Legal Meeting Documentation: What Lawyers Need to Record
Client meetings, depositions, and negotiations require accurate records. Learn what bar associations require, where attorney-client privilege applies, and how AI transcription fits into your documentation workflow.
A lawyer sits down with a client to discuss a potential lawsuit. Twenty minutes into the conversation, a detail surfaces—a conversation that client had with a third party, the tone of a negotiation, a timeline that might matter. The lawyer's instinct is to capture it precisely.
But the moment you record that conversation, you've crossed from memory into documentation. And documentation has legal consequences. What you capture, how you store it, who can see it, whether it's protected by privilege—these questions matter more for a lawyer than for almost any other profession.
Most attorneys rely on handwritten notes. Some use digital note-taking apps. An increasing number are experimenting with AI transcription tools—Whisper, Otter, Fireflies—to convert the meeting into text. Each approach trades accuracy for privacy, completeness for privilege risk.
This guide covers what bar associations actually require from legal meeting documentation, where attorney-client privilege protection begins and where it can end, the genuine limitations of AI transcription for legal work, and how to think about tool choice when your documentation can be subpoenaed.
Automate your meeting notes. MinuteKeep records your meeting and uses AI to transcribe, summarize, and extract action items. 9 languages, no subscription, 30 min free.
What Bar Associations Require from Legal Documentation
Unlike corporate boards, which operate under specific statutory and common law rules about record-keeping, lawyers operate under professional responsibility rules enforced by state bar associations. These rules are not prescriptive about how to document a meeting. They are strict about the effect of what you document.
ABA Model Rule 1.4 requires that you keep clients reasonably informed about the status of their matter. This does not require a written record of every conversation, but it does require you to know what you told them and when.
ABA Model Rule 1.6 prohibits disclosure of information relating to the representation without informed consent. This is the privilege rule, and it is the centerpiece of legal documentation practice.
ABA Model Rule 1.1, amended in 2012 and adopted by 42 U.S. states as of 2026, requires competence in the technology you use. This includes understanding the security and confidentiality implications of any tool you use to record, transcribe, or store client communications. Ignorance of how a tool handles privileged data is itself a breach of this duty.
ABA Model Rule 1.15 requires safekeeping of client property and records. If a client sends you documents during a meeting, or if you create notes from that meeting, those materials must be maintained securely and made available to the client on request.
ABA Opinion 95-371 (Confidentiality of Information About a Former Client) established that a lawyer's obligation to maintain client confidentiality survives the end of the representation.
New York City Bar Association Formal Opinion 2025-6, issued in January 2026, provides specific guidance on AI transcription tools used by attorneys. The key finding: attorneys may use AI to transcribe client communications, but only if:
- The attorney has disclosed the use of AI transcription to the client and obtained informed consent
- The attorney has conducted reasonable due diligence on the vendor's security and data handling practices
- The attorney understands whether recordings or transcripts are retained by the service, used for model training, or shared with third parties
- The attorney has a plan for securely storing or deleting the transcripts according to the client's engagement terms
This is not blanket permission to use Otter or Fireflies on a client call. It is permission if you've done the groundwork.
Where Attorney-Client Privilege Begins and Where It Ends
Attorney-client privilege is the shield that protects communications between a lawyer and client from being used as evidence against the client. It is foundational to the legal system—without it, clients cannot speak candidly with their lawyers, which undermines both the client's interests and the administration of justice.
But privilege is fragile. It can be waived unintentionally, especially through poor documentation practices.
What Communications Are Protected
For a communication to be protected by attorney-client privilege:
A lawyer must be involved. The communication must be to, from, or involving a licensed attorney acting in their professional capacity. A business advisor, consultant, or accountant is not an attorney, even if they are giving you advice.
It must be confidential. The client and attorney must intend the communication to be confidential. If you discuss legal strategy in front of your business partner, an employee, or a third party who is not present for a legitimate reason (e.g., an interpreter, a co-counsel), the privilege may be waived.
It must be for the purpose of seeking or receiving legal advice. Communications about the lawyer's fees, administrative matters, or social conversation are not privileged. Communications about how to commit fraud, evade taxes, or engage in illegal conduct are not privileged (and may create affirmative legal obligations to report).
