Custom GPT: Build a Proposal Assistant Trained on Your Own Templates and Rates

What This Builds

A Custom GPT loaded with your past proposals, your pricing tiers, and your standard terms, so a new proposal starts from your actual business logic instead of a blank page. You give it discovery-call notes and it drafts a scope, timeline, and price already matched to your voice and rate card. A full proposal that used to take 2 to 4 hours to write drops closer to 30 to 45 minutes of review and tweaking, since the structure and pricing math no longer need re-explaining every time.

Prerequisites

• Plus subscription ($20/month), required to build and save a Custom GPT
• 3 to 5 past proposals you're proud of, saved as text or PDF
• Your pricing tiers, typical project types, and standard contract terms written out somewhere, even roughly
• Total ongoing cost: $20/month a month, the same subscription you'd already need to draft proposals in plain ChatGPT

The Concept

A Custom GPT is a saved configuration of ChatGPT: a set of instructions, uploaded reference files, and settings that stay attached every time you open it. Instead of pasting your rate card and explaining your proposal format at the start of every conversation, you do that once, and the GPT remembers it every time after. It works like handing a new subcontractor your standard estimate template and a stack of old bids before their first job, so you only need to give them the specifics of the next one.

Build It Step by Step

Part 1: Gather your business inputs

Before opening the GPT builder, put together a single document with:

1. Your standard proposal structure (sections you always include: scope, timeline, price, terms)
2. Your pricing tiers or hourly/project rate logic, including how you handle common add-ons like extra revision rounds or e-commerce integration
3. Standard terms: deposit percentage, payment schedule, what happens with scope changes
4. 2 to 3 sentences describing the kind of client and project you take on most often

Part 2: Create the Custom GPT

1. In ChatGPT, click Explore GPTs in the sidebar, then click Create
2. In the chat panel on the left, describe what you want in plain language:

I want a proposal-writing assistant for my freelance web development business. Here is my standard proposal structure, pricing logic, and terms: [paste your document from Part 1]. When I give it notes from a discovery call, it should draft a complete proposal: scope of work, timeline, price broken out by phase or deliverable, and my standard terms, in my own voice, without me re-explaining my rates or template each time.

1. Answer the builder's follow-up questions as they come. It refines the configuration as you go, and you can keep adjusting in plain language at any point

Part 3: Tighten the system instructions

1. Click Configure at the top of the builder to see and edit the raw instructions directly. Useful additions:

   • "Never guess at a price. If the discovery notes don't specify enough detail to quote confidently, ask a clarifying question instead of filling in a number."
   • "Match the tone of the uploaded sample proposals: direct, no jargon, confident but not salesy."
   • "Always include a change-order clause in the terms section."
   • "Flag anything you're unsure about in brackets rather than presenting it as final."
2. Under Knowledge, upload your 3 to 5 sample proposals from Part 1. These teach the GPT your actual phrasing and formatting far better than a description of your style would

Part 4: Set visibility and test

1. Save the GPT with visibility set to Only me, unless you have a specific reason to share it
2. Test it with a realistic set of discovery notes:

New lead: local bakery wants a 6-page WordPress site, online ordering via WooCommerce for pickup orders only (no shipping), needs a blog for weekly specials, wants it done in 5 weeks, has existing branding and photos ready to send. Budget mentioned around mid-range.

1. Read the draft closely. If the pricing logic is off, the tone doesn't match, or a section is missing, go back into Configure and adjust the instructions rather than trying to fix it prompt by prompt each time

Real Example: Discovery Notes to Draft Proposal

Setup: Your GPT knows your standard structure (scope, timeline, price by phase, terms), your rate logic (project-based pricing with a 30% deposit, e-commerce add-ons priced separately from the base build), and the tone of three past proposals you uploaded.

Input: "Local bakery, 6-page WordPress site, WooCommerce for local pickup only, weekly blog for specials, 5-week timeline, has branding and photos ready, budget sounds mid-range."

Output: A complete draft proposal with a phased scope (design, build, WooCommerce setup, content integration, launch), a 5-week timeline broken into weekly milestones, pricing that separates the base site from the WooCommerce add-on using your actual rate logic, and your standard 30% deposit and change-order terms, written in the tone of your sample proposals.

Time saved: What used to take 2 to 4 hours writing from scratch becomes 30 to 45 minutes of reading the draft, adjusting anything specific to this client, and sending.

What to Do When It Breaks

• Pricing comes out wrong → open Configure and check exactly how you described your rate logic. Vague pricing instructions produce vague or inconsistent quotes; write the actual math out as a rule, not a general description.
• Tone reads generic or overly formal → upload another sample proposal that has the exact tone you want, and add an instruction pointing at it directly: "Match the tone of [filename] specifically."
• It quotes a price with missing information → strengthen the instruction to ask a clarifying question rather than guess. This is the single most important instruction in the whole build, since a wrong quote sent to a client is expensive to walk back.
• Knowledge files stop influencing the output → re-upload them under Configure; occasionally a file upload doesn't attach correctly on the first try, and the GPT quietly falls back to its general instructions.

Variations

• Simpler version: Keep a single saved prompt in a notes app with your rate card and terms pasted in, and paste it at the top of a new ChatGPT conversation each time. No builder, no Knowledge uploads, but you retype the setup context less than starting from zero.
• Extended version: Build a second Custom GPT for change orders, using the same pattern, so mid-project scope changes get the same fast, consistent drafting treatment as new proposals.

What to Do Next

• This week: Run every new proposal through the GPT and note where it gets the pricing or tone wrong
• This month: Fold those corrections back into the Configure instructions so the GPT keeps improving with use
• Advanced: Connect the GPT's output to a document template through an automation, so an approved draft turns into a formatted, ready-to-send PDF without a manual copy-paste step

Advanced guide for freelance web developer professionals. These techniques use more sophisticated AI features that may require paid subscriptions.

A note on sharing this GPT: your pricing tiers and proposal terms live inside the Knowledge files and instructions of this build. Keep visibility set to Only me. If you ever add a collaborator, publish it to the GPT Store, or share it with anyone outside your business, assume they can see or extract everything you uploaded, including your actual rates.