Artificial intelligence in building inspection: what it really does, and what it will not replace

A building inspector signs a document that engages their professional liability. That is the starting point of any discussion about AI in inspection: the tool can prepare, propose and check; it can neither observe for you, nor decide, nor sign. Once that frame is set, AI becomes very useful — provided you know which one you are buying.

The five concrete uses of artificial intelligence in inspection software

1. Structured dictation

Classic dictation transcribes: your voice becomes raw text you then have to split, classify and reword. Structured dictation understands: "basement, north foundation wall, vertical crack about two millimetres, no visible infiltration, monitor" becomes a classified observation — system, component, location, condition, recommendation — with the photo attached. It is the use that returns the most hours, because it moves the writing from the evening to the moment of the visit. Demo test: dictate a natural sentence and watch what the software does with it without intervention.

2. Photo analysis

On demand, the AI looks at a photo and proposes a probable component, a potential visible anomaly and a draft finding. Useful to go faster, dangerous if applied without looking: the AI only sees what is in the frame, does not know the building's history and can be confidently wrong. The draft must stay a draft until you validate it.

3. Nameplate reading (OCR)

Water heater, electrical panel, furnace, heat pump: the nameplate is photographed, model and serial numbers are read and filed. A modest gain per plate, a considerable one over a year. A detail that matters for Law 25: the reading can happen on the device, without sending the photo to a third party.

4. Review before signature

A 40-to-80-page report is rarely re-read in full. An AI review looks for what a tired eye misses: contradictions between two observations, empty sections, orphan photos, missing recommendations, forgotten limitations, a summary to propose. It is the use that protects the most — provided it stays recommended, not blocking: you read the list, fix what you judge relevant, and sign.

5. Assisted writing

Rewording an observation, proposing a clearer version for the client, translating a finding for an anglophone buyer. Handy, but it is the most sensitive use: every signed word must remain yours. An AI that "writes the report" from three photos and a checklist produces a document that binds you without your having thought it through.

What artificial intelligence will not replace

• Observation. AI does not see what you did not photograph, does not smell the humidity, does not hear the compressor struggling, does not lift the hatch.
• Context. The building's age, past renovations, what the seller said, what the buyer wants to know: none of that is in the photo.
• Judgment. Telling a cosmetic defect from a major one, deciding on a recommendation, choosing a limitation: that is the inspector, and the inspector alone.
• Liability. The report is signed by a professional. No software carries that signature, and none can promise an error-free report. Be wary of those that do.

Adopting artificial intelligence without taking risks

Four simple rules, whatever the software:

1. Systematic validation. Every AI proposal — observation, finding, rewording — is applied only after your confirmation. If the software writes directly into the report, change the setting or the software.
2. Traceability. Who proposed what, who validated, when? An audit log protects you the day a file is disputed.
3. Data. Where do your dictations and photos go? To which model provider, in which country, under which contractual terms? Are they used to train models? Law 25 requires you to know. Demand written answers and read the privacy policy, not the marketing page.
4. Offline. In basements or rural areas, the network drops. Good software keeps capturing and resumes analyses when the signal returns, losing nothing.

How DomiSpect applies these principles

DomiSpect is AI building inspection software built in Quebec around one rule: AI prepares, the inspector decides. Domi turns dictation into structured proposals to apply; Domi Vision analyzes a photo on demand and proposes a draft finding; nameplate reading happens on the device; Review AI reviews the full report before signature, recommended and non-blocking. Every proposal waits for your validation, and the offline mode captures everything even without a network. The model providers used and their contractual framework are described in our privacy policy, which we complete as each element is documented.

Frequently asked questions

Can AI write an inspection report for me?

It can structure your observations, propose findings and review the report. It must not write a report you have not thought through: the document engages your professional liability, and every signed word must remain yours.

Is an inspection report produced with AI assistance valid?

A report is worth the inspector who performed, checked and signed it, not the writing tool. AI is a preparation and verification tool, like a template or a checklist. What matters is that you validated every observation.

Are my photos used to train the AI?

It depends on each vendor and its model providers. Ask in writing and check the privacy policy — it is a Law 25 obligation for you, and an answer any serious vendor must be able to give.

Does AI work without a network?

Rarely in full. Heavy analyses (structured dictation, photo analysis) generally need a connection; what matters is that capture continues offline and analyses resume on their own when the signal returns. Nameplate reading, for its part, can happen on the device.

← Back to the blog · The guide to inspection software in Quebec