By Julie Bays 

Lawyers have spent the last few years discussing how artificial intelligence can be used in the practice of law. That discussion is important, but it is no longer the only AI issue in the attorney-client relationship. Clients are also using AI. They are using it before they call a lawyer, after they hire a lawyer and sometimes while litigation is pending. They may use AI to understand a demand letter, summarize a contract, draft a statement, evaluate a claim, prepare for a meeting or decide whether they think their lawyer is doing a good job.

This behavior is understandable. Legal problems are stressful. Clients want immediate answers. AI tools are available, easy to use and often presented as conversational helpers. But when clients use public or consumer AI tools for legal problems, they may create risks they do not understand. They may disclose confidential facts. They may input privileged communications into a third-party platform. They may create prompts and outputs that become relevant in discovery. They may rely on inaccurate legal analysis. Or they may create a written record that complicates the lawyer’s ability to advise them later.

For lawyers, the practical question is not whether clients should ever use AI. They will. The practical question is how lawyers should manage client AI use in a way that protects confidentiality, preserves privilege where possible, maintains competent representation and improves the quality of legal services.

A recent federal case, United States v. Heppner, illustrates why this issue needs attention. In that case, a criminal defendant used Claude, Anthropic’s generative AI chatbot, and later claimed that the resulting AI materials were protected by attorney-client privilege and the work-product doctrine. The court rejected those claims. The court reasoned that Claude was not an attorney, that the communications were not confidential in light of the platform’s privacy terms and potential third-party disclosures, and that the materials were not prepared at the direction of counsel or reflective of counsel’s strategy. [i]

The lesson from Heppner should not be overstated. It does not mean every AI-related prompt or output is automatically discoverable, and other courts have reached more protective work-product results in different circumstances. But Heppner does offer an important warning: When clients independently use public AI tools about their legal problems, lawyers may later have to deal with privilege, confidentiality and discovery consequences that could have been reduced with better client instructions at the beginning of the representation.

The better approach is practical and preventive. Lawyers should do three things: Ask clients about AI use, instruct clients on safe boundaries, and preserve AI-related materials when litigation is pending or reasonably anticipated.

Ask About AI Use at Intake

Lawyers routinely ask new clients whether they have signed documents, spoken with opposing parties, received letters, posted on social media or saved relevant text messages. AI use should now be added to that intake conversation.

This does not need to be complicated. Lawyers can simply ask, “Have you used ChatGPT, Claude, Gemini, Copilot, Perplexity or any other AI tool to research, summarize, draft or analyze anything related to this matter?” If the answer is yes, the lawyer should ask what tool was used, what information was entered, whether any documents were uploaded, whether any communications with counsel were copied into the tool and whether the client still has access to the chat history, prompts, outputs or uploaded files.

This question matters for several reasons. First, it helps the lawyer understand what the client may already believe about the legal problem. A client may arrive with confidence based on an AI-generated explanation that is incomplete, jurisdictionally wrong or simply invented. The lawyer needs to know whether the client’s expectations have been shaped by an AI tool before meaningful legal advice can begin.

Second, the question helps identify confidentiality concerns. A client may have pasted a demand letter, settlement communication, contract, financial record, medical record or family-law fact into a public AI tool. The client may not appreciate that, “I was just asking a question” can involve transmitting sensitive information outside the attorney-client relationship. Lawyers cannot address that risk if they do not know it happened.

Third, the question helps identify possible privilege and work-product issues. A client who uploads attorney emails, draft pleadings, strategy notes or summaries of lawyer conversations into an AI tool may create waiver arguments or discovery disputes later. Even when privilege is not waived, the existence of the AI interaction may, itself, require analysis.

For many firms, the simplest fix is to add AI questions to the intake checklist. This is especially important in matters where the client may be tempted to ask an AI tool for immediate legal guidance before or during the representation.

Give Clients Clear Instructions

Asking about prior AI use is only the first step. Lawyers should also give clients clear instructions about future AI use during the representation.

Many clients do not distinguish between a public chatbot, a legal research tool, a private enterprise system, a document automation product and an AI feature embedded in software they already use. To the client, all of these may simply be “AI.” Because the technology is confusing, client instructions should be direct and practical.

A lawyer might tell a client:

Please do not put information about your case into ChatGPT, Claude, Gemini, Copilot, Perplexity or any other AI tool unless you talk with us first. This includes documents, emails, texts, letters, court papers, records or anything we have discussed. Using AI for your case could affect confidentiality, privilege or strategy.

Lawyers should also explain why this matters. Clients may assume that a chatbot conversation is private because it feels private. They may not read terms of service, privacy policies or data-retention settings. They may not know whether the tool stores prompts, uses inputs to improve the model, allows human review, retains uploaded files, shares information with affiliates or responds to legal process.

Ethics guidance for lawyers has repeatedly emphasized that lawyers must understand how AI tools handle information, including data retention, data sharing and the use of prompts or outputs for training. Clients need a simplified version of the same warning.

