I just interviewed the person who helped write the IRS cost seg rulebook. Here is what he said about AI-generated studies.

James C. Peacock spent 38.5 years as a General Engineer at the IRS. He was among the first IRS engineers to examine cost segregation. He contributed to the Cost Segregation Audit Techniques Guide from its inception in 2000 through the February 2025 update. Per James, he trained approximately 200 new-hire IRS engineers on cost segregation and Section 179D before retiring in September 2025.

I sat down with him for 90 minutes and asked him one question that almost every real estate investor is asking right now: what does he actually think about AI-generated cost segregation studies?

His answer was direct. And it was more nuanced than the marketing from either side of this debate would suggest.

Quick Answer: Fully AI-generated cost segregation studies create audit risk because 1245 personal property must satisfy the Whiteco test: movable and in fact moved. AI cannot verify that without inspection support. AI is useful for 1250 completeness checks, such as missing steel, slabs, roofing, HVAC, or electrical systems.

"It Goes Against Every IRS Rule"

I asked James directly about AI-generated cost segregation studies, platforms that classify property without a physical site visit, using satellite imagery, construction databases, and property records.

His response was unambiguous.

"It goes against every IRS rule," he said, "because there's no physical inspection."

That phrase, no physical inspection, is not a stylistic preference. It points to a specific legal standard embedded in 40 years of cost segregation case law. To understand why it matters, you need to understand what the IRS is actually looking for when it evaluates personal property reclassification.

The Whiteco Test: What Most AI Vendors Don't Explain

The IRS uses a framework called the Whiteco test to determine whether an asset qualifies as 1245 personal property, the category that accelerates depreciation from 39 years down to 5, 7, or 15 years.

The test has two parts. The asset has to be movable. And it must have actually been moved.

That second requirement is the problem for AI.

The governing case involves roadside billboards. The court found that the billboards qualified as personal property not because they were theoretically removable but because they were physically dismantled and relocated. Movability in principle is not enough. There has to be documented proof that the asset was, in fact, moved.

James explained how this plays out in practice. Take self-storage facilities. Metal partitions screwed into walls are a classic cost segregation classification. Can they qualify as 1245 personal property? Yes, but only if the owner can prove those partitions were actually removed and relocated, not merely that they could be.

"It has to be movable AND in fact moved," James said.

No AI system can determine that from satellite imagery or permit records. The documentation requirement means someone has to examine the property and the records that show what happened to those assets.

The Kitchen Cabinet Problem

This is not a theoretical edge case. James pointed to the 2012 Tax Court case Amerisouth XXXII Ltd. v. Commissioner as a precedent that still catches investors off guard.

Kitchen cabinets, counters, and sinks are often listed in cost segregation studies as 5-year personal property. The court ruled otherwise. Those items were classified as 1250 property because their function is tied to the building and they are not items that get moved and reinstalled elsewhere.

This matters for AI studies because AI systems trained on general cost segregation databases will often flag kitchen cabinets as 5-year personal property, the classification appears in older studies, in marketing materials, and in practitioner guides. But Tax Court has ruled against that position.

An AI system cannot know whether the kitchen cabinets in your specific property have been moved. A physical inspector can review documentation, interview the owner, and make a supportable determination.

The Amerisouth case is a reminder that common classifications are not always safe classifications. And AI systems learn from what is common.

What the IRS ATG Actually Says About Inspection

The IRS Cost Segregation Audit Techniques Guide, the document James helped write, establishes specific requirements for a quality study. Physical inspection is not optional.

The ATG identifies two inspection approaches as acceptable:

Field examination: Direct on-site inspection by a qualified individual who walks the property, documents component conditions, and verifies what was built, what was modified, and what has been moved.

IRS-acceptable alternatives: Blueprint review combined with photographic documentation and owner interviews, but only when the alternatives can substitute for what a field examination would produce.

What is not listed as acceptable: satellite imagery alone. Tax record review alone. Permit database review alone. Construction cost databases applied without property-specific verification.

AI platforms that classify property using these remote data sources alone are not following ATG requirements, regardless of how their marketing describes the methodology.

James was clear about what happens when an examiner reviews a study without field verification: "The first thing an examiner asks is how was the study prepared. If there is no inspection, that is the first flag."

Why AI Cannot Verify the Support Behind the Report

There is a principle that James returned to throughout our conversation: "It's the support not the report."

