Build a Custom GPT that writes your proposals
Build a Custom GPT that writes your proposals
A step-by-step workshop to build a custom GPT trained on your own past proposals, so it drafts new ones in minutes instead of hours. Follow along and build it while you read.
The proposal that took three hours, and what we're building
A potential client asks for a proposal. You block out your evening. Three hours later, you've rewritten the same sections you write every time, your process, your pricing structure, your "why work with me," just with different client details swapped in.
This happens every single time, because you're starting from a blank page instead of your own proven template. The fix isn't writing faster. It's not starting from scratch.
By the end of this workshop, you'll have a custom GPT trained on your own past proposals, one that drafts a new one in your actual voice from a short brief, not a generic template.
Before:
New enquiry → Blank document → 2-3 hours of rewriting your own boilerplate → Send
After:
New enquiry → A few bullet points → Custom GPT drafts it in your voice → 15 minutes of editing → Send
What it will do:
Draft proposals using your real structure, tone, and pricing approach, based on 2-3 of your actual past proposals.
What it won't do:
Know a new client's specific needs without you telling it, or replace the judgment calls only you can make about scope and pricing for a genuinely unusual project.
What you need before you start
What you need | Cost |
|---|---|
ChatGPT Plus, Team, Pro, or Enterprise subscription | Plus is $20/month, this is the entry point |
2-3 of your best past proposals (as documents) | Free, you likely already have these |
20-30 minutes to prepare your content | Free |
Important honesty note:
Building a custom GPT requires a paid ChatGPT plan, there's no free-tier option for this specific feature. Free accounts can use GPTs other people have published, but can't build their own. If you're not already on a paid plan, this is the one prerequisite you can't work around.
Why a custom GPT instead of just asking ChatGPT normally:
A regular conversation starts from zero every time, you'd have to re-paste your writing style and past examples in every new chat. A custom GPT saves all of that permanently, so it's ready the moment you open it, no re-explaining required.
Step 1: Choose your source proposals, then write your GPT's instructions
Part A: Choose your source proposals
Pick 2-3 proposals you're genuinely proud of, ones that reflect how you actually write, not your very first attempt from years ago. Save them as PDF or Word documents, you'll upload these directly.
Why this matters: the GPT learns your voice from these documents. Weak source material produces weak output, this is the single highest-leverage choice in the whole setup.
Part B: Write your GPT's core instructions
This is what tells the GPT how to behave every time you use it. Fill in the brackets based on how you actually work.
Copy this:
You are a proposal-writing assistant for [Your Name/Business], a [what you do, e.g. "freelance Framer web designer"].
Your job is to draft new client proposals using the same structure, tone, and pricing approach shown in the uploaded past proposals. Do not invent services, prices, or timelines that aren't consistent with what's shown in those documents or what I tell you directly.
When I give you a new brief, ask for:
Client name and what they need
Rough scope (what's included)
Timeline
Any pricing specifics that differ from my usual approach
How to write:
Match the tone of my uploaded proposals: [describe it, e.g. "confident, warm, not overly formal"]
Keep the same section structure I've used before (don't reinvent the format each time)
Flag clearly with [CONFIRM] anywhere you're unsure about a price or detail, don't guess silently
Your result:
Your source documents ready to upload, and a complete instruction block that defines exactly how your GPT drafts every proposal.
Step 2: Build the GPT in ChatGPT's builder
This is the actual setup, verified against OpenAI's current GPT Builder, which is a no-code, conversational interface, no technical knowledge required for what we're building here.
Go to "Explore GPTs" in ChatGPT, then click "+ Create." This opens the GPT Builder.
Use the Configure tab. Give it a clear Name (e.g. "My Proposal Writer") and a short Description. Then paste your full instructions block from Step 3 into the Instructions field.
Upload your knowledge files. In the Knowledge section, upload your 2-3 chosen past proposals. The GPT will reference these directly when drafting new ones, this is what makes the output sound like you, not a generic template. You can upload up to 20 files total if you want to add more examples later.
Skip Capabilities and Actions for now. These add extra features like web browsing or API connections, genuinely unnecessary for proposal writing, and Actions specifically require technical setup we don't need here.
Set visibility to "Only me." Since this GPT will reference your real client proposals and pricing, keep it private, don't publish it to the GPT Store or share the link publicly.
Save. Your custom GPT is now ready to use, accessible anytime from your GPTs list.
Your result:
A live, working GPT trained on your real proposals and instructions.
