From Novelty to Evidence: AI Chats Are Now Part of Everyday Life—and Litigation

Artificial intelligence chat tools such as ChatGPT, Claude, Grok, and Gemini (“ChatBots”) are no longer niche productivity aids. They are rapidly becoming part of everyday life. Recent reporting from media and research institutions shows that a majority of Americans now use AI ChatBots for casual conversation, advice-seeking, emotional reflection, and personal problem-solving, not just work tasks.[i] Studies have even observed AI-influenced language patterns creeping into ordinary speech, podcasts, and written communication.[ii] As these tools shift from novelty to routine companions, they are generating large volumes of informal, candid, and sometimes highly revealing communications. This reality raises an increasingly unavoidable litigation question: When, how, and to what extent are AI chat records discoverable?

As AI chat tools become woven into everyday decision-making—where people casually seek guidance, vent about disputes, and even workshop how to respond to conflict—the conversations themselves are increasingly appearing as potential evidence in litigation. In civil litigation, that matters because “informal” AI exchanges can create a timestamped record of what a party believed, when they believed it, what facts they shared, and how they framed their injuries, damages, or intent—often in ways that differ from later pleadings, discovery responses, or deposition testimony. The result is a new and growing discovery frontier: chat logs that look and feel like private brainstorming, but function like any other written communication when relevance, preservation, and production obligations are attached.

The next question is: Can you get such records and how?

No Safe Harbor: Why AI Chats Are Not Privileged Communications

Communications with AI chatbots do not carry attorney-client privilege, psychiatric privilege, or any other type of privilege. OpenAI CEO Sam Altman has publicly acknowledged this gap: “There’s no legal confidentiality for users’ conversations … [and] we haven’t figured that out yet for when you talk to ChatGPT.”[iii] So whether they must be produced in civil litigation usually turns on ordinary discovery standards (likely to lead to the discovery of admissible evidence, proportionality, and other limits) rather than privilege protection.

California privilege law underscores the distinction between the existence of a consultation and the content of communications. For example, Mitchell v. Superior Court explains that merely disclosing that an attorney communication occurred does not itself waive privilege or mask pre-disclosed attorney-client communications. Therefore, discussing with an attorney what was previously discussed with a ChatBot does not cloak pre-attorney ChatBot exchanges. These communications with a ChatBot, if not otherwise objectionable, may be discoverable like any other third-party communication.[iv]

For post-attorney retention, when the client shares information with third-party consultants (such as ChatBots), courts focus on whether third-party involvement was reasonably necessary to accomplish the purpose of legal representation.[v] Absent that tether, AI chat logs are typically evaluated under standard discovery relevance rules rather than the heightened protections that attach to attorney-client communications.

Courts Are Opening the Door: Early Rulings Signal AI Chats Are Fair Game

In a recent 2025 federal court discovery ruling, In re OpenAI, Inc. Copyright Infringement Litigation (S.D.N.Y.), U.S. District Judge Sidney H. Stein compelled OpenAI to produce 20 million de-identified ChatGPT conversation logs to plaintiffs, including the New York Times and other news publishers suing for copyright infringement related to AI training data.[vi] The ruling rejected OpenAI’s objections, finding that users’ privacy interests in these voluntarily submitted chats were adequately safeguarded through de-identification, a protective order, limited sample size (from tens of billions of total logs), and “attorneys’ eyes only” restrictions. Stein distinguished this from cases like S.E.C. v. Rajaratnam (622 F.3d 159, 2d Cir. 2010), noting weaker privacy expectations for data shared with a third-party AI service. He further held that courts need not mandate the least burdensome production method when broader access serves legitimate discovery needs, such as assessing fair use defenses—even for logs not directly reproducing copyrighted material.

The In re OpenAI, Inc. Copyright Infringement Litigation discovery ruling potentially sets a significant precedent in federal courts for treating AI chat records as discoverable electronically stored information, signaling that such logs are not categorically shielded by privacy concerns or overbroad and undue burden objections when relevant to litigation and protected by safeguards. It expands the scope of discoverable AI communications in copyright cases and suggests broader implications: similar records could become accessible in other civil contexts, like personal injury litigation, where chats might bear on issues such as cognitive capacity, damages, or incident circumstances, provided comparable privacy measures are applied.

How to Get the Chats: A Litigator’s Playbook for AI Discovery

Civil litigants have several tools at their disposal to pursue AI chat records, but the most effective ones look familiar: party discovery, testimony, and third-party subpoenas (in narrower circumstances). California discovery reaches any nonprivileged matter relevant to the subject matter (and reasonably calculated to lead to admissible evidence), subject to the court’s power to limit unduly burdensome or intrusive demands.

