AI Prompts for Roofers: Seven Templates for the Office
- Seven prompts that turn ChatGPT or Claude into a decent office assistant: estimate emails, follow-ups, review replies, storm letters, job summaries, and blog posts.
- Copy them as written, swap in your details where the brackets are, and read the output before it goes anywhere.
- Never paste a customer's name, address, phone number, or claim details into a consumer AI tool. Placeholders in, real details after.
- The AI drafts. You decide. That order never flips, especially on price.
- Lesson 3 of 5. Lesson 2 covered the quick wins; lesson 4 is the rollout plan.
A prompt is just instructions. The difference between a useless AI answer and one you’d actually send is almost always the input. Type “write an estimate email” and you get a greeting card. Hand it the job details, the audience, and the rules, and you get something a homeowner in Hartland reads once and understands.
The seven below were written for roofing companies, not adapted from some generic marketing pack. Copy the whole block, replace the brackets, and follow the three rules at the bottom. If you haven’t seen lesson 2 on quick wins yet, these prompts are the manual version of several of them.
Seven prompts, ready to steal.
The estimate cover email
You wrote the number. Now write the email that keeps the homeowner from calling three friends to translate it. Most estimate emails are a PDF and a prayer. This one explains the price in the same paragraph it delivers it.
You write plain-English customer emails for a roofing company. Turn the job details below into an estimate cover email a homeowner can understand without calling us to explain it. Job details: - Roof size: [SQUARE FEET] sq ft, pitch [PITCH, e.g. 8/12] - Material quoted: [MATERIAL, e.g. architectural asphalt shingles] - Tear-off included: [YES/NO] - Total price: [PRICE] - Timeline: [START WINDOW] and about [NUMBER] days on site Rules: include one short paragraph explaining why this price for this roof (the size, the pitch, the tear-off, the material), in words a homeowner uses. No "per square" jargon; translate it. End with the single next step they should take. Under 200 words, friendly, no exclamation points, no pressure.
The quote follow-up sequence
A meaningful share of quotes die quietly. Not because the homeowner picked someone else, but because nobody ever asked twice. Three touches: day 2, day 5, day 14, each one shorter than the last.
Write three follow-up messages for a roofing quote that hasn't gotten a reply. The quote was [PRICE] for [JOB TYPE], sent on [DATE]. - Day 2: an email. Confirm they received the quote, offer to answer questions, and restate the one thing that makes our bid different: [YOUR DIFFERENTIATOR, e.g. we photograph the decking before and after]. - Day 5: a text message. Two sentences maximum. - Day 14: a final email. Let them off the hook politely, tell them the quote holds until [DATE], and leave the door open. Each message shorter than the last. No "just checking in" openers, no guilt, no discounts, nothing desperate. We are a busy company being helpful, not a hungry one asking twice.
The review reply
Works on any Google review, good or ugly. Paste the whole thing in. If you'd rather skip prompts entirely, the review reply tool on this site does this with no setup.
Reply to this Google review as the owner of a roofing company in [TOWN]. Review: [PASTE THE FULL REVIEW] Rules: sound like a person, not a customer-service department. Mention one specific detail from the review so it reads like I remember the job, because I do. If the review is negative: no template apology. Acknowledge the specific problem, say what we're doing about it, and move the details to a phone call. If it's positive: thank them without gushing. Never mention discounts and never ask for referrals inside a review reply. Under 100 words.
The storm-season door letter
For a specific neighborhood after a verified hail event, with the date and hail size pulled from the storm history tool. The hard rule is baked in: facts only. NOAA logged hail up to 2.00 inches at Big Bend on May 15, 2025. That sentence sells better than fear does, and it has the advantage of being true.
Write a one-page door letter for homeowners in [NEIGHBORHOOD OR TOWN] after a verified hail event. The facts, from NOAA storm records: hail up to [SIZE] inches fell at [LOCATION] on [DATE]. Hard rules: stick to the verified facts above. Do not claim or imply that their roof is damaged; we haven't seen it. No fear language, no urgency, no insurance-money promises. The offer is plain: we're a local company, we'll look at the roof and say honestly whether the hail did anything, and "nothing" is an answer we actually give. Include our credential line: Wisconsin Dwelling Contractor [NUMBER]. Under 250 words, signed by [OWNER NAME].