The communication must remain confidential. Once you disclose privileged information to someone outside the relationship—or allow it to be disclosed—the privilege may be waived. This can happen unintentionally through a poorly protected email, a cloud file shared with the wrong permissions, or a transcript uploaded to a third-party service.
Where Privilege Ends: Three Scenarios
Scenario 1: The Third-Party Recording Bot
You are on a Zoom call with your client discussing settlement strategy. Unbeknownst to you, a colleague has Fireflies connected to their calendar, and the bot joins the call automatically. The bot records the audio and sends it to Fireflies' servers for transcription.
Question: Is privilege waived?
Under the reasoning in United States v. Richey (N.D. Ill. 2013) and similar cases, allowing a third-party bot into a privileged conversation—even without your knowledge—demonstrates a lack of intent to maintain confidentiality. Opposing counsel could argue that by permitting a bot to listen and record, you have implicitly disclosed the communication to a third party. Courts have found this to be a waiver of privilege.
This is why the New York City Bar Association specifically warned attorneys to ensure they have full control over who (or what) participates in a privileged conversation. If you are using a tool that joins calls as a bot, you cannot guarantee privilege protection.
Scenario 2: Using Cloud Transcription Without Data Handling Clarity
You record a client call on your iPhone, upload the recording to Otter.ai or a similar service for transcription, and leave the transcript in your Otter account. You assume Otter deletes the recording after transcription.
In reality, Otter retains transcripts and audio in its systems, uses that data to improve its models, and stores recordings across multiple data centers. Your client's confidential information is now part of Otter's training dataset.
Question: Is privilege waived?
This is unsettled law, but the emerging consensus is that if you send privileged communications to a third-party service without understanding or controlling what happens to that data, you have likely waived privilege. The reasoning: privilege requires reasonable care to maintain confidentiality. Uploading to a service without knowing whether your data will be retained, used for training, or shared constitutes lack of reasonable care.
Moreover, if a dispute arises and opposing counsel argues that by using Otter's service you've disclosed the transcript to Otter's staff and training systems, a court could exclude it from privilege protection.
Scenario 3: Storing Notes Without Encryption or Access Control
You record a client meeting, transcribe it yourself (or use a local tool), and store the transcript in an unencrypted file on your cloud account with default permissions. A colleague accidentally gains access. Or your cloud account is breached.
Question: Is privilege waived?
Breach of confidentiality through inadequate safeguards can constitute waiver. Courts have found that failure to use reasonable security measures (encryption, access control, password protection) is evidence that you did not maintain confidentiality, and therefore did not intend to maintain privilege.
The Test: Reasonable Care Under the Circumstances
Across all three scenarios, courts apply a "reasonable care" standard. What security and confidentiality measures would a competent attorney take given the sensitivity of the communication and the technology available?
For client meetings discussing strategy, settlement negotiations, or sensitive personal or business information, reasonable care now includes:
- Understanding exactly what a vendor does with your data (retention, training use, third-party access)
- Disclosing to the client that you are recording and how the recording will be stored
- Using tools that process transcripts locally or with explicit data deletion policies
- Encrypting any stored transcripts or recordings
- Limiting access to stored materials to people with a legitimate need to know
- Having a documented deletion schedule (e.g., "client meeting recordings deleted after 90 days unless litigation hold applies")
AI Transcription: What It Gets Right and Where It Fails
AI transcription tools—particularly Whisper (OpenAI's speech-to-text model) and newer legal-specific tools—have reached levels of accuracy that are genuinely useful for legal documentation. Modern systems can achieve 95-97% accuracy for clear English speech, with specialized legal terminology training pushing accuracy even higher.
Where AI Transcription Helps Lawyers
Accuracy in capturing exact language. When a client says "I told him the contract was unenforceable," you want to capture that exact phrasing, not a paraphrase. A human note-taker, even experienced, introduces interpretation. A transcript is verbatim.
Completeness for later review. A full transcript lets you revisit a conversation weeks later with confidence. Did the client actually say "the deadline is March 15" or "around mid-March"? The transcript tells you.
Speaker identification. If you are meeting with multiple parties—co-counsel, a client, third parties—a transcript that identifies who said what is far more useful than handwritten notes.
Defensibility in litigation. If you are later called to testify about what was said in a meeting, a contemporaneous transcript is more defensible than your memory or handwritten notes, even if the transcript is an AI output.
