Estimate Drafts That Sound Like You: Training AI on Your Voice

Estimate Drafts That Sound Like You: Training AI on Your Voice

A generic AI estimate draft is better than a blank page — but only slightly. It reads like every other contractor’s estimate because it is every other contractor’s estimate: average phrasing, average structure, average exclusions. This article is about closing that gap: how to train AI on your past estimates so drafts come out in your format, your tone, and your line-item structure — while keeping your numbers firmly in your hands.

The problem: the uncanny generic estimate

You’ve seen it — maybe you’ve sent it. An estimate that says all the right things but sounds like nobody. “We propose to furnish all labor and materials necessary to complete the following scope of work.” Your customers have seen that sentence from five contractors this month. It doesn’t build trust; it builds a pile of identical proposals where the only differentiator is price.

Your best estimates don’t read like that. They read like you: how you describe work, the order you list things, the exclusions you’ve learned to include the hard way, the plain-spoken cover note. That voice is an asset — it’s part of why customers pick you. The goal isn’t to let AI flatten it. It’s to teach AI to draft in it.

How it works: examples, not instructions

Here’s the key insight: AI learns your voice from examples, not from descriptions. Telling it “write in a friendly but professional tone” produces generic friendly-professional text. Showing it two of your actual estimates produces drafts that mirror your structure, your phrasing, and your habits.

The technique is simple and has a technical name — few-shot prompting — but all it means is: paste in 2–3 examples of your best past estimates, then your new job notes, and ask for the new estimate in the same format and style. The AI pattern-matches against your examples instead of its generic training.

Step 1: Pick your example estimates

Choose 2–3 estimates you’re proud of — ones that won the job, read clearly, and represent how you want to sound. They should be:

• Recent: your pricing, terms, and format evolve. Use last year’s, not 2019’s.
• Varied: if possible, pick different job types (a small repair, a mid-size project) so the AI sees your range, not just one template.
• Clean: correct exclusions, clear line items, your real tone. The AI will copy flaws as faithfully as strengths — audit the examples first.

Step 2: Strip or keep the numbers deliberately

You have two options. Keep the numbers if you want the AI to learn your price-book patterns (e.g., how you price linear feet of trench, your standard markup structure). Replace them with placeholders ([PRICE]) if you’d rather the AI learn only format and tone, and supply fresh numbers each time. For most contractors, placeholders are safer — it forces the “numbers come from you” discipline every single draft.

Step 3: Build your standard prompt

Save one prompt that includes your examples and your rules. Reuse it for every estimate — consistency compounds. (The copy-paste prompt below is built for this.)

Step 4: Review for voice drift

Every few estimates, compare a new AI draft against your example estimates. Is the structure holding? The tone? The exclusions? If the drafts are drifting generic, your examples may need refreshing — or your instructions need tightening. Voice is maintained, not set-and-forget.

What to standardize in your format

Beyond tone, the highest-value thing to teach AI is your line-item structure — the anatomy of how you break down a job. Most experienced contractors have an implicit system:

• Order: demolition first, then rough work, then finish? Or by trade? By area? Whatever your order is, the AI should follow it.
• Granularity: do you itemize every fixture or group by room? One line for “tile labor” or separate lines for demo, prep, set, and grout? Your granularity reflects how you think about the work — teach it.
• Exclusions block: this is the most valuable paragraph in your estimate. Your standard exclusions (permits, engineering, hazardous materials, paint touch-up by others, etc.) exist because of past pain. Make them permanent in the prompt so no draft ever goes out without them.
• Terms block: deposit, payment schedule, warranty language, validity period (“this estimate is valid for 30 days”). Standardize once, reuse forever.
• Cover note style: two sentences or two paragraphs? First person or company voice? Warm or strictly business? Show the AI, don’t tell it.

