A general-purpose AI chatbot should not be the authority behind a completed cost segregation report. It can produce clear prose, tables, citations, and calculations within minutes. None of that proves that the property facts, costs, classifications, or schedules are correct.
A supportable study begins with the actual property. It uses documented basis, property-specific facts, a stated cost method, supported classifications, and calculations that another qualified reader can follow.
Judge the evidence behind the report, not how polished the pages look. Start with the fictional failure example, then use the report-quality checklist to examine a provider's work.
Where limited AI assistance can help
AI can help with work around a study. An owner might use it to draft a document request, organize questions for a CPA, or summarize a source that the owner then checks. A provider might use software to make an intake easier to complete.
Those tasks do not make the model the report authority. A general chatbot should not establish depreciable basis, choose a final land allocation, estimate components from missing facts, classify property for tax purposes, or complete the depreciation schedules. Those decisions depend on records, physical details, tax authority, and a method that remains available for review.
The useful question is not whether a service uses the label "AI-powered." Ask what the model does, which facts it receives, how its output is checked, and whether the final report stands on evidence outside the generated text.
Why fluent output is not evidence
Language models generate likely responses based on their training and the information available during a request. That information may include retrieved web pages, uploaded documents, and instructions from the user. More information can help, but it does not guarantee that an answer fits a particular property. OpenAI explains why hallucinations remain a problem even as models become more capable.
A peer-reviewed Nature study on semantic entropy examines one way to detect plausible falsehoods known as confabulations. The researchers did not claim that the method guarantees factual output or catches every kind of mistake.
In cost segregation, a model can quote a real tax source and still apply it to the wrong asset. It can also perform correct arithmetic with an unsupported starting number. The citation and calculation may be real while the conclusion remains wrong for the property.
A fictional chatbot failure
Suppose a rental owner gives a chatbot the purchase price, address, square footage, and several exterior photographs. The owner asks it to produce a complete cost segregation report.
The model assumes that land represents 20 percent of the purchase price. It applies a generic percentage to shorter-life assets, estimates component costs from broad construction averages, and produces depreciation schedules with IRS citations.
A disclosed temporary land assumption can be useful in a preliminary estimate when the owner does not yet know the land value. It should not become the final report's land allocation without support from the owner's records or another appropriate source.
What looks complete and what is still missing
A land percentage needs a documented source. A plausible percentage does not replace a filed schedule, assessor ratio, appraisal, or other support for the final allocation.
A starting basis needs complete records. Check whether acquisition costs, later improvements, prior depreciation, and other relevant facts were accounted for.
Component quantities and costs need property evidence. Exterior photos do not establish concealed systems or interior finishes. Generic percentages do not explain quantities or the cost method.
IRS citations need an explanation of how they apply. A real citation does not establish the right classification for a particular component and its use.
Completed schedules need reconciliation and responsibility. Allocated costs should tie to the property's depreciable basis and renovation records, with an identified preparer responsible for the work.
The problem is the missing path from the owner's records to the final schedules. Correct-looking arithmetic cannot repair unsupported inputs.
What the finished study should show
The IRS Cost Segregation Audit Technique Guide, Publication 5653 describes principal elements of a quality study and report. Its guidance discusses preparer expertise, methodology, supporting documents, legal analysis, unit costs, engineering take-offs, cost reconciliation, assumptions, certification, and exhibits.
A property owner or CPA should be able to review the following parts of the work.
Property and scope. The property, relevant dates, building use, project scope, and records received.
Depreciable basis. The amount analyzed, the treatment of land and later improvements, and any connection to an existing depreciation schedule.
Cost method and quantities. The records, estimates, measurements, photographs, interviews, or stated assumptions behind component costs.
Tax classification. The authority and property facts supporting each material classification.
Cost reconciliation. How allocated component costs tie back to the total depreciable cost being studied.
Assumptions and limits. Missing information, estimates, unresolved questions, and limits on the work.
Preparer and deliverables. Who prepared and reviewed the study, along with the component detail, schedules, and supporting exhibits.
Not every report presents these items in the same format. The work should still be traceable. A saved prompt shows what someone asked a model, but it does not establish that the inputs were complete or that the answer was correct.
Owners who need help with the starting number can read how to determine depreciable basis. Owners reporting later work can use the renovation documentation guide to organize dates, costs, invoices, and category details.
How Basis draws the line
Basis does not use AI to calculate or complete the final cost segregation report. The property facts, cost calculations, tax classifications, and completed study must stand on documented information and professional review.
The report does not decide how a taxpayer files a return. The owner and tax professional must evaluate how the depreciation schedules fit the return, elections, timing, and broader tax situation.
You can review a fictional Basis sample report before buying. Look at how it explains the property, classifications, and report sections. Proprietary calculation details are withheld in the public sample.
When comparing services, use the residential provider comparison to check differences in intake, review, support, and property fit. The web-based tool guide shows what to look for in the online process as well as the delivered report.
Decide whether to investigate further
If you are still deciding whether a study makes financial sense, start with the cost segregation eligibility guide. You can also run a free Basis property estimate before ordering a report. A preliminary estimate can help screen the opportunity, but it does not replace the records and analysis behind a completed study.
© 2026 Basis Cost Seg. For informational purposes only and not tax or legal advice. Depreciation treatment depends on the property and the taxpayer. Consult a qualified tax professional regarding your situation.
Further reading
- IRS Cost Segregation Audit Technique Guide, Publication 5653 IRS examination guidance describing principal elements of quality cost segregation studies and reports.
- Why language models hallucinate OpenAI research explaining why capable models can still produce confident errors.
- Detecting hallucinations in large language models using semantic entropy Peer-reviewed research examining one method for detecting confabulations and its limits.
Stock photo source: Pexels