Step 3: Test it properly, then fix what's broken
Part A: Test it
Run these test briefs through your new GPT and check the output against each expectation:
A brief matching a past client closely
Should sound very close to your original tone and structure
A brief for a slightly different service than your uploaded examples
Should adapt reasonably, not just copy-paste the wrong sections
A brief with an unusual or unclear budget
Should flag it with [CONFIRM] rather than inventing a number
A very short, vague brief ("I need a website")
Should ask clarifying questions rather than guessing
Two different briefs back to back
Should stay consistent in tone across both, not drift
Part B: Fix what's broken
Output sounds generic, not like you
The instructions didn't specify tone clearly enough.
Go back and describe your actual voice more precisely (e.g. "direct, no corporate jargon, short sentences")
GPT invents pricing or scope details
Instructions didn't emphasize this rule strongly enough.
Re-add the "don't invent, flag with [CONFIRM]" rule more explicitly
Structure changes every time
The uploaded proposals may have inconsistent formats themselves.
Consider uploading proposals that share a more consistent structure, or explicitly describe your standard section order in the instructions.
GPT doesn't ask for missing details
The "when I give you a brief, ask for" section wasn't clear.
Make this list more explicit and put it near the top of your instructions.
Your result:
A tested, reliable proposal drafter, refined against your actual real-world briefs.
Real business examples, and what it should never do
Freelance designer:
A new enquiry comes in for a 5-page website. Instead of starting from scratch, a quick brief goes into the GPT, name, scope, rough budget, and a first draft comes back in the designer's own structure and tone within minutes, ready for a final read and small edits before sending.
Marketing consultant:
A prospective client wants a quarterly retainer proposal. The GPT drafts it using the consultant's standard retainer structure, adjusted for the specific scope discussed on the discovery call, saving the usual hour of reformatting a template by hand.
Small agency:
A junior team member needs to send a proposal but isn't confident writing one from scratch yet. The GPT gives them a strong first draft in the agency's established voice, which a senior team member reviews before it goes out, faster turnaround without sacrificing quality control.
(These are illustrative workflows based on how this setup is designed to work, not measured results from a specific client, since real outcomes depend on your own proposals and use.)
What it should never do
Don't send a GPT-drafted proposal without reading it first, always review before sending
Don't let it invent pricing, timelines, or scope details you haven't confirmed
Don't upload sensitive client information you wouldn't want referenced in future drafts
Don't treat it as a replacement for your own judgment on genuinely unusual or high-stakes proposals
What you've built and troubleshooting
What you've built:
Before, every new enquiry meant 2-3 hours rebuilding your own boilerplate from memory. Now, a short brief produces a real first draft in your own voice within minutes, leaving you with editing time instead of writing-from-scratch time.
Troubleshooting:
Output doesn't match my voice at all
Your uploaded proposals may be too different from each other, or too old → Choose more recent, more consistent examples and re-upload
GPT forgets my instructions over a long conversation
Start a fresh conversation with the GPT for each new proposal rather than one long ongoing thread
I want to update it with a new proposal example
Go back into the GPT's settings and add the new file to Knowledge, you don't need to rebuild it from scratch
FAQs:
Do I need coding knowledge?
No, the GPT Builder is entirely conversational and no-code. The only part of custom GPTs that needs technical knowledge is Actions (API connections), which this workshop doesn't use.
Is this really $20/month, nothing more?
For the setup in this workshop, yes, ChatGPT Plus covers building and using custom GPTs. Team and higher tiers exist mainly if you want to share the GPT with people who don't have their own ChatGPT accounts, not needed here.
Can I use this for other document types too, not just proposals?
Yes, the same approach (upload real examples, write clear instructions) works for quotes, contracts summaries, or any recurring document with a consistent structure.
Will it work if my past proposals are inconsistent in format?
It'll still work, but less precisely. The more consistent your source documents, the more reliable the output.
Can I share this GPT with my team?
Yes, but that requires a Team plan and changing the visibility setting, and you should think carefully about whether your team should have access to the client proposal examples uploaded as Knowledge.
Where to go from here
You now have a working proposal-writing assistant trained on your real work, not a generic template pretending to be personal.
Use it on your next 2-3 real enquiries and see how much editing time drops. You'll likely want to tweak the instructions once or twice as you see real output, that's expected, not a sign something's wrong.
If you'd rather have a custom AI tool like this built and refined for your specific business, let's talk.
Rahul Singh
Framer Designer & AI Consultant
Written by a Framer designer and AI consultant based in Pune. I write about design, AI implementation, and building a freelance business. More about Rahul.
You may also like to read
Subscribe now!
Get latest insights first. No spam thats my promise.