A cost segregation report is just a document. What makes it defensible is the underlying support, the photos, the drawings, the field notes, the component-by-component analysis that justifies each classification. Without that support, the report is a number with no foundation.

AI systems generate reports. They cannot generate the support that makes a report defensible.

When an IRS examiner opens a cost segregation audit, they issue Information Document Requests (IDRs), formal requests for the documentation that backs up the claimed classifications. "The least amount of IDRs, the easier the audit goes," James said.

An AI-generated study will produce IDRs about the physical inspection that never happened. The taxpayer cannot produce documents that do not exist. That is when the examination gets difficult.

The One Area Where AI Is Genuinely Useful

James's critique of AI is not a blanket rejection of technology in cost segregation. He drew a specific line.

The problem is AI replacing physical inspection for 1245 personal property classification. The legitimate application is different.

"AI can help ensure completeness of 1250 structural components," he said.

Here is what that means in practice.

When an engineer performs a cost segregation study, they need to account for every component of the building, not just the items that might qualify for personal property treatment but the full structural inventory. Missing a structural component means the reconciliation to total basis will not balance. It also means potential misclassification in the other direction.

AI tools trained on construction documentation, RS Means cost codes, and building specification databases can identify structural components that a human reviewer might overlook. They can flag gaps. They can verify that the component list is complete against what the blueprints and permits describe.

That is a verification and completeness function. The AI is not making the primary classification decision, it is checking whether the human analysis missed anything.

This distinction matters because it describes exactly how AI should be used in cost segregation and why some AI-assisted studies are defensible while fully automated studies are not.

RS Means Codes and Why They Matter

One area where AI has genuine value is in cost allocation. The IRS expects cost segregation providers to assign unit costs to each component using defensible cost data. The standard reference is RS Means, a construction cost database that provides national averages by component type, adjusted for local market conditions.

Applying RS Means codes accurately requires knowing what construction type you are analyzing, what the current regional cost factors are, and how the component in question maps to the available cost data. This is exactly the kind of structured lookup task that AI handles well.

An AI system that automates RS Means application as part of a human-reviewed study is adding efficiency without compromising defensibility. An AI system that uses RS Means lookups as a substitute for site verification is using the tool correctly in the wrong context.

The distinction matters for investors evaluating AI-assisted studies. Ask your provider: what is the AI doing, and what is the licensed engineer doing? If the AI is handling cost allocation based on human-verified component lists, that is defensible. If the AI is making the component classification decisions, that is not.

How Examiners Actually Find AI-Generated Studies

James explained the sequence that leads to cost segregation problems. It is different from what most investors expect.

Cost segregation studies are not flagged by the IRS's DIF (Discriminant Function) scoring system in isolation. The DIF system flags returns for audit based on deviation from expected patterns. A real estate investor with significant depreciation deductions will have a higher DIF score, but that alone rarely triggers audit.

"Cost seg studies don't get discovered until there is already an audit open," James said.

What opens the audit is usually something else: a Schedule C loss, an inconsistency in reported income, a related entity issue, a third-party referral. Once an examiner is on a return, they look at everything. And if they see significant bonus depreciation from a cost segregation study, they will examine that study.

At that point, the examiner will ask for documentation. A fully AI-generated study with no physical inspection will immediately raise questions. The examiner will want to know who performed the inspection, when, and what they found. If those records do not exist, the study's defensibility collapses.

This sequence means investors who assume they are safe because they were not "targeted" for cost segregation are misunderstanding the risk. You do not get audited for cost segregation. You get audited for something else, and then the examiner finds the cost segregation study.

The 1250 Completeness Case: Where AI Belongs

Let me make the positive case concrete.

A commercial property has dozens of structural components that has to be correctly classified as 1250 property, roofing systems, HVAC equipment, plumbing, electrical distribution, fire suppression, elevators, structural steel. These classifications are not contested in the way personal property is. They are 39-year assets. The question is not whether to reclassify them, it is whether the study accounts for all of them and allocates costs correctly.

Missing a building system component creates a reconciliation gap. It also means the total basis calculation is wrong, which undermines the entire study.

AI trained on construction cost databases can verify that a study accounts for all expected building systems given the property type, vintage, and location. It can flag missing components against what the permit records and blueprints describe. It can verify that the allocated costs are consistent with RS Means benchmarks for the construction type.

