AI in building inspections, Mid-2026

Vendor-published figures for AI defect detection from imagery cluster around 90 to 95 per cent on visually distinct defect classes such as water staining and roof membrane damage, and drop substantially on subtle, internal, or occluded defects. No independent benchmark confirms that range at the precision vendors quote it; it is a consistent pattern across vendor marketing rather than a single peer-reviewed source, which is itself worth knowing before repeating the figure to a client. Either way, the distribution matters because the weak category is precisely where inspection liability lives. The technology is strongest at the defects an inspector would have found easily and weakest at the defects an inspector might have missed. The exposure profile is the opposite of what the marketing implies.

That mismatch, more than any other single fact, shapes how AI usefully fits into a building inspection practice in 2026. The other shifts of the past twelve months are real and worth understanding, especially in voice-to-structured-record capture on site and in firm-level knowledge synthesis across accumulated reports. But the defect detection accuracy distribution is the load-bearing fact that determines which AI capabilities can substitute for the inspector’s eye and which ones cannot.

This is for principals of building inspection firms and the senior inspectors who set the operational tone inside them. The deeper evidence base for everything below is in the State of AI in Mid-2026 literature review, which surveys ten capability categories across applied AI and went through three independent fact-check passes before publication. This piece pulls the threads that matter to building inspection specifically.

What’s stable in building inspection

An inspector attends a property, examines it against a defined scope, photographs what matters, applies professional judgement about what is visible and what is not, and produces a report on which a purchasing or maintenance decision will rest. That work is not going to change in the next eighteen months. The report is read by clients who do not have the inspector’s training, by conveyancers and solicitors who assess its commercial implications, and occasionally by tribunals when the matter goes wrong. The inspector’s name is on the document and the professional indemnity attaches to that name.

The liability profile is stable too. Inspection liability concentrates around the defects that were present, were within scope, and were missed. The defects an inspector could not have reasonably seen (concealed, internal, occluded, behind cabinetry, under coverings) sit on one side of the standard scope. The defects that should have been visible to a competent inspector sit on the other. The professional and legal distinction between those two categories is one of the load-bearing features of the work, and the place where the published accuracy distribution of defect detection matters most.

The shape of the client relationship is also stable. The inspector is engaged at a moment of urgency, often within a short pre-settlement window, and the engagement frequently includes a tail of follow-up questions as the client moves into the property and discovers things. The post-inspection period is part of the value proposition, not separate from it; for some firms it is twelve months of advice included in the original engagement.

What’s changing

Defect detection from imagery is now a real category, with an accuracy pattern that should be understood precisely. Vendor case studies converge on roughly 90 to 95 per cent accuracy for visually distinct defect classes such as water staining and roof membrane damage, though no independent benchmark has confirmed that range at the precision claimed. The same systems perform poorly on subtle, internal, or occluded defects. As noted above, that distribution matters because the second category is precisely where inspection liability lives. A defect detection product that is excellent at the obvious cases and weak at the hard cases is not, on balance, useful as a substitute for an inspector’s eye. It is potentially useful as a second pass on photographs already gathered for other reasons, with the inspector’s professional judgement still authoritative on the final report.

Vision against standards is the emerging category that matters most, and the evidence is still thin. Checking visual material against reference documents, photographs against specifications, plans against codes, has moved from theoretical to demonstrated in the past eighteen months. Several products now exist for automated building plan review against US building codes: CodeComply.AI, CivCheck (acquired by Clariti in October 2025), and PlanCheckPro.AI. Florida legislated in 2025 to permit software-based plan review, via House Bill 683, effective 1 July 2025. No independent accuracy evaluation of any of these products has been published, and the regulatory pathway in Australia has not opened. The capability is real enough to watch. It is not yet evidenced enough to depend on for any inspection that carries professional liability.

Construction progress tracking against BIM models is in real production use on large projects. Buildots and OpenSpace ship products that compare site photography against the design model and flag deviations. These are commercial deployments with vendor case studies and no published independent accuracy figures. The technology is more relevant to inspection firms that work on the construction side, large-project dilapidation, and progress certification than to the pre-purchase inspector working on existing residential stock. It is worth knowing about. It is not the central frontier for most practices.

Aerial property measurement has matured commercially. EagleView and similar platforms now offer measured roof and exterior data at scale. EagleView’s own claim, confirmed against its published release, is 98.77 per cent accuracy on roof line measurement, benchmarked by an independent LiDAR survey it commissioned from CompassData in Denver in 2025. That is a named methodology, commissioned by the vendor, which is a step up from an unnamed internal benchmark but still short of independent academic replication. For practices doing repeat work across roof condition and exterior surfaces, this is a tooling category that has crossed from novelty into usable.

