Artificial intelligence is moving into the real estate appraisal profession fast. At seminars, conferences, and professional discussions, appraisers are increasingly being encouraged to bring ChatGPT and other AI tools into their work.
There’s nothing wrong with that on its own. Used well, AI can be a genuinely useful tool for appraisers.
The trouble starts when AI is pitched as something you can simply ask to determine a property’s value — sometimes without even inspecting the property. That proposition deserves closer scrutiny.
AI Does Not “Know” Property Value
Ask an AI system for a property’s value and it will often produce an estimate readily — one that can look remarkably precise.
But where did that number come from?
Market value isn’t information AI inherently possesses. A credible opinion of value has to be supported by relevant market evidence, and AI is only as useful as the information it has to draw on.
This matters most in markets where reliable transaction data is thin. Where actual selling prices, property characteristics, transaction dates, and similar information are publicly available and systematically recorded, automated valuation systems have a lot to work with. Where that information is fragmented, privately held, hard to verify, or simply unavailable, the picture changes completely.
Online listings may be abundant, but an asking price is not a transaction price. A property listed at ₱20 million may eventually sell for ₱16 million. Another may never sell at all. A transaction may involve unusual financing, related parties, package deals, distress, or other conditions that never make it into the listing.
An AI system that doesn’t know any of this can still produce a confident-sounding estimate. That’s exactly where the danger lies.
Precision Is Not the Same as Reliability
Suppose an AI system concludes that a parcel of land is worth ₱18,437 per square meter. The figure looks scientific because it’s precise.
But suppose the data behind it consists mostly of asking prices, duplicated listings, outdated postings, mislocated properties, and transactions whose actual consideration was never verified.
No amount of sophistication in the calculation can make up for weak evidence underneath it. Worth remembering:
Precision of output is not reliability of value.
A sophisticated algorithm run on unreliable information just produces a sophisticated-looking but unreliable conclusion. Garbage in, garbage out didn’t go away because AI showed up.
ChatGPT Is Not an Automated Valuation Model
This distinction gets missed constantly.
ChatGPT and similar generative AI tools are general-purpose systems, built to understand, organize, analyze, and generate information. An Automated Valuation Model (AVM), by contrast, is purpose-built to estimate property values using defined property databases, transaction data, statistical techniques, and valuation models.
They are not the same thing.
Even a well-designed AVM has limits — its reliability depends heavily on the quantity, quality, recency, and representativeness of its underlying data. If a specialized valuation model struggles when market information is inadequate, there’s even more reason for caution when a general-purpose AI system is asked to value a specific property without being handed sufficient, reliable evidence to work from.
AI can process information. It cannot manufacture reliable market evidence where none exists.
Can AI Replace Property Inspection?
An equally concerning idea is that AI has made physical inspection unnecessary.
There may be legitimate cases for desktop or limited-scope valuation, depending on the applicable standards, the purpose of the assignment, the evidence available, and the agreed scope of work. But that’s a very different claim from saying AI removes the need for inspection altogether.
Consider what an appraiser actually finds on-site that no database captures reliably:
- actual road access and road width
- topography and elevation
- physical condition of improvements
- neighborhood influences
- encroachments, easements, and rights-of-way
- flooding or drainage conditions
- transmission lines and other infrastructure
- actual frontage
- surrounding land uses
- quality of views
- occupancy
- inconsistencies between documents and actual conditions
- other physical characteristics affecting utility and marketability
A database may show a property fronting a road. Inspection may reveal the “access” is a narrow passage shared with several other lots. A map may show a regular, developable parcel. Inspection may reveal severe topographical limitations. Records may describe a residential improvement in good condition. Inspection may reveal serious deterioration.
AI cannot analyze a property characteristic it was never given. Technology doesn’t eliminate the need to actually understand the property being valued.
AI Cannot Cure Inadequate Appraisal Evidence
This may be the single most important idea in the whole discussion: AI cannot cure inadequate appraisal evidence.
Technology can process evidence faster. It can surface relationships within that evidence. It can organize thousands of data points at once. What it cannot do is turn unreliable information into reliable market evidence just by running it through a model.
The difference plays out like this:
- AI + poor data + no verification + no appropriate inspection → a potentially misleading estimate
- AI + reliable data + professional verification → genuinely useful analytical assistance
- AI + reliable data + appropriate inspection + sound methodology + professional judgment → powerful appraisal support
The difference isn’t the sophistication of the AI. It’s the quality of the appraisal process around it.
Can AI Apply the Correct Valuation Method?
Even with sufficient data, another question remains: can AI determine the appropriate method for valuing a given property?
AI can certainly run the calculations. Given verified comparable sales, transaction dates, property characteristics, and defensible adjustments, it can assist with the Sales Comparison Approach — computing unit values, applying adjustments, analyzing ranges, testing alternative assumptions. Given reliable rents, vacancy rates, operating expenses, capitalization rates, and growth assumptions, it can run the Income Approach, including capitalization and discounted cash-flow analysis. Given reliable land values, construction costs, depreciation, and obsolescence data, it can assist with the Cost Approach.
But performing a method correctly is not the same as selecting the right method. That distinction is fundamental.
A Correct Calculation Can Still Produce the Wrong Appraisal
Take a beachfront resort property. An AI system might pull nearby land listings, calculate price per square meter, apply mathematical adjustments, and produce an indicated value — flawlessly.
