Plenty of reps draft proposal copy in ChatGPT, and honestly: it works better than the blank page. But most people prompt it the way they'd prompt a search engine (“write a business proposal for a software company”) and get back the average of every mediocre proposal on the internet. The difference between generic filler and a usable draft is almost entirely in what you feed it.
Here are the prompts we've seen produce genuinely useful proposal copy, the mistakes that make AI proposals sound like AI, and, since we build a tool in this space, an honest note on where prompting stops scaling.
The setup prompt that changes everything
ChatGPT only knows what you paste into the box. So before asking for any section, front-load the context once, in a structured way:
1) SELLER: [your company, what you sell, in two sentences]
2) BUYER: [company, industry, size, who the decision-makers are]
3) WHAT THEY TOLD US: [paste your actual discovery notes: their goals, pains, objections, timeline, in their words]
4) THE OFFER: [what you're proposing, pricing structure, timeline]
Don't write anything yet. Confirm you have it and ask me for whatever's missing.
Part 3 is the one people skip and the one that matters most. The words your buyer used on the call are the strongest raw material you have. A proposal that opens with their phrasing reads like listening; one that opens with your boilerplate reads like a mail merge.
Six prompts for the sections that matter
1. The executive summary
2. The challenges section
3. The approach, mapped to their problems
4. Pricing framing (not the pricing itself)
5. The de-risking pass
6. The cut pass
The mistakes that make AI proposals sound like AI
- Prompting for the whole proposal at once. You get uniform mush. Section by section, with a review between each, is slower and dramatically better.
- Letting it write about your company. ChatGPT is at its worst on “about us” copy. That's where “leading provider of innovative solutions” lives. Write that part yourself, once, and reuse it.
- Accepting its case studies. It doesn't know your customers. Any customer story it writes is fiction wearing your logo.
- Skipping the read-aloud test. If a sentence would sound absurd said across a table (“we are thrilled to present this comprehensive solution”), it goes.
Where prompting runs out
Used this way, ChatGPT is a real writing assistant. But notice what the workflow actually is: you gather the context by hand, paste it in, prompt section by section, police the facts, then take the text somewhere else to design, price, and send it. For one proposal, that's fine. For every proposal, it's a second job, and it resets to zero each time, for every rep, with quality depending on who wrote the best prompts that day.
That gap, between AI that can write and a system that knows your deals, is exactly what we built Deck'd to close. The deal is the prompt: CRM data and meeting notes load automatically, the AI can only cite case studies from your approved library, and the output is a designed, trackable proposal rather than text to reformat. We wrote up the full comparison in Deck'd vs ChatGPT, including the cases where ChatGPT is still the right tool.
Either way: stop prompting “write a business proposal.” Feed it the call. The call is where the proposal was, all along.