ChatGPT for Construction: Practical Workflows and Prompts

ChatGPT can help construction teams prepare draft reports, organize meeting notes, compare supplied records and write clearer questions. It is most useful when the task is narrow, the inputs are current and a responsible person checks the result. It does not automatically know your live project status or turn an uploaded file into an approved project record.

For commercial owners, developers and GCs, start with work that already has a reviewer: a weekly update, an RFI draft or a list of changes between two reports. The examples below are illustrative, proposed workflows and prompts, not tested customer results or promises of a Jet.Build integration.

Set up the inputs before writing the prompt

Use only a workspace and files approved for your company's information. Confirm the provider's current data-use, access and retention terms rather than assuming every account has the same settings. Remove information that is unnecessary for the task.

Label each input with its project, record type, revision and status date. Supply the specific files needed, not an uncurated folder of conflicting versions. Tell ChatGPT which file controls if two sources differ. If a drawing, table or scanned page is not readable, obtain a usable source rather than accepting an inferred answer.

OpenAI's file documentation describes document summarization, comparison and spreadsheet analysis. Availability and file handling depend on the service and settings. The prompts below define a review process; they do not add access to systems or files you have not supplied.

Prompt 1: Draft a weekly construction report

Inputs: approved daily reports, current open-item log and reporting period. Use the daily report template if the team needs more consistent inputs.

Using only the supplied daily reports and open-item log, draft a weekly project update for the stated reporting period. Separate completed work, open issues, decisions needed and next-period activities. For each material statement, cite the source file and date or record ID. Do not invent progress percentages, causes, owners or due dates. Mark missing information as “not provided.” Finish with a list of items the project manager must verify before distribution.

Review: compare each progress statement with the dated observation. Confirm that an issue described early in the week was not resolved in a later report. Keep unknowns visible. The project manager approves the summary before it reaches an owner or lender.

Prompt 2: Prepare an RFI draft from a documented ambiguity

Inputs: the current specification excerpt, relevant drawing reference and the team's observed question. Do not substitute AI interpretation for the designer's response.

Draft an RFI from the supplied question and references. Include the observed ambiguity, the specific information requested, source document and revision references, and the decision needed. Do not invent a design solution, cost impact or schedule impact. Separate supplied facts from assumptions and list missing references. Return a draft for project-manager review, not an approved RFI response.

Review: verify the references, confirm the answer is not already available and remove leading language that assumes approval. Submit through the established RFI workflow. Track the issued record and response in the RFI log.

Prompt 3: Compare two cost reports

Inputs: two dated reports using the same cost-code basis, a description of each column and any approved changes relevant to the review.

Compare the two supplied cost reports by cost code. List changed values with the prior amount, current amount, difference and source row. Keep approved and pending changes separate. Do not infer why a value changed unless a supplied record supports the explanation. Flag missing codes, duplicate rows, inconsistent units and totals that do not reconcile. Produce a review table and questions for the project controls lead.

Review: check arithmetic and formulas in the source spreadsheet, reconcile totals and confirm status dates. A plausible narrative is not evidence of the cause of a variance. Preserve the reviewed conclusion with the project cost record.

Prompt 4: Turn meeting notes into an action register

Inputs: approved notes or a transcript the team is authorized to use, plus the existing action list.

Extract proposed action items from the supplied meeting notes. Include the action, stated owner, stated due date and source reference. If an owner or date is absent, say “not assigned.” Identify potential duplicates in the existing action list, but do not close an action without an explicit completion statement. Separate decisions made from proposals discussed. Return a draft for the meeting chair to confirm.

Review: the meeting chair confirms wording, owners and dates. Record accepted actions in the team's operating workflow. AI should not quietly turn discussion into a contractual instruction.

Prompt 5: Summarize open items for a look-ahead meeting

Inputs: current RFI, submittal and procurement logs, the look-ahead period and relevant schedule activities.

From the supplied logs, identify open items due within or before the look-ahead period. For each item show its ID, status, due date, responsible party and referenced schedule activity if supplied. Separate confirmed relationships from possible impacts that need review. Do not label an item critical path unless the supplied schedule evidence establishes that relationship. List missing information needed for the coordination meeting.

Review: the scheduler and project team confirm dependencies and field status. An overdue review is a reason to investigate, not proof that the completion date will move.

What to check before using any output

• Can each material statement be traced to a supplied source?
• Are the revision, date, units and approval status correct?
• Are unknowns and contradictions visible?
• Have calculations been independently checked?
• Did the responsible person accept the result?
• Is the accepted result stored with the relevant project record?

If any answer is no, correct the draft before using it. Do not use an AI response alone to approve design, certify payment, authorize a change or direct safety-critical work.

Can ChatGPT estimate construction costs or read drawings?

ChatGPT can help organize supplied scope descriptions or explain the assumptions in an estimating worksheet. It should not be treated as an independent source of current labor rates, local material prices, verified quantities or bid-ready totals. A construction estimate needs an estimator to confirm scope, units, quantities, rates, productivity, waste, exclusions and the commercial basis.

For drawings, verify that the relevant pages, scale, symbols and revisions are readable. Do not assume a PDF upload means every drawing detail was understood. AI-generated counts or measurements require checking against the source. Start with a validated takeoff and estimating workflow, then use an assistant to help organize reviewed information.

Organize the supplied, estimator-approved quantities and rates into a scope review. Preserve units and source references. List missing assumptions, exclusions and inconsistencies. Do not invent prices or quantities, and do not present the result as a bid-ready estimate.

Claude for construction and other AI tools

Claude can also work with documents and context supplied to its project workspaces. The same input and review discipline applies; this article does not claim one assistant is more accurate than another. Construction-specific tools may add project-document processing or workflow features. Compare the task, source access and review process using the AI construction software guide.

Keeping accepted work connected to Jet.Build

A draft generated from exported files is a snapshot. Later edits in the operating system will not appear unless the inputs are refreshed or an approved connection provides them. Record the source date and have a person transfer accepted work through the appropriate workflow.

Jet.Build connects project operations, financials and field information, with Jenny available as its AI assistant. Review the relevant workflow on the platform. If your team wants an external AI or document system connected, define the required records and access through integration planning. This article does not establish a ready-made ChatGPT or Claude integration.

Bring one recurring reporting or review task to a Jet.Build demo, together with the records and approval steps it needs to support. For the wider picture of where AI fits in project workflows, see AI in Construction: Practical Uses, Benefits and Limits.