Back to Guides and resources

Human review

Review an AI construction output before it informs a decision

A source-first review workflow for quantities, prices, schedules and document summaries, with visible uncertainty and human approval.

Original review checklist. It does not certify accuracy, compliance, contractual entitlement or suitability for a particular project.

Separate the source from the generated interpretation

Retain the original document and its revision. Next to the output, record the file or drawing reference, date, units and the question asked. A fluent explanation is still an interpretation: a reviewer needs the underlying evidence to check it.

CAIS describes QTO as experimental and requires independent validation. Treat generated quantities, prices, logic and summaries as working material until the relevant professional has checked them.

Use checks appropriate to the output

Choose checks around the decision you need to make, rather than asking whether the entire output merely looks reasonable.

Suggested review checks by output type
OutputReview focus
QuantitiesDrawing revision, scale, geometry, unit basis, omissions and double counting.
PricingRate date and origin, local scope, inclusions, unit consistency and arithmetic.
ScheduleCalendar, durations, predecessors, sequencing, constraints and resources.
Document summaryExact source passage, conflicting versions, missing context and uncertainty.

Build an exception register

Keep one short record for each issue: the item, source, discrepancy, possible consequence, responsible reviewer and next action. Distinguish a verified correction from an unresolved assumption. Do not hide an unknown behind a more precise-looking number.

Prioritise items that can change cost, duration, payment or safety decisions. A sample check can reveal problems, but it does not prove that unchecked items are correct.

Make approval a separate action

Record which version was reviewed, the changes made and the decisions still open. The person making a project commitment should see those limits alongside the output. A generated report, citation or download does not itself constitute professional approval.

For payment or contract questions, use the actual agreement and authorised project records. An AI summary and an instructional example cannot establish entitlement or replace appropriate professional advice.

Keep the review file proportionate

Use only the documents necessary for the question, confirm permission to share them, and avoid including personal data without a reason. A public article or demo is not a place to publish a client file.

This checklist is a proposed editorial method. The product links below explain current public scope and limitations; they do not promise a specific review result.

Review an AI construction output before it informs a decision