Could Your Project-Level AI Prompt Become the Other Side’s Smoking Gun?

A dispute is brewing on the project. Wanting a quick assessment, the project manager opens ChatGPT, Claude, or another generative artificial intelligence (“AI”) platform. The project manager types the dispute background into the prompt, including concerns, explanations, and a frank discussion of what may have gone wrong on the contractor-side. The AI response identifies potential weaknesses in the contractor’s position, including possible failures to meet contractual obligations.  Unhappy with the answer, the project manager starts a new chat, revises the prompt, and tries again. If the dispute escalates to arbitration or litigation, can the owner force the contractor to provide these prompts and the resulting responses?

Frustratingly, the answer is: it depends. Courts have reached different conclusions. For now, the safest course is to involve your counsel before using AI to evaluate a demand letter, default notice, or threatened lawsuit, or before using AI to draft dispute-related emails, claim letters, or other communications that contain project-specific facts. Otherwise, what feels like a private conversation may later become key evidence for the opposing party.

Even though that concern is only one of the risks associated with contractors using generative AI on a project, this article does not attempt to address every risk associated with its use. Instead, it is limited to the dispute-related risks that arise when project or commercial managers use generative AI for quick feedback or gut checks on real-time project issues. 

AI Conversations May Be Discoverable

Generative AI has quickly become a common business tool in recent years. One leading AI platform is now being used by more than 800 million people weekly.[1] Construction professionals use AI to draft emails, change-order requests, RFI responses, notices, claim letters, and countless other project documents. In doing so, users often share project facts and strategic thinking with AI platforms. Because users interact with a chatbot rather than another person, the conversation may feel confidential. Users may also assume that a closed or abandoned chat simply disappears. However, those assumptions may not be correct.

While direct communications with your attorney are generally privileged (meaning you do not have to provide them to an opposing party), using AI does not create a new privilege or automatically shield a conversation from discovery in a later dispute. AI prompts and outputs may become discoverable in litigation just like routine emails, text messages, and other electronic records. When employees use AI as part of their day-to-day project responsibilities, those communications will generally be subject to disclosure.

However, a more complicated question arises when AI is used to analyze or discuss issues related to anticipated or ongoing litigation.

Potential Protections for AI Communications

When an opposing party seeks AI conversations discussing an anticipated or ongoing dispute, parties typically look to two standard litigation protections: the attorney-client privilege and the work-product doctrine. The attorney-client privilege protects confidential communications between attorneys and clients made to obtain or provide legal advice. Importantly, that protection does not apply simply because a user asks an AI platform for legal advice.[2] Further, the contractor may expose its lawyer’s confidential legal advice by entering attorney-client communications into an AI platform. Construction professionals therefore should not assume an AI conversation is protected merely because it involves legal issues, nor should they upload attorney communications or legal advice without first consulting counsel.

Contractors often prepare materials to assist counsel when disputes arise. The work-product doctrine generally protects certain materials prepared in anticipation of litigation. In a construction dispute, those materials might include a project chronology prepared for an attorney, a summary of project records, or an assessment of alleged defects. Today, many of those materials may be drafted with the help of AI. To provide useful responses, users often feed the platform detailed facts about the dispute, including their view of the strengths and weaknesses of a claim.

As a result, the prompt itself may be more revealing than the output. It may contain internal concerns, unfavorable facts, or alternative explanations that never appear in the final email, summary, or project-related documents. If disclosed, those prompts and responses could be very damaging for a case.

Conflicting Decisions on AI Work Product

Courts have not yet agreed on whether the work-product doctrine protects a contractor’s AI prompts and outputs in litigation. A Texas court recently refused to require a company to disclose a company representative’s ChatGPT conversations about case-related documents he uploaded to the platform.[3] But the company still had to disclose every document he uploaded to ChatGPT, including confidential materials.[4]

Conversely, two courts required parties to disclose dispute-related AI conversations created on the user’s own initiative rather than at counsel’s direction.[5] In one case, a criminal defendant entered information he had learned from his lawyers into Claude to generate reports discussing legal arguments and defense strategy.[6] He later shared those reports with his lawyers.[7] Still, the government was able to obtain 31 documents containing his conversations with the AI platform, including information that revealed aspects of his legal strategy.[8]

These decisions suggest that a represented party’s independent use of AI to investigate a case or develop strategy may present different work-product concerns than AI use directed or supervised by counsel. But in any event, there is a not a clear, universal answer to the question.

Best Practices for Dispute-Related AI Use

Until the uncertainty is resolved, construction professionals should involve counsel before using AI to evaluate a demand letter, default notice, claim or threatened lawsuit. The same holds true for using AI to draft dispute-related emails, claim correspondence, or other communications that require project-specific facts. Counsel’s involvement cannot guarantee protection, but it may strengthen the argument that the work was performed in anticipation of litigation rather than as part of an employee’s ordinary business activities.

Construction professionals should also not assume that their AI conversations will remain private. For now, every prompt should be treated as a record that the opposing party may someday read. Users should proceed with the assumption that any unfavorable facts, concerns, questions, or drafts entered into an AI platform could someday appear in court or arbitration.

Finally, construction professionals should not upload counsel’s legal advice, protected discovery materials, or confidential project information to an AI platform without prior approval and direction by counsel. They should use only company-approved and counsel-approved AI tools that include appropriate security, confidentiality, and data-protection safeguards.

In sum, AI can help construction professionals evaluate project issues, organize information, and work more efficiently. But it can also create records that expose weaknesses in a party’s position. Because courts disagree over whether those records receive legal protection, construction professionals should take a cautious approach when using AI with dispute-related materials. Without the proper safeguards, a seemingly private AI conversation may become the opposing party’s smoking gun.

“The Construction Industry Team at Jones Walker LLP is one of the most highly regarded and award-winning construction law practices in the nation. Our experienced construction attorneys understand the complex dynamics between — and the unique priorities of — project participants and can craft effective solutions that minimize disputes, manage risks, and help keep projects moving from conception to completion.” 

The views expressed in this article are not necessarily those of ConsensusDocs. Readers should not take or refrain from taking any action based on any information without first seeking legal advice. 

[1] See Bond, Trends – Artificial Intelligence (2025) https://www.bondcap.com/report/pdf/Trends_Artificial_Intelligence.pdf at 59 (accessed October 14, 2026).

[2] Whisman v. Rajabi, No. 6:25-CV-06063-SOH-MEF, 2026 U.S. Dist. LEXIS 217896, at *2-3 (W.D. Ark. July 15, 2026).

[3] Tate Grp. Auto., LLC v. Legacy Auto. Cap., LLC, No. 25-BC11B-0020, 2026 TXBC LEXIS 43 (Tex. Bus. Ct. June 3, 2026).

[4] Id.

[5] U.S. v. Heppner, 820 F. Supp. 3d 292 (S.D.N.Y. 2026); Shealy v. Seaside Invs., LLC, No. 2684CV00799-BLS2, 2026 Mass. Super. LEXIS 135 (Mass. Super. Ct. June 16, 2026).

[6] Heppner, 820 F. Supp. 3d at 295.

[7] Id.

[8] Id. at 299.