Photo notes to job documentation
Your crew already captions photos in CompanyCam or whatever you use. Those captions are a warranty file waiting to be assembled. This prompt does the assembling.
Turn these job-site photo notes into a clean job summary for the customer file and warranty documentation. Photo notes, in order: [PASTE YOUR CAPTIONS, e.g. "tear-off complete north face", "rotted decking SE corner, replaced 2 sheets", "ice and water shield at eaves and valleys", "ridge vent installed", "final magnetic sweep done"] Format: job date [DATE], crew lead [FIRST NAME]. Then a chronological summary in complete sentences: what was found, what was replaced or repaired and why, what materials went on, and what was verified at the end. Written so an insurance adjuster, or the homeowner three years from now, can follow every step. No abbreviations a civilian wouldn't know.
The interview prompt
The reason most contractor blogs read like a robot wrote them is that a robot wrote them, from nothing. Flip it: the AI interviews you about a real job, then writes from your answers. The post ends up full of things only you know.
You're interviewing me, the owner of a roofing company, about one specific job we finished. Ask me 10 questions, one at a time, and wait for my answer before asking the next. Cover: what the customer first called about, what we actually found up there, the decision point, what it cost and why, what surprised us, and how it ended.
After my last answer, draft a 500-word blog post built from my answers, keeping my words and rhythm wherever possible. If my phrasing is specific ("the decking was soft enough to push a thumb through"), keep it exactly. Fix grammar only. Title it with the town name and the problem. No marketing language, no "call today." Just the story, and one line at the end about who we are.The price-objection reply
The homeowner says another bid came in lower. The worst replies either cave on price or trash the other company. This one does neither.
A homeowner replied to our roofing quote saying it's higher than another bid they got. Draft my response. Details: our quote is [PRICE], their stated other bid is [PRICE, IF THEY SAID], the job is [JOB TYPE], and here's what our quote includes that cheap bids commonly leave out: [YOUR LIST, e.g. full tear-off, ice and water shield to code, decking inspection with photos, dump fees, workers comp on every person on the roof]. Rules: no trash-talking the other company. Explain what's inside our number in concrete terms, invite them to compare the two bids line by line, and make clear we're fine either way. Do not offer a discount. Under 150 words.
Prompt 03 has a no-typing version: the review reply tool takes a pasted review and hands back a draft. And prompt 04 wants real dates and hail sizes, which is exactly what the storm history tool holds for Waukesha County.
Three rules before you paste anything.
Keep customer details out of the box
Names, street addresses, phone numbers, and claim numbers don't belong in a consumer AI tool. Draft with the bracketed placeholders, then add the real details back in your email program. If you're on a business plan with training on your data turned off, the risk drops, but the habit is still the right one.
Read it before it ships
AI is at its most confident right before it's wrong. It will invent a shingle warranty term or an install date with total composure. Thirty seconds of reading catches it. Nothing generated goes to a customer unread.
The AI drafts, you decide
The draft is the machine's job. The judgment is yours: whether to send, what the price is, whether that customer gets the day-14 email or a phone call instead. Especially price. The number never comes from the robot.
Take prompt 02, open whichever AI tool you have, and run it against one real quote from the last two weeks that never got an answer. Read the day-2 email, fix anything that doesn’t sound like you, and send it. One quote, one email, tonight.
Asked before you had to ask.
Which AI tool should I use for these prompts?
Any of the big ones: ChatGPT, Claude, Gemini. The prompts matter more than the tool. For office use, get a paid plan so you can turn off training on your data, and keep the whole office on the same one so the prompt library from lesson 4 actually gets shared.
Can I just have the AI send the follow-ups automatically?
Eventually, and that's what the tools in lesson 5 are for. Start with drafts you read and send yourself, then automate the sequence once you trust the words. One caution: automated texting has consent rules attached, which is a good reason to run it through a real platform instead of duct tape.
Will Google punish AI-written blog posts?
Google cares whether the page is useful, not who typed it. What gets punished is generic filler published at scale. The interview prompt exists precisely so your post contains things no other roofing site has: your job, your decking, your decision. A robot can't fake that, which is why it interviews you first.
Not sure whether your follow-up problem is the words or the plumbing behind them? The audit grades your site, profile, and reviews in about a minute. It’s $0 and doesn’t ask for your email.