The Construction AI Prompt Library: 40 Prompts That Work
Forty copy-paste AI prompts for estimators, project managers, and construction admin teams. Spec review, RFIs, contracts, job costing, and system cleanup.
Most contractors who try AI give up because they used it the way you would use a search engine. You ask a vague question, you get a vague answer, you conclude it is overhyped.
The prompts below are the opposite of vague. They are the ones that actually survive contact with a construction week: reading a spec section, comparing three quotes that arrived in three different formats, drafting an RFI from a field note, finding which jobs are quietly running over budget.
Copy any of them from the board below, or take the whole set as a file at the bottom of the page.
How to use these
Replace the brackets. Anything in [BRACKETS] is where your content goes.
Change the wording to match your team. A prompt using your terminology, your document formats, and your cost code structure will beat a generic one every time. These are starting points.
Pick three, not forty. The contractors who get value from this are the ones who built three habits, not the ones who bookmarked a library. Choose the three tasks your team does weekly that are mostly reading and drafting, and run only those for a fortnight.
Notice the recurring instruction. A lot of these end with some version of "tell me what you could not find" or "list what this analysis did not cover." That is deliberate, and it is the most important pattern in the whole library. An AI reports what it finds and has no idea what your records are missing. Ask for the gaps and a confident summary becomes something you can audit in ten seconds.
40 prompts shown
Build a scope sheet from a spec section
You are assisting a construction estimator. Read the specification section below and produce a scope sheet as a table with columns: Scope Item, Included, Excluded, Clarification Needed. Quote the spec language that supports each row. Do not infer scope that is not written. Where the spec is ambiguous, say so explicitly rather than choosing an interpretation. [PASTE SPEC SECTION]
Why it works: The "do not infer" instruction matters. Without it the model fills gaps with plausible assumptions.
Normalize vendor quotes into one table
Compare the subcontractor quotes below. Produce a single table with one row per scope item and one column per vendor, so I can see coverage side by side. Add a final section listing: items only one vendor included, items nobody included, and any exclusions or qualifications that materially change the price. Do not calculate a recommended winner. [PASTE QUOTES]
Why it works: Asking for gaps is the point. The pricing comparison is easy; the coverage gaps are what cost you.
Draft qualifications and exclusions
Based on the scope documents below and our included scope, draft a qualifications and exclusions page for our proposal. Group into: Clarifications, Exclusions, Assumptions, and Allowances. Keep each item to one sentence in plain language. Flag anything you are unsure should be excluded rather than silently including it. Scope documents: [PASTE] What we are including: [PASTE]
Compare an addendum against original scope
Compare the addendum below against the original documents. List every change as: what changed, where it appears, and whether it is likely to increase cost, decrease cost, or is neutral. Separate changes that affect our scope from changes that affect other trades. Do not estimate dollar amounts. Original: [PASTE] Addendum: [PASTE]
Pre-submission bid review
Act as a second set of eyes before we submit. Review the bid package below against the ITB requirements and list: missing required forms, unaddressed addenda, scope items in the ITB that do not appear in our proposal, and any inconsistency between our cover letter, pricing, and exclusions. Be blunt.
Why it works: One of the highest-value uses. Cheap to run, catches the kind of miss that loses a bid on a technicality.
Extract a submittal register
Read the specification sections below and build a submittal register as a table: Spec Section, Submittal Item, Type (product data, shop drawing, sample, certificate, O&M), and Responsible Trade. Include only submittals the spec actually requires. Cite the paragraph number for each row.
Why it works: Cite-the-paragraph is what makes this auditable in two minutes instead of re-reading the spec.
Pull testing and inspection obligations
From the specifications below, list every testing, inspection, and special inspection obligation. For each: what is tested, who pays, who performs it, when it must happen relative to the work, and the spec reference. Flag any where responsibility is ambiguous.