Where AI Transcription Breaks Down
Legal terminology and jargon. General-purpose transcription models (including Whisper) struggle with Latin phrases, technical legal terms, foreign language segments, and acronyms. "Force majeure" becomes "for sure measure." "Habeas corpus" becomes "have us corpus." These errors are not catastrophic if you are reviewing and editing the transcript, but they require human correction.
Multiple speakers and overlapping dialogue. In a deposition or negotiation where multiple parties are present and voices overlap, even good transcription systems make errors. One speaker's words are attributed to another. Interruptions are lost. A court-reporter-generated deposition transcript is still more reliable than AI transcription in this context.
Nuance, tone, and context. A transcript of words cannot capture hesitation, sarcasm, emphasis, or implicit agreement. "We might be able to do that" and "We could do that if we really had to" convey different meanings to a human ear but might appear identical in a transcript.
Privilege and confidentiality of the transcript itself. Here is the critical issue: once you create a transcript—whether AI-generated or not—you have created a new document. That document is subject to discovery in litigation. If opposing counsel can obtain the transcript, they can read every word the client said, every question you asked, and potentially infer your legal strategy.
An attorney note-taker can choose what to write down. They can omit tangents, false starts, and privileged strategy discussions. An AI transcript captures everything. This creates a new risk: a complete transcript, if discovered, reveals more than you would have voluntarily disclosed through notes.
Best Practices for AI Transcription in Legal Work
Disclose to the client. Before recording, tell the client: "I'd like to record this meeting and have the recording transcribed for accuracy. The transcript will be stored securely and treated as attorney work product. Do you consent?"
Use a tool that processes locally or with explicit data control. MinuteKeep, for example, records on your iPhone and processes via Supabase without storing the audio file. Whisper.cpp can run locally. Google's speech-to-text and Azure Speech Services allow private deployments. Avoid tools that route audio to public cloud services without understanding their data retention policies.
Edit the transcript before using it. Review the AI output carefully. Correct errors, verify speaker attribution, and remove any portions that are clearly off-topic or privilege-sensitive. The edited transcript is your record.
Treat the transcript as work product. Under Fed. R. Civ. P. 26(b)(3) and similar state rules, attorney work product—documents created by an attorney in anticipation of litigation—is protected from discovery. A transcript annotated with your notes and edits is more defensible as work product than a raw AI output.
Delete or destroy transcripts according to your retention policy. If a client meeting is not litigation-related, you likely don't need to keep the transcript indefinitely. Establish a retention schedule (e.g., "Transcripts of client meetings are deleted 90 days after the engagement ends unless litigation hold applies") and document that you follow it.
Understand the ethics rules in your jurisdiction. New York has issued guidance. California, Illinois, and other states are likely to follow. Check your state bar association's recent opinions on AI use in client communications.
The Privacy Question: Which Tools Are Safe for Privileged Conversations?
Not all AI transcription tools treat your data the same way.
Bot-based tools (Fireflies, Otter.ai, Read AI, Sembly): These join your meeting as a visible participant. The audio is sent to cloud servers for processing. Retention policies vary, but many tools retain transcripts and audio for their own infrastructure, analytics, and model training. For attorney-client conversations, these tools present unacceptable privilege and confidentiality risks.
Cloud API services with explicit data handling (Google Cloud Speech-to-Text, Azure Speech Services, AWS Transcribe): These can be configured to not retain audio or transcripts for training purposes. You maintain explicit control over where data is stored and when it is deleted. If you use these services, you can verify through their terms of service that audio is not used to train their models.
Local processing (Whisper.cpp, on-device transcription, privacy-first apps): Tools that transcribe on your own device, without sending audio to external servers, eliminate the third-party data handling question entirely. MinuteKeep, for example, records on your iPhone and processes transcription via Supabase API without storing your audio files. This local-first approach is the most defensible for privilege protection.
Legal-specific tools: A growing number of vendors are building tools specifically for legal professionals, with features like speaker identification for depositions, redaction capabilities for privilege-sensitive content, and compliance with HIPAA, GDPR, and legal data handling standards. These tools are newer and more expensive, but they are designed for the exact use case.
