AI in property valuation, Mid-2026
AVMs are reliable enough for triage, not for the final word on contested property. What has shifted in valuation work by mid-2026, and what has not.
A valuer’s report is a document. A court might one day read it. The valuer’s name is at the bottom of it. That structure shapes which AI capabilities can usefully live inside the work, and which ones produce risk that the professional indemnity policy is the eventual collector of.
The twelve months to June 2026 have produced material movement in three directions worth being precise about. Automated valuation models now anchor at honest accuracy figures rather than the inflated ones that SEO content still circulates. Frontier models can hold a firm’s accumulated valuation experience in working memory at a scale they could not a year ago. And the RICS standard that became mandatory for regulated surveyors on 9 March 2026 has drawn a line under AI governance generally, requiring risk-based oversight and disclosure rather than naming specific use cases as high-risk. The marketing around all of it has moved further still.
This is for senior valuers and the firms that employ 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 valuation specifically.
What’s stable in valuation
A valuer inspects, assesses against the relevant standard, draws on comparable evidence, applies professional judgement, and produces a document for which they remain personally responsible. That work is not going to change in the next eighteen months. The document is read by a lender, an owner, a court, or an opposing expert; the reader’s reliance on it depends on the valuer’s name and credentials, not on any tool that contributed to its production.
The regulatory frame has tightened rather than loosened. RICS made its responsible-AI standard mandatory for regulated surveyors on 9 March 2026. The standard itself does not single out automated valuation models or defect detection by name; it requires risk-based governance, client disclosure of material AI use, and clear human accountability for the outcome, for any AI application a regulated surveyor relies on. In Australia, the API and the state property regulators have not yet matched RICS with a binding professional standard, but the direction of travel is visible: regulated AI use, mandatory transparency about what the technology did and did not do, and clear human accountability for the final figure. Anyone preparing for the next five years should assume the Australian floor will rise to meet the RICS one, and shape their practice accordingly.
The product at the core of the work remains a document. That fact is what determines which AI capabilities are interesting and which are noise.
What’s changing
Automated valuation models, narrowly defined. CoreLogic’s own account, given by a company spokesperson to trade media rather than in a formal published report, puts almost 90 per cent of its Australian AVM estimates within 15 per cent of sale price since early 2024. That is a vendor figure rather than an independently audited one, but it is the most credible accuracy claim in the Australian market and a useful anchor for honest expectations. PropTrack covers over 12 million properties with its own AVM machinery. There is no credible peer-reviewed benchmark of Australian AVM accuracy as of mid-2026, and a widely circulated figure of “94.2 per cent within 5 per cent” could not be traced to any credible source, vendor or independent, and should be treated as marketing rather than evidence.
The honest interpretation is that AVMs are now reliable enough to be useful as a triage and sanity-check layer, and not yet reliable enough to be the final word on any contested property. For high-volume mortgage origination on standard suburban stock, the case is strong. For anything with thin comparables, unusual features, or contested value, the AVM is one input among several, weighted by a valuer.
Image-augmented valuation. A small body of academic work suggests that adding photograph analysis to comparable-sales modelling produces materially better accuracy on certain property types. The headline accuracy claims for these systems in vendor marketing rest on thin evidence. The underlying capability, vision models that can read property photographs as structured features, is real. The question is whether the marginal accuracy gain in any specific market segment justifies the model risk, and that is a procurement question for a regulated firm, not a technology question.
Vision against standards. Checking visual material against reference documents, photographs against specifications, plans against codes, is the emerging category that matters most to valuation-adjacent work, especially where the engagement crosses into defect identification or compliance verification. 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 rely on.
Defect detection from imagery. This is where the marketing diverges most sharply from the evidence. 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 confirms that range at the precision claimed. The same systems perform poorly on subtle, internal, or occluded defects. That distribution matters because it is precisely the subtle, internal, and occluded defects that produce the liability tail in inspection work. A defect detection product that is excellent at the easy cases and weak at the hard ones is not, on balance, useful as a substitute for trained eyes; it is potentially useful as a second pass on photographs already gathered for other reasons.
