Commercial real estate transactions have traditionally relied on manual drafting and review of a familiar stack of documents: offering memoranda, rent rolls, operating statements, leases, amendments, estoppel certificates, and subordination and non-disturbance agreements (“SNDA”) to name a few. AI has changed that.
AI is a great tool for commercial real estate practice. More firms are now utilizing AI as drafting assistants. AI can assist in abstracting commercial leases, which includes but is not limited to pulling out key terms of the lease, for example, key dates, rental rates, CAMs, rent escalations, renewal, options, operating expense structures, insurance requirements, assignment provisions, and tenant improvement allocation. These tasks previously required manual work, which was slow, and required several hours being spent by an attorney on a single lease. It is claimed that AI lease abstraction can save up to 50% time.
These are real gains, but it raises the question, what portion of the work that matters in a commercial lease can be performed by AI?
Landlords and tenants are increasingly turning to AI to draft and negotiate leases, attracted by costs savings, far lower than traditional legal fees associated with review and drafting of such documents. However, AI still sometimes misinterprets clauses or overlooks nuanced language, extracts text without its underlying legal implication, or fails to recognize contradictions or unusual provisions.
The short-term savings often hide the longer-term legal costs of fixing those mistakes if the lease is ever tested. These hidden defects in an AI-generated lease would be identified by a qualified attorney.
What AI Should Be Used For
AI is decent at generating standard boilerplate initial drafts for term sheets, summarizing long leases, and running side-by-side comparison of various lease versions. AI can take agreed upon deal points and convert them into a standard term sheet, purchase sale agreements (“PSA”), lease amendments, estoppels, subordination and non-disturbance agreements (“SNDAs”). But such requests typically result in a simple variation of the required document which raises the question of its validity and reliability.
AI performs best when given source material, a defined task, a clear output format, and followed by a qualified professional review.
AI is reliable for repeatable document-heavy work like initial review of leases. The problem starts when users treat AI-generated output as the final draft. AI is limited by its training data. If the training data contains biases, those biases will be replicated in the documents produced by AI, leading to unfair results. That is where a qualified human reviewer makes the difference by identifying AI shortfalls and hallucinations.
[1] See Jonathan West, “AI Lease Abstraction: Faster Lease Review Without Losing Rigor,” (July 8, 2026), https://www.layer3labs.io/guides/ai-lease-abstraction.
[1] See Anubha Bakhsi, “Why Accurate Lease Abstraction Matters in Litigation: The Hidden Risks of AI-Generated Lease Abstracts,” (March 15, 2026), https://www.completelegaloutsourcing.com/lease-abstraction-services-benefits-ai-risks/.
[1] See Tony Alfonso, “Can ChatGPT be Your Next Real Estate Lawyer?,” Daily Business Review (Feb. 15, 2023), https://www.law.com/dailybusinessreview/2023/02/15/can-chatgpt-be-your-next-real-estate-lawyer/.
AI Lease Drafting
AI does not understand a commercial lease. AI generates text by pattern-matching against its training data, meaning, AI generates leases based on its library of leases. Leases generated by AI look professional because the AI model draws on structure, vocabulary, and clause sequencing from the library of examples in its training data. The AI-generated lease covers the expected topics, uses appropriate defined terms, and reads like the work of an experienced legal practitioner.
What the AI-generated lease does not consider is specific deal points. It cannot tell that the landlord’s lender requires an SNDA with particular subordination language, or that a tenant’s business model depends on assignment flexibility, or that the property is in a jurisdiction that restricts landlord self-help remedies. AI simply produces text; it does not tailor the text to what a particular client needs.
These are not examples of AI failure, but characteristics of the technology itself. The errors AI produces are not random but are systematic and predictable and follow directly from the gap between pattern-matching and professional judgment.
The purpose of this article is not to discourage people from utilizing AI, but to encourage using AI for work it does well and supervising where it does not.
Common Pitfalls in AI-Drafted Leases
Below are a few examples our clients have experienced with AI-drafted leases, including leases that contradict themselves, contain hallucinated provisions, or miss important deal points, followed by other risks to consider when using AI to negotiate leases.
Contradictory Provisions
Clients are more frequently utilizing AI to draft their leases and then relying on their professional appearance to utilize them irrespective of whether these leases are legally sound or not. For example, the tenant wants the lease to be assignable, so AI includes an assignment clause that allows the tenant to assign the lease without the landlord’s approval, but later in the lease there is a default clause that states that such assignments would require landlord approval and any failure to obtain such approval will result in a default. Such errors can result in enforceability issues, which can significantly impact the rights of the parties involved.
[1] See Christine M. Walker, “The Impact of Artificial Intelligence on Commercial Real Estate Transactions,” Daily Business Review (Feb. 6, 2025), https://www.law.com/dailybusinessreview/2025/02/10/the-impact-of-artificial-intelligence-on-commercial-real-estate-transactions/.