CTA: Document Client Meetings with Confidence
If you are documenting client meetings—whether strategy sessions, client consultations, depositions, or negotiations—you need a tool that:
- Captures accurate, searchable transcripts
- Handles legal terminology correctly
- Processes without routing audio through third-party servers
- Stores transcripts securely under your control
- Lets you edit transcripts before filing or sharing
MinuteKeep on iOS records client meetings on your device and generates accurate transcripts without any bot joining your call or audio being sent to external servers for storage. You get a complete, searchable record of the conversation. You edit it before using it as documentation. You control who sees it and when it's deleted.
No subscription required. 30 minutes free on install. Pay-per-use from $0.99.
Download MinuteKeep from the App Store
Frequently Asked Questions
Can I record a client call without telling the client?
It depends on your state's recording laws. In one-party consent states (36 U.S. states), you can record a conversation if you are a party to it, without notifying the other party. In all-party consent states (California, Illinois, Florida, Maryland, Montana, Nevada, and 7 others), you must obtain consent from all parties before recording. For video conferences with out-of-state participants, apply the most restrictive rule: if any participant is in an all-party consent state, you need consent from everyone. Beyond legality, bar associations increasingly recommend disclosure to clients as a matter of ethics and professional courtesy. Transparency builds trust.
If I record a client call and don't use the recording, can it still be discovered?
If a recording exists and litigation arises, yes. During discovery, opposing counsel can request all audio files, notes, and work product related to the matter. Failure to produce them can result in sanctions. The safest approach is to delete recordings after transcription (if the transcript is sufficient) or to clearly label them as attorney work product and maintain a secure location. Never assume a recording "won't be needed" and leave it lying around unsecured.
What if I record a client call but the client later claims privilege was waived because I used Otter?
If you used a tool that sends audio to a third-party server without explicit data control, a court could find that you failed to maintain the confidentiality required for privilege to attach. The remedy would be that the opposing party can access the transcript. To avoid this, use tools where you understand and control what happens to your data, disclose your transcription method to the client upfront, and get documented consent.
Can I use the same AI transcription tool for both attorney-client calls and business meetings?
You should not. Use separate tools with different security profiles: a local or private-cloud tool for privileged attorney-client conversations; a standard tool (Otter, Fireflies, or your meeting platform's built-in transcription) for routine business meetings. This separation makes it clear that you are treating privileged communications differently, which strengthens the confidentiality argument if your documentation is later challenged.
What's the difference between a transcript and a summary for legal documentation?
A transcript is a word-for-word record of what was said. A summary is a condensed account of the key points. For legal work, transcripts are more defensible because they are verbatim and less subject to accusations of editorial bias. Summaries are useful for quick reference, but they should not be your sole record of a client meeting. If you use AI summarization, also keep the full transcript (or a detailed transcript summary) as your official documentation.
If I delete a transcript after the meeting, does that hurt me if litigation arises later?
If you have a documented retention policy (e.g., "Transcripts of non-litigation client meetings are deleted 90 days after engagement end"), and you follow that policy consistently, deletion is routine and defensible. If you selectively delete transcripts of certain meetings but not others, that selectivity can be used against you in discovery (as evidence of consciousness of guilt). Establish a clear retention schedule and apply it uniformly.
Key Takeaways
- Bar associations require that attorneys understand the security and confidentiality implications of any technology used to record or store client communications. Ignorance is not a defense.
- Attorney-client privilege requires confidentiality. Privilege can be waived if you use a tool that sends audio to third-party servers without explicit data handling agreements, if you fail to disclose recording to the client, or if you store transcripts without adequate security.
- AI transcription can achieve 95-97% accuracy for clear speech, but it struggles with legal jargon, multiple speakers, and nuanced communications. Always review and edit transcripts before using them as documentation.
- Bot-based transcription tools (Fireflies, Otter, Read AI) join your call as a third party and route audio to cloud servers. For attorney-client conversations, these tools present unacceptable privilege and confidentiality risks.
- Use local-processing or private-cloud tools for privileged conversations. Ensure you understand the vendor's data retention and use policies. Disclose to clients that you are recording and how the recording will be handled.
- Treat transcripts as attorney work product. Edit them, secure them with encryption and access control, and delete them according to a documented retention policy.
- Recording laws vary by state. In all-party consent states, you must get consent from all participants before recording. Even in one-party consent states, transparency is a matter of professional ethics.