Prompt Engineering
Prompt engineering is the practice of writing instructions, examples, and context for an AI language model so it produces reliable output, such as telling it exactly how to classify exclusions in a construction subcontractor proposal.
Why it matters in construction
The same model can give a vague answer to “summarize this proposal” or produce a more useful leveling table when it has clear instructions, a required format, and rules for missing information. The prompt and the documents supplied to the model both affect the result.
For a GC, this matters in two ways. Anyone using a general assistant for precon work can improve the output with a few habits. Anyone evaluating purpose-built software should know that the vendor’s prompt design shapes the product, even when it is invisible.
How it works
A prompt is the full set of text sent to the model, and the useful parts are predictable.
- Role and task. “You are a senior estimator reviewing a plumbing proposal against the attached scope of work.” This narrows what the model pays attention to.
- Context. The actual documents: the proposal, the bid package scope list, the relevant spec sections. Without these the model works from general knowledge and guesses.
- Output format. A table with fixed columns, JSON with named fields, or a bulleted list with a required structure. Fixed formats make output checkable and let software consume it.
- Rules for uncertainty. “If the proposal does not mention an item, return ‘not stated.’ Never infer an inclusion.” This instruction can reduce unsupported assumptions.
- Examples. One or two worked examples of a correctly classified line item teach the model the firm’s conventions faster than a paragraph of description.
- Testing. Run the prompt against proposals where the answers are known, count the misses, and revise. A prompt gets iterated like any other estimating template.
Good vendors version their prompts, test them against labeled document sets, and change them carefully. A small wording change can change the output enough to require retesting.
Example in practice
Consider a precon coordinator at a mid-size GC using a general AI assistant to summarize RFPs before go/no-go. Her first prompt is “summarize this RFP.” She gets three paragraphs of overview and no dates.
She rewrites it: “You are a preconstruction manager. From the attached RFP, return a table with these rows: bid due date and time, pre-bid meeting date, bond requirements, insurance limits, liquidated damages, MBE/WBE goals, and any phasing or occupied-building constraints. For each row, quote the passage and page number. If a row is not addressed, write ‘not stated.’”
The revised prompt returns the 2:00 PM deadline, the mandatory pre-bid on the 8th, a $2,500 per day LD clause on page 41, and a 25 percent MBE goal she had not noticed. The model and document are unchanged; the instructions are more specific.
Frequently asked questions
Do I need to learn prompt engineering to use AI in preconstruction?
For a general chat assistant, a little goes a long way. For purpose-built precon software, the vendor has already done the prompt work, and your job is to judge whether the output is right, not to craft the instructions.
What makes a construction prompt good?
Specific role, specific task, specific format, and the actual source document. Tell the model it is an estimator checking a Division 23 proposal against a scope list, ask for a table with a source quote per row, and instruct it to say not stated when the proposal is silent.
Is prompt engineering a lasting skill or a phase?
The tricks change as models improve, but the underlying skill is writing clear, complete instructions with the right context. That is the same skill used to write a good scope letter.
Go deeper
- From the blog The Pros and Cons of ChatGPT in Construction ChatGPT in construction: a 2026 look at what it does well, where it falls short in preconstruction, and why your data beats the public internet.
- From the blog GPT Models for Contractors: A Plain-English Guide A breakdown of GPT-5.6 for general contractors: what Sol, Terra, and Luna do, how effort levels work, and which model fits each precon task.
- From the blog AI for Estimators: 5 Practical Uses that Work Today (Not Someday) Construction's drowning in AI hype, but here are five use cases AI can actually do for estimators today, if it has access to your preconstruction data instead of just whatever's on the internet.