What to never let AI invent

This deserves its own section because it’s where the money is. When training AI on your examples, draw a hard line:

• Prices. The AI may notice your examples price tile at $X/sq ft and apply it to the new job. That’s pattern-matching, not pricing. Every price on every draft comes from your notes or your price book, verified by you.
• Measurements and quantities. Square footages, linear feet, counts, hours — all from your site notes. If the notes don’t have them, the draft gets [NEED FROM OWNER], not a guess.
• Material selections. “Quartz counters” in your example doesn’t mean quartz in this job. The AI will happily carry selections across jobs if you let it.
• Timeline commitments. Your example said “two-week duration” for a different job. This job’s timeline comes from your current schedule, not a past document.
• Customer-specific details. Names, addresses, and job-specific conditions never carry over. This sounds obvious until an AI draft addresses the new customer by the old customer’s name.

The principle: examples teach format and voice; notes supply facts. Never let the two streams cross.

An illustrative example

Picture a typical 5-person painting contractor. His estimates have a distinctive structure he’s refined over years: prep work itemized separately from painting (because customers undervalue prep), a “what we don’t do” section he’s expanded after three bad experiences, and a short cover note that always mentions the crew lead by first name.

He feeds two past estimates into his standard prompt as examples — with prices replaced by placeholders — and adds his exclusions and terms as permanent blocks. The first AI draft comes back: prep separated from paint, exclusions intact, cover note in his voice mentioning the crew lead’s name from his notes.

Review time drops because the structure is right on the first try — he’s only checking numbers and job-specific details instead of reorganizing the whole document. And customers keep getting the estimates that won jobs before, because the format that won is now the format every draft starts from.

Try it: a copy-paste prompt

This is the voice-training version of the estimate prompt. Paste your 2–3 example estimates where indicated, then reuse:

Draft an estimate for my [TRADE] business, matching the format, structure, and tone of my example estimates below. This is a FIRST DRAFT — I will verify every number.

=== EXAMPLE ESTIMATE 1 ===
[PASTE YOUR FIRST EXAMPLE ESTIMATE HERE — prices replaced with [PRICE] placeholders]

=== EXAMPLE ESTIMATE 2 ===
[PASTE YOUR SECOND EXAMPLE ESTIMATE HERE]

=== MY STANDARD EXCLUSIONS (include on every estimate) ===
[LIST YOUR STANDARD EXCLUSIONS]

=== MY STANDARD TERMS ===
[DEPOSIT, PAYMENT SCHEDULE, WARRANTY, ESTIMATE VALIDITY PERIOD]

=== NEW JOB NOTES ===
[PASTE YOUR SITE NOTES HERE — include all prices, measurements, quantities, materials, and dates]

Rules:
1. Match the examples' section order, line-item style, and tone. If the examples separate prep from finish work, do the same.
2. Use ONLY the facts in my job notes. Do not carry over prices, materials, measurements, timelines, or customer details from the examples.
3. Anything missing gets [NEED FROM OWNER] — never invented.
4. Keep my exclusions and terms blocks intact on every draft.

Before you hit send: the human-review checklist

• Voice check: read the draft next to one of your example estimates. Same structure? Same tone? No generic drift?
• Numbers: every price, quantity, and measurement from my notes — nothing carried over from the examples.
• Customer details: correct name, address, and job-specific conditions. No leftovers from example jobs.
• Exclusions and terms: present, current, and complete.
• Scope: describes this job’s work — materials, selections, and timeline from my notes only.
• Flags: all [NEED FROM OWNER] items resolved or removed.

Where this goes next

Voice-trained drafting extends past estimates: change orders in your format, proposal letters in your tone, follow-up emails that sound like they came from your office — because they did. The examples-and-rules pattern is the same; only the document changes.

On a free AI Workflow Review, we’ll take one of your real estimates and build your voice-trained prompt with you — live, on your own work, so you leave the call with something usable. And if you want every workflow in the office set up this way — estimating, follow-up, documentation, all trained on how you operate — that’s our $2,500 AI Audit: a full implementation roadmap, built around your business.

Book your free AI Workflow Review — bring your best estimate; we’ll make the AI write like you.