AI property inspection report drafting.

What the workflow produces

Inspection report drafting takes the raw material of a routine visit (40 or so photos and five minutes of spoken notes) and returns three drafts: a room-by-room report, a maintenance list, and a short owner summary. Each line points back to the photo or note it came from. The property manager reviews on a tablet before leaving the street or back at the office, edits what is wrong, and signs it off.

The value is consistency as much as time. Reports written at 5pm on a Friday tend to say “tidy” and “no issues”. A draft that compares every room with the ingoing report, and asks about each difference, makes it harder to miss the new stain or the extractor fan that has stopped working.

How the workflow runs

Inputs

Inspection photos tagged by room, the property manager’s voice notes, the ingoing property inspection report and photos, the last routine report, open maintenance jobs, and the property’s healthy homes compliance record.

Systems

The property management system, a phone or tablet capture app, a transcription model, a vision-capable language model, and the report template.

Steps

Transcribe notes; group photos by room; compare each room with the ingoing report; draft observations with photo references; list maintenance items with suggested urgency; draft owner summary and tenant letter; flag anything uncertain.

Human review

The property manager confirms or edits every line, removes photos showing people or belongings, decides follow-up actions, and approves both letters before they are sent.

Outputs

Signed inspection report with dated photos, maintenance items raised as jobs, owner summary and tenant letter.

Measures

Minutes from visit to signed report, share of draft lines edited, maintenance items found at the next inspection that were missed at this one, and owner queries per report.

Risks

Confident descriptions of things not visible, rooms mixed up, fault implied in wording, and private images kept or shared.

Illustrative example: a Dunedin portfolio of 180 homes

Illustrative example, synthetic numbers: a Dunedin property manager inspects 180 homes quarterly, about 60 inspections a month. Each visit takes 20 minutes and the report another 35 minutes at a desk.

Assumptions: transcription with gpt-transcribe at US$0.0045 a minute (OpenAI pricing, as listed on 19 October 2026), about 3 US cents per visit. Drafting with Claude Sonnet 5.5 at US$2 per million input and US$10 per million output tokens (Anthropic pricing, same date), assuming 40 resized photos at roughly 1,600 tokens each plus 8,000 tokens of notes, prior report and instructions (72,000 input tokens) and 2,500 output tokens, about 17 US cents. The saving in this illustration is desk time, and the quality gain is a report that has been compared with the ingoing record every time.

NZ rules for inspections and records

• Frequency and notice. Tenancy Services sets inspections between 8am and 7pm, no more than once every 4 weeks, with at least 48 hours’ notice (and no more than 14 days in advance). The workflow can draft notices, but scheduling stays with the property manager.
• The ingoing report is the baseline. Tenancy Services suggests bringing a camera and the property inspection report completed at the start of the tenancy, and taking photos in case of later disputes.
• Collect only what you need. The Privacy Commissioner’s guidance for landlords says inspection photos should not include people or focus on personal items unless those items breach the tenancy agreement, and should not be reused to sell the property without consent.
• Keep, then dispose. The same guidance notes inspection reports must be kept during the tenancy and for 12 months after it ends. Set retention on the tenancy file, not in the model provider’s systems.
• Healthy homes. Observations about heating, ventilation or moisture link to the healthy homes standards and the compliance statement in the tenancy agreement.

What only the property manager can do

Only the property manager was in the house. They decide what is fair wear and tear, what is damage, what needs a 14-day notice to remedy, and what the owner is told. They also decide which photos are kept. Our lesson on planning human review for AI output covers how to make line-by-line confirmation quick rather than skipped.

Where drafting goes wrong

• Seeing what is not there. Shadows become “mould” and reflections become “water damage”. The prompt should require a photo reference for every defect, and the property manager confirms each one. Our note on OCR versus vision models covers what image models handle well.
• Room mix-ups. Two similar bedrooms swap findings. Tag photos by room at capture, not afterwards.
• Wrong baseline. Comparing against a superseded ingoing report after a renovation produces false differences. Link the current baseline in the property record.
• Tested versus seen. A draft that says “smoke alarms tested” when the property manager only looked at them is a false record. Use separate fields for observed and tested.
• Polite blame. Phrases like “tenant has damaged” pre-judge a dispute. Describe the condition and leave cause to the property manager.

Inspection findings become jobs in maintenance request triage, and the signed reports become searchable through tenancy document search. See the property management sector guide and our AI document processing service for the wider picture.

Written by the Levy Street team in Wellington. Last updated 19 October 2026.