How to use ChatGPT in a contracting business without making a mess

ChatGPT is a chat assistant built by OpenAI. You type a request, it writes a response. For a contracting business, that makes it a decent office helper for the writing nobody has time for: the reply to a long customer email, the rain-delay policy, the description on a quote, the job ad for a second crew lead. It is a poor estimator, a poor scheduler and a poor lawyer, because it doesn't know your numbers and it can state wrong things with full confidence. This guide covers how to set it up, which jobs to give it, how to ask, and where to stop.

What is ChatGPT actually good at in a contracting business?

It is good at language work where you supply the facts. Hand it rough notes and it returns clean sentences. Hand it a long, angry email and it returns a short summary of what the customer actually wants. Hand it your bullet points about a warranty and it returns a readable paragraph. The pattern is always the same: you bring the knowledge, it does the typing and the tidying.

It is weak anywhere the answer depends on information it doesn't have. It doesn't know what you pay for materials this month, how long your crew takes to strip a deck, which homeowner already got a discount, or what your county's permit office requires. Ask about those and it will still produce an answer, often a plausible one. NIST's 2024 generative AI risk profile has a name for this: confabulation, the production of confidently stated but false content. Plan your use around that weakness instead of hoping it won't show up.

How should you set it up before the first real use?

1. Use a company account. Sign up with a business email address the company controls, not a crew member's personal Gmail. When that person leaves, the chats and the settings stay with the business.
2. Read the data controls. OpenAI's help center describes a setting called "Improve the model for everyone". When it is off, your new conversations won't be used to train OpenAI's models. Decide on that setting deliberately before anyone pastes in job details.
3. Know the plan differences. OpenAI's business data page says that by default it does not use data from its business plans or its API for training. Consumer and business plans are handled differently, so read the current terms for the plan you actually pay for.
4. Write standing instructions. ChatGPT lets you save background about yourself and how you want answers written. Put in public facts only: your trade, the towns you serve, your tone ("plain, short, no sales talk"), and spelling preferences. Leave out anything you wouldn't print on a flyer.
5. Write a one-line rule for staff. Something like: "No customer names, addresses, phone numbers, card numbers, gate codes or employee records go into any AI chat." Post it where the office can see it.

Which jobs should you hand it first?

Start with low-risk writing where a mistake is easy to spot and cheap to fix. Use it for a few weeks before trusting it with anything customers see without a careful read.

If you want ready-made wording for these requests, our AI prompts for contractors guide has copy-paste versions.

How do you ask so the answer is useful?

Most bad output comes from thin requests. "Write a message about the delay" produces a generic apology. A useful request has five parts:

• Who you are. "I own a four-person fence company."
• The situation. "Our crew is two days behind because of rain. This customer's job was set for Thursday."
• The facts it must use. "New date is Monday. Posts are already set. No change to price."
• The format. "A text under 300 characters."
• The limits. "Don't apologize more than once. Don't offer a discount."

Then treat the first answer as round one. Tell it what's wrong ("too formal, and drop the second sentence") and let it revise. Two or three rounds usually get you something you'd send. If it keeps drifting, start a fresh chat and put the facts and limits at the top.

Ask it to flag what it doesn't know. Adding "If you're missing information you need, ask me instead of guessing" cuts down on invented details. It won't stop them entirely.

What should you check before trusting the output?

Read every draft as if a skeptical customer will forward it to a lawyer. Specifically:

• Numbers. Prices, quantities, measurements, dates, times. If a number appears that you didn't supply, find out where it came from or delete it.
• Promises. Arrival windows, warranty lengths, "guaranteed" anything. The assistant likes to be helpful and will promise things you never offered.
• Rules and codes. Building code sections, permit rules, tax treatment, licensing. These vary by state and town. Check the actual rule with the authority that enforces it.
• Names and specifics. Product names, model numbers, manufacturer warranty terms. Confirm them on the manufacturer's own documentation.
• Claims about your company. If a draft says you're "licensed," "certified" or "the most experienced" anything, make sure it's true and you can prove it. In a 2024 enforcement announcement, the FTC made clear that existing consumer protection law applies to AI use. A false claim is yours whether you typed it or a chatbot did.

Where should ChatGPT not make the call?

Pricing. It can format a quote and explain line items. It cannot know your labor rate, your overhead, your material costs this week, or how tough the access is on that lot. Your price comes from your own numbers. The estimating and quotes guide shows which parts AI can touch.

Safety. Whether a roof is safe to walk, whether a panel needs to be shut off, whether a tree is a hazard to the house: those are judgments for a trained person on site. An assistant can help you write a safety briefing from your own rules. It should not write the rules.

Legal and tax language. Contract terms, lien notices, employee classification, sales tax on services. Use the assistant to understand the vocabulary before you meet your attorney or accountant, then let them decide. This page is general information too, and it is not legal or tax advice.

Hiring and firing. It can draft a job ad. Decisions about people need a person who knows them and the law that applies.

How does it fit with the rest of your systems?

ChatGPT works on whatever you paste into it. It doesn't see your job history, your calendar or your call log unless someone copies those in, which is exactly where the privacy problems start. The longer-term answer for most shops is to keep customer records in a system you control and use AI features inside that system, under rules you set. A custom CRM can hold your price book, job notes and customer history so that drafts start from real data without anyone copying it into a consumer chat app.

Phones are a separate question. A chat assistant can't answer your business line. If missed calls are the bigger problem, that is a job for call routing, a missed-call text and possibly an AI receptionist set up with your own rules. Our AI tools for contractors list covers where each type of tool fits.

A reasonable first month looks like this: one company account with training settings decided, a written rule about customer data, and daily use for scopes, emails and policies. After that month you'll know which tasks it saves you real time on and which ones you'd rather keep doing by hand.

Frequently asked questions