The on-site documentation capability has shifted substantially. Voice transcription and structured field-record capture have moved from clunky to competent in the past year. The platforms are mature, sub-second latency is routine, per-minute costs are low enough that the cost is no longer the constraint, and the leading systems can capture an inspector’s narration in the field and produce a structured draft that the inspector edits rather than types from scratch. The closest validated analogue from the broader literature is clinical AI scribes, where a multi-site study across five academic health systems found around 16 minutes saved per 8 hours of patient care among adopters, with only about a third of clinicians using the tool often enough to see the largest gains; the error profile is dominated by omission rather than fabrication. Both of those findings translate. Voice-to-structured-record is genuinely useful inside the inspector’s workflow, and the verification step is non-negotiable.

Knowledge synthesis at the firm level is now technically feasible in a way it was not a year ago. Current frontier models can hold an organisation’s accumulated material in working memory at a scale that meaningfully changes what a junior inspector can do with a question. A firm with a hundred and seventy-five reviews, several years of completed inspection reports, and twelve months of post-inspection client questions on every job has a corpus that is uniquely valuable. The same questions recur across properties of similar age, construction type, and location. The same defects cluster around the same building eras. The same client concerns surface in the same first weeks after settlement. A firm that makes that material legible to its own people, with verified retrieval, has a structural advantage that no off-the-shelf product can replicate, because no off-the-shelf product has access to the firm’s own work.

Where it lands

AI fits the work where the inspector’s eye remains authoritative on what was seen and what it means, and where verification is built into the workflow rather than into individual discretion. Several places fit that shape.

Voice-to-structured-record in the field, where the inspector narrates findings on-site and a draft report is assembled by the time the inspector is back at the office, with the inspector editing rather than re-typing. Photograph categorisation and second-pass review of imagery already captured during the inspection, with the inspector’s judgement authoritative on the final defect list. Internal knowledge retrieval, where the firm’s prior reports, post-inspection client questions, and accumulated commentary on building eras and defect patterns become searchable in plain English by every inspector in the firm. Client communication during the 12-month post-inspection advice window, where routine questions are answered consistently against the firm’s own prior reasoning rather than from memory or from scratch each time. First-cut report drafting from a structured field record, with the inspector editing the language and confirming every load-bearing assertion. Translation of technical defect descriptions into plain language for the client-facing summary, with the inspector verifying the precision has survived.

Across the list, what stays human is the eye on the site, the judgement on what the photographs mean, and the signed name on the report. AI does the assembly. The inspector does the work that the indemnity attaches to.

Where it doesn’t yet land

And several places where the marketing implies a capability the published accuracy distribution does not support.

Any reliance on defect detection systems for the subtle, internal, or occluded defects that produce inspection liability. The published accuracy distribution does not support it, and the failure modes concentrate precisely where the risk does. Anything that would otherwise be the inspector’s eye on the site, replaced by photograph-based review by a system whose training data is opaque. Vision against standards as the basis for any inspection finding in Australia, where the regulatory pathway has not opened and the products have no independent accuracy publication. Any vendor whose published evidence rests on a single secondary source or an unspecified internal benchmark; that is a procurement red flag rather than a technology view.

Anything client-facing that has not been verified against the original report. The dominant failure mode of grounded summarisation across the literature is omission rather than fabrication, and an omitted qualification in a building inspection context, especially during the post-inspection advice window, is not a recoverable error. The lit review walks through the broader pattern of perception-versus-measurement across office AI; people feel faster than the measurements confirm, and that gap recurs here as it does everywhere.

What this means for the operator’s posture

The posture that ages well in inspection work treats the firm’s own material as the asset most worth protecting. Treat AI as a layer that thickens the firm’s accumulated experience and accelerates the assembly of inputs, not as a layer that produces inspection findings without verification. Keep the inspector’s name on every report and the professional judgement on every load-bearing assertion, and keep the workflow consistent with that. Treat the firm’s own material (completed reports, post-inspection client questions, accumulated knowledge of building eras and defect patterns) as the asset most worth making legible, before reaching for any off-the-shelf product.

Be specific, internally, about which tools are in use, on what kinds of jobs, with what verification routine attached. Build the verification routine into the workflow, not into the individual inspector’s discretion. A firm that can describe what AI does and does not do in its inspection process, in plain language to a client or to a tribunal, is a firm whose work will age well. A firm that cannot is taking on a kind of risk that is hard to measure and easy to underestimate.

The frontier in inspection AI is moving quickly enough that a senior practitioner does not need to be the first to adopt anything. It is moving slowly enough that anyone who waits eighteen months without forming a view will be visibly behind a peer who spent those eighteen months running small structured pilots and learning which capabilities to trust on which kinds of properties.

The question that helps a practice settle into this is not “which AI tool should we license.” It is “what does our accumulated inspection experience look like once it is legible to every inspector in the firm, what is the verification routine around any AI-assisted output, and how would we describe both to a tribunal if it ever came to that.” We work through that with practices that have decided it is time to think about it deliberately, as part of our trades and construction work.