But what if buyers of comparable resort properties actually base their decisions on income-generating capacity, development potential, tourism demand, or redevelopment opportunity, not raw land comparables? The math can be correct while the underlying methodology is entirely wrong for the asset.
The same trap applies to hotels, industrial properties, special-purpose properties, leasehold interests, partial takings, landlocked parcels, properties burdened by transmission-line easements, environmentally constrained sites, and properties with significant redevelopment potential.
AI can calculate almost anything. The professional question is whether that’s what should have been calculated in the first place. A correct calculation using the wrong valuation method is still a wrong appraisal.
Highest and Best Use Comes Before the Method
There’s a deeper layer still. Before an appraiser even selects a valuation approach, they must determine the property’s highest and best use.
A vacant parcel might physically resemble the residential lots around it. But its zoning, accessibility, location, development trends, physical characteristics, and market demand may point to an entirely different use. Get the highest and best use wrong, and everything downstream can be technically sophisticated and still conceptually wrong.
The proper sequence looks like this:
- Identify the property and property rights
- Define the appraisal assignment and valuation date
- Inspect and investigate as appropriate
- Analyze physical, legal, economic, and market characteristics
- Determine highest and best use
- Select the appropriate valuation approach and method
- Apply relevant, verified market evidence
- Reconcile the value indications
- Form the opinion of value
AI can assist at nearly every one of these stages. But if the whole process starts and ends with a single prompt — “What is the value of this property?” — most of the essential valuation questions never actually get answered.
AI May Spot Similarity — The Appraiser Determines Comparability
Comparable-property analysis is a good illustration of where AI genuinely helps and where it can’t take over.
Given enough reliable data, AI can screen hundreds or thousands of properties and flag those that look statistically similar to the subject. That’s valuable. But statistical similarity is not the same as appraisal comparability.
Two properties can have nearly identical lot areas and sit a short distance apart, yet differ substantially because one has better road access, wider frontage, superior topography, a better view, flood exposure, a transmission-line easement, development restrictions, stronger commercial exposure, or a different highest and best use altogether.
AI may identify statistical similarity; the appraiser determines comparability. Selecting comparables isn’t a search for similar numbers — it requires understanding which characteristics actually drive buyer and seller behavior in that specific market.
Where AI Can Truly Help
None of this means appraisers should reject AI. Quite the opposite — they should learn to use it well.
AI can help organize large datasets, screen potential comparables, analyze market trends, review documents, summarize regulations, flag inconsistencies, run statistical analyses, prepare sensitivity tests, work through income and expense figures, check calculations, and sharpen the clarity and consistency of appraisal reports. It can cut the time spent on repetitive work dramatically.
It’s also an effective quality-control tool — spotting inconsistencies between sections of a report, stress-testing assumptions, checking math, comparing scenarios, and flagging items that need further investigation.
Used well, AI frees the appraiser to spend more time where professional expertise actually matters most: verification, interpretation, highest and best use, methodology, comparability, reconciliation, and judgment. The right relationship looks like this:
AI assists → the appraiser verifies → the appraiser analyzes → the appraiser judges → the appraiser takes responsibility.
The Appraiser Still Signs the Report
This point shouldn’t get lost in the enthusiasm.
If an appraisal contains an unsupported adjustment, an inappropriate comparable, a wrong assumption, the wrong valuation method, or a flawed conclusion, the appraiser cannot fall back on “the AI generated it.” The professional who adopts the analysis and signs the report remains accountable for the opinion of value.
Artificial intelligence does not assume professional accountability. The appraiser does.
The Wrong Question About AI and Appraisal
The debate shouldn’t really be whether AI can produce a property value. Of course it can — a calculator can produce a number, a spreadsheet can produce a number, a regression model can produce a number, an AVM can produce a number, and so can ChatGPT.
The question that actually matters is: can the appraiser demonstrate that the resulting opinion of value is supported by sufficient, relevant, verified market evidence, appropriate methodology, and sound professional judgment?
That’s what separates a numerical estimate from a defensible professional appraisal.
AI Will Change Appraisal — Not Its Foundations
AI is going to reshape this profession. Data gathering will get faster. Market databases will get more sophisticated. Comparable searches will become increasingly automated. Statistical analysis will become accessible to far more practitioners. Report preparation and quality control will get dramatically more efficient. All of that is worth welcoming.
But the fundamentals of valuation don’t move. The appraiser still has to understand the property. Still has to understand the market. The evidence still has to be relevant, and the data still has to be verified. Highest and best use still has to be analyzed. The methodology still has to be appropriate. The conclusion still has to make economic sense. And someone still has to exercise professional judgment and take responsibility for the resulting opinion.
So the better message to appraisers isn’t “use AI to determine property value.” It’s this: use AI to strengthen the appraisal process, but never let it substitute for the evidence, verification, appropriate inspection, methodology, market analysis, and professional judgment an opinion of value actually depends on.
The future of appraisal isn’t AI versus the appraiser. It belongs to the appraiser who knows how to use AI effectively — and just as importantly, knows its limits and when its output shouldn’t be trusted.
Because the challenge in appraisal has never really been producing a number. It’s being able to explain and defend why that number represents value.
Real LexEcon 