[1] See ABA Comm. on Ethics & Prof’l Responsibility, Formal Op. 512, at 3–4 (2024).
Hallucinated Cross-References
On the other hand, the landlord may be open to the tenant assigning the lease but wants the right to approve any assignment and set certain conditions. So, AI drafts the following clause: “Notwithstanding anything to the contrary in this Section 11.6, any assignment executed by the Tenant must strictly comply with the fee schedules and background check protocols outlined in Section 49.1 below.” Here, AI has created a phantom cross-reference to Section 49.1 in a 38-section lease. Because the lease has no other sections past Section 38, the referenced Section 49.1 does not exist. This is called a hallucination. “AI hallucinations are instances where an AI system produces outputs that sound plausible but are factually wrong, irrelevant or entirely fabricated.” As a result, the landlord’s conditions for assignment were never included in the lease and the tenant has no way of knowing what it must satisfy for the assignment.
CAM Expenses
AI-drafted leases may also fall short on Common Area Maintenance (“CAM”) expenses, such as missing caps on controllable expenses, unidentified expense categories, or missing capital expenditure exclusions. Without a clear capital expenditure exclusion, a landlord may pass the cost of replacing major building systems to tenants through CAM expenses, leaving tenants with expenses they had not contemplated or budgeted for.
Negotiation Strategy
During lease negotiations, parties may want to modify certain contractual rights based on then current market conditions, relationship dynamics, tenant’s financial strength, landlord’s risk tolerance etc., but AI is not privy to this information, and even if it is, AI will not be able to determine which rights to concede and which to protect, such as whether to grant contraction rights or which rent escalation structure to use. That is professional judgment, not pattern matching.
Confidentiality and Privilege
One of the main items that tenants and landlords overlook is confidentiality and privilege. Unless you have turned off model training or turned on privacy settings, the information you share with public AI tools can place your confidential deal information in the public domain. Uploads to AI can disclose parties’ personal information, financial records, and confidential deal points; therefore, secure handling of this information is important to prevent any unauthorized disclosures.
As for whether communications exchanged with AI are privileged, courts have reached differing results. The Southern District of New York held that a user’s exchanges with a public AI platform were neither privileged nor work product, and that sharing privileged information with the platform waived the privilege. The District of Colorado, addressing only work product, held that AI use does not automatically waive that protection, but it still barred uploading confidential information to AI platforms unless the platform or the company behind the platform is contractually prohibited from using that information for training data or sharing it with third parties. Until the law settles, we recommend keeping confidential information out of public AI tools.
[1] See Chrystal R. China, “What are AI Hallucinations?” (updated July 31, 2026), https://www.ibm.com/think/topics/ai-hallucinations.
[1] See David Saltman, “AI in Commercial Leasing: What It Can and Can’t Replace,” (Nov. 20, 2024), https://leasepilot.co/blog/ai-commercial-leasing-what-it-can-cannot-replace.
Attorney Review of AI-Generated Leases Is a Must
Based on above, AI cannot exercise the professional judgment that determines whether a commercial lease actually serves your interests. That judgment involves understanding your specific deal, your specific property, the governing jurisdiction, and your business objectives. It involves recognizing when a provision is ambiguous and evaluating whether the ambiguity helps or hurts your position. It involves stress-testing the document under adversarial conditions, asking not just what the lease says, but what happens when the other side tries to use it against you. That is what a commercial real estate attorney does, and it is something that AI cannot replace.
We cannot deny AI’s place in commercial real estate; however, what makes it effective is having an attorney review the AI output and use their legal judgment. AI should handle the first pass, extracting terms, comparing versions, flagging non-standard language. The attorney should then take it from there, verifying accuracy, correcting errors, negotiating the provisions that matter, and ensuring the document holds up when it is tested. The attorney can always utilize AI to change certain provisions or draft new provisions based on their experience and knowledge. This process captures the cost savings AI offers while ensuring you are not exposed to the risks it creates.
A commercial lease governs a multi-year financial commitment, and the internal consistency and integrity of the lease document determines your exposure when something goes wrong. AI-generated documents are speedy and low cost when compared to attorney-prepared documents, but speed does not translate into accuracy, and low cost may expose you to a defective lease that can cost you over the life of the deal. If you enter into a commercial lease, whether as a landlord, tenant, investor, developer, or bank, do not blindly rely upon AI-generated documents.
Hallucinations, incorrect cross-referencing, contradictory provisions, missing terminology, and possible disclosure of confidential deal points are all risks an attorney can address. Have these documents reviewed by a commercial real estate attorney.
[1] See Walker, supra note 4.
[1] See United States v. Heppner, 820 F. Supp. 3d 292, 296–97 & n.3 (S.D.N.Y. 2026).
[1] See Morgan v. V2X, Inc., No. 25-cv-01991-SKC-MDB, 2026 U.S. Dist. LEXIS 67939 (D. Colo. Mar. 30, 2026).