Start with the Party: Narrow, Focused Requests Beat Broad Demands

Against that backdrop, the best entry point is typically targeted written discovery served on the party—interrogatories and requests for production that first confirm whether the party used an Large Language Model (“LLM”) like ChatGPT/Gemini /Claude/Grok (or similar) to discuss the incident, injuries, damages, liability, or how to respond to discovery, and then narrow production to defined time windows, defined topics, and defined platforms/devices. This matters because “produce your entire chatbot history” will almost always trigger strong privacy, overbreadth, undue burden, and harassment objections—and California’s constitutional privacy framework forces courts to balance the seriousness of the invasion against the requesting party’s need, typically favoring narrow tailoring and protective measures (confidentiality designations, redactions, and in camera review where appropriate).

Locking the Record: Using Testimony to Establish AI Use and Relevance

Depositions and requests for admission (“RFAs”) can be used to lock down the foundational facts, what AI tools were used, when they were used, for what purpose, and whether specific statements or topics were discussed. Establishing these facts not only sharpens relevance and proportionality, but also forecloses attempts to characterize the discovery as a fishing expedition. And where responsive AI records are not produced in written discovery, deposition testimony often supplies the evidentiary predicate needed to seek court authorization for targeted third-party subpoenas or other party-focused relief.

Why Subpoenaing the AI Provider Is Usually a Dead End

Subpoenas to AI providers are a different animal and at least initially a dead end for content. The Stored Communications Act (“SCA”) generally prohibits providers of electronic communication services and remote computing services from voluntarily disclosing the contents of stored communications except under specific statutory exceptions, and courts have applied the SCA to quash civil subpoenas seeking private message content from service providers.[vii] While these providers seem to comply with a court’s order, it is uncertain what content will be provided, and what it will take to have a court order a provider to produce the same.

Practically, that means the cleaner route is usually to obtain the chats from the party (who can export, download, screenshot, or otherwise retrieve what’s in their account), and—if necessary—seek party-focused relief that compels a reasonable search and production (or, in appropriate cases, consent-based access) rather than assuming the provider will produce content in response to third-party process.

Getting Creative: Using Prompts and Platform Tools to Capture Relevant Chats

While this is still a developing industry, the parties will likely need to “get creative” to obtain these records. One avenue may be to request the party use the platform’s search/export tools to locate and produce existing conversations. You can accomplish this with a narrowly tailored prompt instruction that requires the party to (i) run searches within their existing chat history for defined keywords/topics and date ranges, (ii) identify responsive conversation titles/URLs/IDs (if available), and (iii) export/produce those chats in a specified form—while preserving the underlying history to avoid later authenticity fights.

Sample Prompt-Based Discovery Request

A sample request is to ask the responding party to input the below prompt (or whatever prompt you want) into their ChatBots and print out the results:

I need an audit report of my AI chatbot usage related to my lawsuit.

Scope:

• Time range: [INSERT START DATE] to [INSERT END DATE]
• Sources: exported chat logs / pasted chats / conversation history available to you

Search for any of the following concepts (including synonyms):

A) Incident facts: accident, collision, crash, rear-end, impact, liability, fault

B) Injuries/medical: pain, injury, symptoms, diagnosis, treatment, PT, chiropractor, MRI, x-ray, surgery, meds, prognosis

C) Damages/value: settlement, demand, valuation, case value, specials, general damages, wage loss, future care, policy limits

D) Litigation steps: lawsuit, claim, complaint, answer, motion, court, deadlines, service, filing

E) Discovery: interrogatories, requests for production (RFP), requests for admission (RFA), subpoena, objection, privilege, meet and confer

F) Testimony: deposition, testimony, “how should I answer,” “what do I say,” prep questions, credibility

G) Attorney search / counsel: lawyer, attorney, “do I need a lawyer,” “find an attorney,” referrals, contingency fee, retainer, consultation, representation, “should I hire,” communications with counsel

Deliverable:

1) A chronological list of every matching conversation entry with:

• Date/time (or “timestamp unavailable”)
• Platform/tool (ChatGPT / Gemini / Claude / etc., if known)
• The user’s question (quote up to 2 short sentences max)
• The chatbot’s response summarized (2–4 bullets)
• Tag each entry with one or more categories: A/B/C/D/E/F/G
• Flag “Drafting Assistance” if the chatbot drafted or revised text     intended for the lawsuit, discovery, declarations, letters, or deposition prep

2) A short totals section:

• Number of matching conversations
• Count by category (A–G)
• Count marked “Drafting Assistance”

Constraints:

• Do not invent timestamps or content.
• Do not include private unrelated content.
• If you cannot access logs, tell me exactly what you need (e.g., export file or pasted text).