Reconcile drawing notes against the spec
I am giving you general notes from the drawings and the relevant specification section. Identify every place they conflict, every place the drawings add a requirement the spec does not contain, and every place the spec adds one the drawings do not show. Present as a table with the conflicting language quoted from both sources.
Why it works: Drawing-versus-spec conflicts are where change orders are born. This finds them before you price the job.
Assess a product substitution request
Compare the specified product against the proposed substitute below. Produce a table of the performance criteria the spec requires, what each product provides, and whether the substitute meets, exceeds, or falls short. List what additional information we would need to approve it. Do not make the approval decision.
Summarize a division for the field
Summarize the specification section below for a foreman who will not read the full document. One page maximum. Cover: what has to be installed, the quality standards that will be inspected, the submittals that gate the work, and the three things most likely to cause a rejection. Plain language, no spec jargon.
First-pass subcontract risk review
Review the subcontract below and flag provisions that shift risk to us. Specifically address: indemnity scope, consequential damages, pay-if-paid or pay-when-paid, notice periods and their length, termination for convenience, retainage terms, liquidated damages, change order process, and dispute resolution. For each, quote the clause and state plainly what it means for us in practice.
Why it works: A first pass, not legal advice. It tells you which clauses deserve your attorney time.
Build a notice deadline list
From the contract below, extract every obligation that has a deadline attached: claim notices, change notices, delay notices, payment applications, lien deadlines, and closeout submissions. Present as a table: Obligation, Deadline, Trigger Event, Contract Reference, Consequence of Missing It.
Why it works: Missed notice periods are one of the most expensive unforced errors in construction. This is ten minutes well spent.
Diff two contract versions
Compare these two versions of the same agreement. List every substantive change, ignoring formatting and numbering. For each change, quote the old and new language and state which party it favors. Highlight anything that changed in the indemnity, payment, or termination provisions first.
Check insurance requirements
Extract all insurance and bonding requirements from the contract below: coverage types, limits, additional insured obligations, waiver of subrogation, primary and non-contributory language, certificate requirements, and how long coverage must be maintained. Then list what we would need to confirm with our broker.
Translate a clause into plain English
Explain the clause below in plain English to someone who runs projects but is not a lawyer. Then give two concrete scenarios: one where this clause costs us money, and one where it protects us. Do not soften it. [PASTE CLAUSE]
Draft an RFI from a field question
Turn the field note below into a formal RFI. Structure it as: Background, Specific Question, Reference Documents, Proposed Solution (if we have one), and Schedule or Cost Impact if unanswered by a date. Keep it to one question. Neutral professional tone, no argument. Field note: [PASTE]
Why it works: One question per RFI. The model will happily bundle three, and bundled RFIs come back half-answered.
Summarize an open RFI log
From the RFI log below, produce: items open more than 14 days, items whose answer blocks work scheduled in the next three weeks, items with no assigned responder, and a suggested escalation list ranked by schedule impact. Present as short tables.
Write a follow-up on a stale RFI
Write a short, polite follow-up on an RFI that has gone unanswered. State the RFI number and date, restate the question in one sentence, note the work it is holding and the date that becomes a problem, and request a response by a specific date. Four sentences maximum. Firm but not adversarial.
Check a submittal against the spec
Compare the submittal below against the specification requirements. Produce a table: Spec Requirement, What the Submittal Provides, Complies / Does Not Comply / Cannot Tell. List exactly what additional information would resolve each "cannot tell". Do not approve or reject.
Draft a submittal transmittal
Write a transmittal for the submittal package described below. Include the spec section, submittal number, contents list, any deviations from the specified product clearly called out, and the review time we are requesting per the contract. Professional and brief.
Find jobs heading over budget
Using our live job data, list every open job where actual cost is tracking above projected cost. For each: job name, projected cost, actual cost, variance in dollars and percent, and which cost groups are driving it. Then list every job you checked where the cost data was missing or incomplete, so I know what this analysis did not cover.