The instruction should also cover AI meeting tools. More clients are using AI notetakers, transcription tools and meeting summaries. If a client records or transcribes a lawyer-client conversation using the client’s own AI tool, the lawyer may not know where the recording is stored, how long it is retained, whether it is used for training, whether the summary is accurate or whether it could later be obtained in discovery. Recent ethics and risk-management guidance has cautioned lawyers to address these tools directly with clients because they may create confidentiality, privilege, consent, retention and accuracy risks.

Lawyers may need a standard client instruction for the beginning of representation, but the instruction should not be treated as a one-time formality. It should be repeated when the risk becomes more concrete. For example, a lawyer might remind a client not to use AI before the client drafts a statement, responds to a government inquiry, prepares for deposition, reviews discovery, communicates with witnesses or makes social media posts about the dispute.

Preserve AI Materials When Litigation Is Pending or Anticipated

The third practical duty is preservation. Once litigation is pending or reasonably anticipated, lawyers should consider whether AI-related materials may be relevant electronically stored information.

AI materials may include prompts, outputs, uploaded files, chat histories, AI-generated summaries, AI notetaker transcripts, exported conversations, account histories and metadata. Whether those materials are discoverable will depend on relevance, proportionality, privilege, work product, possession, custody, control and the facts of the case. But lawyers should not assume AI materials are irrelevant or immune from discovery simply because they were created in a chatbot.

The discovery issues are already developing. Courts and commentators are beginning to address whether AI prompts and outputs may be protected work product when created by counsel as part of legal strategy. Other cases show that AI conversations can become evidence in a dispute. In Fortis Advisors, LLC v. Krafton, Inc., for example, the Delaware Court of Chancery described a corporate executive’s use of an AI chatbot in connection with a disputed business strategy as part of the factual narrative in a contract dispute.  Fortis is not a privilege decision, but it is a useful warning. The court’s discussion of a business executive’s AI chatbot use shows that AI prompts and outputs may become part of the evidentiary record when they relate to disputed decisions, motives or strategy.

Preservation advice should be tailored to the matter. A client who used AI once to ask a general question before hiring counsel may present a different issue than a client who uploaded key documents, used AI to draft a statement, summarized privileged communications or relied on AI to make business decisions at the center of the dispute. But lawyers should at least ask enough questions to decide whether preservation is necessary.

In litigation matters, lawyers should consider adding AI-related materials to litigation hold notices. A simple preservation instruction might say:

Do not delete any AI chats, prompts, outputs, uploaded documents, AI-generated summaries or AI meeting transcripts related to this matter. This includes materials created in ChatGPT, Claude, Gemini, Copilot, Perplexity, Zoom AI Companion, Teams, Otter, Fireflies or similar tools. Please let us know what tools you used and whether you can access or export the related history.

This instruction serves two purposes. It helps prevent spoliation arguments, and it helps the lawyer assess whether any AI materials should be collected, reviewed, withheld as privileged, logged or produced.

Preservation can be technically challenging. Some AI tools allow users to delete conversations. Some allow chat history to be turned off. Some retain data for a limited time. Some store uploaded files separately from chat text. Some meeting tools generate recordings, transcripts and summaries in different locations. Clients may not know where these materials are stored or whether they still exist. The lawyer may need to work with the client, IT personnel or an e-discovery professional to understand the relevant sources.

Lawyers should also be cautious about advising clients to delete AI materials. A general instruction to stop using AI going forward is very different from an instruction to delete past AI conversations. Once litigation is pending or reasonably anticipated, deletion may create avoidable risk. The better practice is to preserve first, then analyze relevance, privilege and work product.

Make AI Part of the Client Conversation

The rise of client AI use is not just a technology issue. It is a client communication issue. Clients need guidance in plain language. Many clients will not understand the difference between asking a friend, searching Google, using a legal self-help website and uploading documents into a generative AI chatbot. Lawyers are in the best position to explain why those differences matter.

This is also a risk-management issue. A lawyer who never asks about AI use may later learn that the client has uploaded sensitive information, created damaging summaries, relied on incorrect legal conclusions or deleted relevant AI conversations. By contrast, a lawyer who asks early, gives clear instructions and addresses preservation has a much better chance of protecting the client and managing the representation competently.

Clients are going to use AI. Lawyers should assume that some already have. The firms that handle this well will not be the firms that pretend AI use is not happening. They will be the firms that make it part of competent client counseling.

Ms. Bays is the OBA Management Assistance Program director. Need a quick answer to a tech problem or help solving a management dilemma? Contact her at 405-416-7031, 800-522-8060 or julieb@okbar.org. It’s a free member benefit. 

[i] Further reading and analysis see Elizabeth X. Guo, United States v. Heppner, Harv. L. Rev. Blog (Mar. 23, 2026).

Originally published in the Oklahoma Bar Journal — August, 2026 — Vol. 97, No. 6