: buyer, products, approved price, known constraints. Write the one decision you want: a meeting, a signature, a clarification. Ban new numbers and new products. Ask for an outline with [missing] on commercial gaps. Generate once. Fill gaps from the CRM and delivery, by people, not by another prompt. Then ask for a retone of your filled draft, still with a ban on new commitments. Read every sentence that contains will, include, by, and percent. If it is not on the quote line, cut it. Send the PDF only after the named signer has seen those sentences. The CRM and the signed PDF are the record of the deal.

Save a brief per proposal type: renewal, new logo, services add-on. Do not start from a blank write me a proposal when the quarter is tight. Tight quarters are when invented weeks appear. Keep legal clauses in the approved template, not in the chat. If the model rewrites a limitation of liability to sound friendlier, that is not friendliness. That is a different contract. Put the template language back. Your review is commercial and legal, not literary. A prettier PDF with the wrong cap is a worse artefact than an ugly PDF with the right cap.

Role: You are outlining a sales proposal I will price and sign.
Task: Produce a section outline and a short summary from the opportunity facts I paste.
Context: Buyer, products on the opportunity, and the one decision I want next.
Constraints:
- Do not invent prices, discounts, dates, products, or case studies.
- Write [missing] wherever a commercial figure or a delivery week is required.
- Do not rewrite legal or liability language. Leave a placeholder for the template.
Output: Outline, summary, and a list of commitments a buyer could think we made.
Quality checks: Which sentences would bind us if this PDF were sent unchanged.

Fill every [missing] from the CRM and from delivery, in that order. Run a retone only after the commercial core is true. Search the draft for will, include, by, and percent, and match each hit to a quote line. If it does not match, delete it. Put the approved legal template back wherever the model softened a cap. Then send the PDF to the named signer. The outline was useful. The invented week would not have been.

Invented discounts, fake weeks, and a PDF that becomes the deal

A discount in a sentence is a discount in the negotiation. Buyers screenshot. Implementation weeks in a summary become the plan, even if delivery never saw them. A case study the model invented becomes a representation. Your catch is mechanical: search the verbs and the numbers. Then ask delivery to initial the weeks. Then ask deal-desk to initial the price. If either refuses, the proposal is not late. It is wrong. Wrong should not go out to make a forecast date. Forecasts recover. Signed fiction does not. Also watch pasting a customer's pricing from a rival bid into consumer ChatGPT to compare. That may be their confidential information. Classification still applies when you are trying to win.

Do not let the model write the email that says we can do this cheaper unless that cheaper is already approved. The email is a proposal too. Do not chain chats from an old deal's PDF to a new logo with different terms. The old limitation clause will wander in. Start from this opportunity's fields. If your organisation uses a configure-price-quote system, the proposal narrative must follow that system, not the other way round. ChatGPT does not become the CPQ. People who treat it as one create a second, unofficial price list in their download folder. That folder will leak. The CRM will not match. Finance will notice after the win, which is the expensive time to notice.

Related reading on StudyGrid

Read next: Writing Someone Will Sign How to Review AI Drafts AI for Competitive Analysis. Those essays sit beside this one. Use them when you need the neighbouring skill, not as a substitute for the check you still have to make.

What to do this week

Take one live opportunity. Paste only allowed CRM fields into an approved tool. Generate an outline with [missing] on price and weeks. Fill those from the CRM and from delivery. Retone once. Search for promises. Get the named signer to initial the commercial sentences. Send that version. Keep the brief if it did not invent a discount. That is ChatGPT for sales proposals as a method, not as a faster way to guess.