Why it works: The second half is the important half. Without it you cannot tell a clean result from a thin one.
What did we charge last time
Search our completed jobs for work matching [DESCRIBE SCOPE]. For each match, give the job name, date completed, the relevant cost items with unit and quantity, what we charged, and what it actually cost us. Note how many jobs you found so I know the sample size.
Why it works: Needs a connected system. This one question is often the whole reason to set up the connector.
Prepare WIP schedule inputs
For every open job, pull: contract value, approved change orders, total projected cost, cost to date, and billings to date. Present as a table sorted by contract value. Flag any job where cost to date exceeds percent complete billing, and any job missing one of these figures.
Find which cost groups keep running over
Across all jobs closed in the last 12 months, identify which cost groups most frequently exceeded budget. For each: how many jobs it ran over on, average overrun percent, and the largest single overrun. Rank by total dollars lost, not by percentage.
Why it works: Ranking by dollars rather than percent stops small line items dominating the list.
Draft a monthly owner report
Using the job data, draft this month's owner report for [JOB]. Cover: work completed this period, work planned next period, current schedule status against baseline, open RFIs and submittals affecting progress, approved and pending change orders, and any issues requiring owner decision. Factual tone, no sales language. Flag anything you could not verify from the data.
Write a change order narrative
Write the narrative for a change order. Explain what changed, why it is outside the original scope with reference to the contract documents, what the impact is on cost and schedule, and what we need from the owner to proceed. Factual and neutral. Do not argue or apologize. Details: [PASTE]
Draft a delay notification
Draft a delay notice per the contract notice requirements. State the delay event, the date it occurred, the contract clause requiring notice, the anticipated impact on the critical path, and that we will follow with a time impact analysis. Preserve our rights without assigning blame.
Why it works: Have your PM or counsel check this against the actual notice clause before it goes out.
Weekly client update from project activity
From the daily logs, task updates, and open items below, write a weekly client update. Lead with progress, then what is coming, then anything we need from them, with dates. Four short paragraphs. Do not include internal problems the client does not need to see, but do not hide anything that affects their schedule or cost.
Chase a subcontractor without burning the relationship
Write a short follow-up to a subcontractor who has not responded about [ISSUE] for [X] days. Reference the previous message, state what we need and by when, and explain the consequence for their scope if it slips. Direct but collegial. Three sentences.
Turn meeting notes into an action register
Convert the meeting notes below into an action register: Action, Owner, Due Date, Status. Then list separately any decision that was made, and any item discussed where no owner was assigned, because those are the ones that get dropped. [PASTE NOTES]
Build a three-week look-ahead
From the schedule data below, build a three-week look-ahead. For each activity: description, responsible trade, start and finish, and any predecessor that is not yet complete. Flag activities that cannot start because of an open RFI, a pending submittal, or an undelivered material.
Explain schedule variance in plain language
Compare the current schedule against the baseline below. Write a short narrative explaining what slipped, by how much, what caused it where the data shows a cause, and whether the critical path changed. Write it for an owner, not a scheduler. Say clearly where the data does not explain a variance.
Generate progress-input requests
From the activity list below, produce a per-trade message asking each trade for the percent complete and any constraints on their activities in the next two weeks. One short message per trade, listing only their activities.
Why it works: Turns one scheduler chasing eight trades into eight messages drafted in a minute.
Extract a constraint log
Read the meeting minutes and daily logs below and extract every constraint mentioned: missing information, undelivered material, access restriction, permit or inspection dependency, or unresolved design question. For each: what it blocks, who owns resolving it, and whether a date was committed.
Outline a time impact analysis
Outline a time impact analysis for the delay event described below. Cover: the baseline used, the delay event and its documented start, the fragnet to be inserted, the affected activities, and the resulting projected completion. Identify every piece of evidence we would need to support it. Do not draw the conclusion.
Why it works: Outline only. A TIA that goes into a claim needs a scheduler, not a chat window.