This is verification work. It does not require the AI to make classification decisions. It requires the AI to check whether the human-prepared analysis is complete and internally consistent.

That is a legitimate and useful application of AI in cost segregation. And it is exactly the opposite of what fully automated AI study platforms are doing.

What to Ask Any Cost Segregation Provider

Based on James's framework, there are three questions that will tell you whether a study is defensible before you file.

Question 1: What was the inspection methodology?

A defensible answer describes either a field examination or a documented alternative (blueprints plus photographs plus owner interviews). An answer that references satellite imagery, public records, or AI-based property analysis alone is a warning sign.

Question 2: Who made the classification decisions?

The ATG requires that studies be prepared by individuals with engineering and tax knowledge. "Our AI system" is not an acceptable answer. You want to know the name of the licensed engineer or qualified professional who reviewed the classifications, how they verified the personal property determinations, and what support documents they generated.

Question 3: How do you handle the Whiteco test for personal property items?

If your provider does not know what the Whiteco test is, stop there. Any engineer who has studied the relevant case law will recognize the term immediately. A blank response tells you everything about the depth of their analysis.

The STR Exception Worth Knowing

One area where the AI debate intersects with a common investor scenario deserves specific attention: short-term rentals.

STR properties are classified as 39-year non-residential real property for depreciation purposes, which changes the analysis for both personal property reclassification and qualified improvement property treatment. The interaction between the 30-day average stay test, the QIP rules, and cost segregation personal property claims creates a specific set of classification issues that many AI systems are not trained to handle correctly.

An AI system that applies residential rental (27.5-year) rules to a property that should be classified as non-residential will generate incorrect classifications at the foundational level. The error compounds from there.

James flagged STR classification as an area where he saw significant inconsistency in studies during his final years at the IRS. Investors who own STRs and are considering AI-generated studies should specifically ask how the provider handles the non-residential classification, and whether a human engineer with STR experience reviewed the work.

For a more detailed breakdown of how STR cost segregation works and where QIP treatment applies, see our companion article: Short-Term Rental Interior Improvements: QIP and Bonus Depreciation.

The Honest Position on AI in Cost Segregation

The marketing from AI cost segregation platforms tends to make two claims: that AI studies are as good as engineering studies, and that physical inspections are unnecessary for most property types.

Both claims conflict with what the IRS actually requires.

The marketing from traditional engineering firms tends to make a different claim: that all AI-assisted cost segregation is dangerous and any study without a mandatory site visit is not defensible.

That claim is too broad.

James's framework is more precise. Physical inspection is required to support personal property classification under the Whiteco test. AI that substitutes for that inspection creates real audit risk. AI that verifies completeness and cost allocation for structural components, as part of a study where a human engineer made the classification decisions, is a legitimate and efficient use of the technology.

The question for investors is not "AI or no AI." The question is what the AI is doing in the study, and whether a licensed engineer with physical inspection experience is making the substantive classification decisions.

If the answer is yes, AI assistance is fine. If the answer is no, you are taking on audit risk that your provider is not acknowledging.

For a free estimate on your property, see FreeCostSeg's calculator. For an assessment of whether your existing study is defensible, use the audit risk guide at FreeCostSeg.

About the Source

James C. Peacock, PE spent 38.5 years as a General Engineer at the IRS and retired in September 2025. He was among the first IRS engineers to examine cost segregation studies and contributed to the IRS Cost Segregation Audit Techniques Guide from its inception through the February 2025 update. Per James, he trained approximately 200 new-hire IRS engineers on cost segregation and Section 179D during his career.

James now consults independently on cost segregation methodology and IRS compliance through JPeacockCSA.com. He can be reached on LinkedIn.

This article is based on a 90-minute interview conducted in June 2026. Direct quotes are reproduced verbatim. All interpretations and editorial framing are Overline's.

Disclaimer: This content is for informational purposes only and does not constitute legal, tax, or financial advice. Consult qualified tax and legal professionals regarding your specific circumstances.

See Your Property's Tax Savings

Drop your address. The AI estimates your depreciation savings in 60 seconds, backed by a certified engineer study.

Overline Property AI● Live
Overline AI

Free cost segregation estimate, engineer-certified studies, and ongoing depreciation tracking.
Backed by $1B+ in supported tax depreciation.