Proceed.

The New ESI Frontier: Precision, Not Panic, Wins AI Discovery Battles

AI chat discovery is moving fast from “novel issue” to routine ESI fight. For litigators, the practical takeaway is simple: assume AI conversations exist, treat them like any other potentially relevant written communication, and pursue them with precision. Broad “turn it all over” demands will (and should) trigger privacy, proportionality, and burden objections—particularly in California—so the winning approach is a tight relevance showing, a defined time window, topic limits keyed to pleaded issues, and reasonable protective measures. At the same time, the absence of privilege and the early federal trend toward production with safeguards should put both sides on notice that these records are not categorically off-limits. The lawyers who adapt now—by asking the right threshold questions, tailoring requests, and building a clean record for enforcement—will be best positioned to capture (or defend against) the next generation of “smoking gun” evidence hiding in plain sight.

Keep Reading

From Classroom to Courtroom: My Personal Journey to Becoming an Attorney Setting Goals & Holding Ourselves Accountable New York State’s AVOID Act Will Drastically Change Third-Party Litigation This Spring From Zero to AI‑Augmented Litigator: My Journey from Avoidance and Fear to Delighted Learner Extraordinary Review: The Life of a Writ in California How I Use the Core Four as the Parent of Teenagers New Laws for The New Year – How To Stay on Santa’s “Good” List in 2026 A Spoonful of Error Makes the Verdict Go Down: The Importance of Contemporaneous Objections An Associate’s Day in the Life

More by this author

Private Thoughts, Public Evidence: AI Chat Conversations and the Next Wave of Discovery The Law of Unintended Consequences: Including Insurance Brokers in Litigation Strategy Communication May Waive the Attorney-Client Privilege 5 New Ways to Evaluate Cases in Light of Skyrocketing Social Inflation To The Victor Goes the Spoils… All of the Spoils Florida Jury Awards $50 Million in Habitability/Mold Case Leveling the Playing Field: Defense Counsel Used “Honest Fouls” in California and Beat a $16 Million Jury Demand Landlords and Tigers and Bears – Oh My! How Landlords are Becoming the Villains of COVID-19 and What They Can Do About It California’s Supreme Court Opens a Path So Plaintiffs Can Recover Greater Nuclear Verdicts™ Against “Deep Pocket” Defendants The California Court’s “Fastpass” to Trial within 120 days – Considerations in Approaching California’s Preference Statute in a Post-Pandemic World

Sources

[i]Horvath, Bruna Over half of American adults have used an AI chatbot, survey finds, NBC News (March 12, 2025)  NBC News https://www.nbcnews.com/tech/tech-news/half-american-adults-used-ai-chatbots-survey-finds-rcna196141?referrer=grok.com (Last visited January 14, 2026);
DeVynck, Gerrit, How people use ChatGPT, according to 47,000 of its conversations, Washington Post (November 12, 2025) https://www.washingtonpost.com/technology/2025/11/12/how-people-use-chatgpt-data/?referrer=grok.com (Last visited January 14, 2026).

[ii] Ramirez, Vanessa, ChatGPT Is Changing the Words We Use in Conversation, Scientific America (July 11, 2025) https://www.scientificamerican.com/article/chatgpt-is-changing-the-words-we-use-in-conversation/?referrer=grok.com (Last visited January 14, 2026).

Chatterji, Aaron, How people Use ChatGPT, Open AI/Duke University/Harvard University (September 15, 2025) https://cdn.openai.com/pdf/a253471f-8260-40c6-a2cc-aa93fe9f142e/economic-research-chatgpt-usage-paper.pdf?referrer=grok.com (Last visited January 14, 2026).

[iii] Perez, Sarah, Sam Altman warns there’s no legal confidentiality when using ChatGPT as a therapist, TechCruch (July 25, 2025) https://techcrunch.com/2025/07/25/sam-altman-warns-theres-no-legal-confidentiality-when-using-chatgpt-as-a-therapist/ (Last visited January 14, 2026).

[iv] Mitchell v. Superior Court (1984) 37 Cal.3d 591.

[v] Behunin v. Superior Court (2017) 9 Cal.App.5th 833; OXY Resources California LLC v. Superior Court (2004) 115 Cal.App.4th 874.

[vi] In re OpenAI, Inc., Copyright Infringement Liti., No. 23-CV-08292, 2025 WL 1652110 (S.D.N.Y. May 30, 2025).

[vii] 18 U.S.C. § 2701.