Knowledge synthesis at the firm level. The most underrated shift in the past year, and the one most relevant to a senior-heavy firm like a valuation practice, is that current frontier models can hold an organisation’s accumulated material in working memory in a way they could not a year ago. Million-token context windows are now standard across the leading providers. A firm with twenty years of valuation reports, expert witness statements, methodology notes, and internal commentary on edge cases can now build a tool that lets the next junior valuer ask the firm’s accumulated experience a question and receive a grounded answer with citations into the source documents.
The accuracy ceiling on this kind of grounded summarisation is now reasonable rather than transformational. The best independent evidence still shows leading systems hallucinating in roughly 13 per cent of responses on this task. That means the answer must be verified before it is used, which is the same standard a valuer would apply to any other intermediate input. The value is not that the tool replaces the senior valuer’s judgement. The value is that the firm’s accumulated judgement becomes legible to every other valuer in the firm, on demand, at the point of need.
Where it lands
AI lands defensibly inside the practice when it accelerates inputs and drafts and does not author the signed report. Several places fit that shape.
Comparable-sales retrieval and triage, with the valuer reading the underlying evidence. Document generation from a structured field record, with the valuer dictating findings rather than typing them, and the model producing a draft that the valuer edits rather than approves wholesale. Internal knowledge retrieval, where the firm’s prior reports, methodology notes, and expert witness materials become searchable in plain English. Client communication, where routine progress updates and document requests are templated and consistent. Plan-against-standard checking as a second pass for the valuer’s own eye, not as a substitute for it.
Across all of them the pattern holds. The AI assembles inputs and drafts. The valuer’s name sits at the bottom of the document, the court reads a human signature, and the opposing expert tests the comparable evidence as they always have.
Where it doesn’t yet land
And several places where the marketing has run ahead of what the evidence supports.
Fully automated valuation as the final word on any contested property. The accuracy distribution does not support it, and the RICS standard now explicitly flags it. Defect detection as a substitute for a trained inspector, especially for the subtle and internal defects that produce inspection liability. Court-facing material produced by AI and not verified line by line by the expert whose name appears on it; the legal profession has now learned this lesson publicly enough that the valuation profession has the chance to learn it privately. Anything from a 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.
It is also worth being specific about what the largest controlled evaluation of office AI to date actually found. The UK Government Digital Service’s three-month cross-government evaluation of Microsoft 365 Copilot across about 20,000 civil servants reported high user satisfaction and self-reported time savings of around 26 minutes per day, but the evaluation was not a randomised controlled trial and did not establish productivity gains under controlled measurement. Users felt faster. The measurements were more equivocal. The same pattern recurs across most categories of office AI. Anyone planning practice-level investment on the basis of “our people feel faster” should sit with that finding for a moment before committing.
What this means for the operator’s posture
The posture that ages well here is methodological more than technical. Treat AI as a layer that thickens the firm’s accumulated experience and accelerates the assembly of inputs, not as a layer that replaces professional judgement on the output. Keep the valuer’s name at the bottom of every document, and keep the engineering choices consistent with that. Prefer tools that make the firm’s own material more useful to its own people over tools that promise to replace any part of the work a regulator or a court has identified as professional. Treat any vendor accuracy claim that traces to a single secondary source as marketing.
Be specific about what is being adopted, why, and what the verification routine is around it. The firms that come through the next three years in the strongest competitive position will be the ones that built that routine early, made it visible to clients and regulators, and used it as a mark of seriousness rather than a constraint to be minimised.
The frontier 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 behind a peer who spent those eighteen months running small structured pilots inside a regulated practice and learning what to trust.
The honest question for a senior practitioner now is not “which tool should we buy.” It is “what does our accumulated valuation experience look like once it is legible to everyone in the firm, and which of the next twelve engagements would benefit from that being true.” That is the question we work through with practices that have decided it is time to form a view, as part of our professional services work.