Audit the cost catalog
Review our full cost catalog and report: duplicate items under different names, items with inconsistent units for the same type of work, cost groups that are out of sequence, items with no cost set, and items not used on any job in the last two years. Present each as its own table. Do not change anything.
Why it works: One of the best uses of a connected system. Miserable by hand, quick this way.
Find naming inconsistencies
Look across our jobs, cost items, and document templates and identify where naming conventions are inconsistent: abbreviations used in some places and not others, different capitalization patterns, and the same concept named more than one way. Propose one convention and list what would need renaming to reach it.
Find incomplete job records
List every open job missing any of: a populated budget, an assigned project manager, a location, a customer contact, or a start date. Present as a table with one column per missing field so I can see at a glance which jobs need the most work.
Why it works: Run this monthly. It is the difference between data you can query and data you cannot.
Find jobs that have gone quiet
List every open job with no daily log, task update, document, or comment in the last 21 days. For each: job name, last activity date, last activity type, and assigned manager. Sort by longest silent first.
Spot cost and document mismatches
Find jobs where the numbers disagree: documents marked sent with no corresponding payment recorded, purchase orders with no matching cost entry, or actual costs recorded with no supporting document. List each with the job name and what specifically does not reconcile.
Why it works: Finds bookkeeping gaps, not accounting errors. Reconciliation stays with a person.
The two rules that keep this safe
A person reads every output before it reaches a client. This is not a limitation of the technology, it is ordinary professional practice. The first time you skip it you will send a client something fluent and wrong.
Decide what the tool may never see, before someone asks. Anything under an NDA, any employee record, any client confidential document. Make the rule in advance, because in the moment the answer is always that it is probably fine. If you do not have a written position on this yet, that is worth fixing before your team gets further ahead of you than it already is.
The prompts that need your job data connected
About a third of the library, mostly the job costing and system cleanup sections, assumes the assistant can read your live project data rather than what you paste in. Those are the ones that answer questions like "what did we charge last time we did this" in four seconds instead of twenty minutes of digging.
JobTread publishes an official connector for this that works with both Claude and ChatGPT. We wrote the complete setup guide, including the permission tiers that matter and why you should start read-only.
One honest caveat that applies to every data-connected prompt here: they are only as good as the account behind them. We ran a cost query against a live construction account and got empty values back, because those jobs were leads that had never had budgets populated. Completely normal, and exactly how you end up with an average computed over a fraction of your work, presented in the same confident prose as a complete answer. Get the underlying structure right first.
Where to go next
If you want the wider picture of what AI does and does not do for contractors, we mapped it in AI in construction. If your team is on ChatGPT rather than Claude, everything here still applies.
And if you would rather someone set this up properly, including deciding what the tool should and should not be allowed to see, that is what we do.
Get the The Construction AI Prompt Library (TXT)
Enter your email and we will send you all 40 prompts as a plain text file you can keep, search, and paste from. No spam, no sales sequence.

Written by
Kenneth Muriuki
Construction PM & Construction Tech Expert
Kenneth is a construction project manager and construction technology expert who specializes in JobTread implementation, with experience helping contractors improve project delivery, system configuration, reporting, and operational visibility. He writes about practical construction technology, workflow design, and the operating disciplines that help teams deliver work consistently.
More articles
AI for the Construction Back Office: What to Hand Off
AI for construction admin, task by task: AP coding, PO matching, lien waivers, COIs and pay apps. What AI does well, what needs a human, and what stays out.
ReadAI for Construction Project Management: 8 PM Workflows
AI for construction project management, workflow by workflow: RFIs, submittals, minutes, daily reports, change orders, schedules, and where PMs must check.
ReadAI Tools for Contractors: An Honest 2026 Buyer's Guide
The AI tools for contractors worth paying for in 2026, sorted by the job they do: assistants, takeoff, document review, field capture, scheduling, and sales.
Read