AI Transcription for Lawyers — Drafts, Discovery Review, and Internal Workflows
By VexaScribe Editorial · Published June 23, 2026
TL;DR. AI transcription has a real role in legal practice, but not as a replacement for certified court reporters. Where AI fits: deposition prep, discovery review at scale, client intake interviews, internal team calls, CLE recordings, research interviews, and multilingual witness statement drafts. Where AI does NOT fit: certified court transcripts, sworn depositions for filed records, evidence filed in court, and high-privilege strategy discussions where cloud exposure is a concern. The honest framing is that AI handles the first 90% of the work cheaply and fast; the remaining 10% — the filed record, the certified transcript, the deposition transcript — still requires a human, often a specifically-certified one. This guide covers what fits where, attorney-client privilege considerations, a cost comparison across AI cloud, hybrid, and certified human services, and an honest landscape including Rev, Verbit, Ditto, GMR, Languages Unlimited, Sonix, VexaScribe, Voibe, and self-hosted Whisper.
Key takeaways
- →AI is not court-admissible as the official record. US courts require AAERT-certified court reporters or equivalent for filed transcripts. AI is a draft tool, not a court record tool. Plan accordingly.
- →The economics of AI transcription change what's feasible. AI at $0.30/hour effective vs certified human at $120-$300/hour. The 100-400× cost gap is why discovery review at scale, client intake transcription, and internal call notes are economically viable in 2026 in ways they weren't in 2020.
- →Attorney-client privilege is a real cloud consideration. Non-privileged material (client intake with consent, discovery audio, internal team calls) is generally fine in cloud AI with strong privacy posture. High-privilege strategy discussions benefit from offline or self-hosted tools where audio never leaves your device.
- →Hybrid AI+human services bridge the gap. Verbit and Rev Pro offer AI first pass + certified human review for legal-grade output at 30-50% the cost of pure human transcription. The credible path for deposition records where you want AI in the loop.
- →Legal terminology accuracy is good but not perfect. Whisper Large-v3 handles common legal vocabulary (deposition, voir dire, prima facie) well. Plan for 15-30 minutes of human review per hour of audio for legal use cases — case names, statute citations, and specialty terminology need careful verification.
- →Solo and small firms are where AI changes practice most. The economics work for solo and small firms in ways they don't for BigLaw, where certified human transcription is already a budget line item. AI lets a solo practitioner run workflows that previously weren't affordable.
Where AI transcription genuinely fits in legal workflows
Eight categories where AI transcription is appropriate, useful, and economically sensible for legal practice. Each is non-court-filing, non-sworn, non-certified — meaning the AI output is for internal use, prep, triage, or working draft purposes, not for the official record.
| Legal workflow | Why AI fits |
|---|---|
| Deposition prep | Internal-use only; you're preparing your witness, not creating the record. AI gives you the transcript fast and cheap; the actual deposition is recorded by a certified reporter. |
| Discovery review at scale | Hours of recorded calls, voicemails, body-cam audio, and meeting recordings need triaging fast. AI transcription makes them searchable for $0.30/hour instead of $120/hour. |
| Client intake interviews | Initial consultations recorded with client consent. AI gets you a searchable transcript for case file notes; no court filing involved. |
| Internal team calls | Strategy discussions, partner meetings, case status calls. Internal use, your own consent. AI is the obvious cost match. |
| Recorded phone calls (with consent) | Witness interviews, settlement negotiations recorded with proper consent. AI transcription for your file; not the court record. |
| CLE recordings and legal podcasts | Continuing legal education content you want notes from. No privilege issues; speed and cost matter more than certification. |
| Research interviews | Subject matter expert interviews, background research, legal scholarship. Same workflow as academic qualitative research. |
| Multilingual witness statement drafts | Non-English speaker statements that need a working draft before certified translation. AI gets you the gist; certified human translates for filing. |
The common thread: working draft, internal use, or triage of large audio volumes. The AI output is a productivity tool for the attorney, not a document being filed with a court or sworn to under oath. When the use case fits this pattern, AI transcription is the right answer for cost, speed, and workflow.
Where AI transcription does NOT fit
Six categories where AI is not the right answer — either because the law requires certified human transcription, the privilege threshold makes cloud AI inappropriate, or local rules specifically prohibit it.
| Workflow | Why AI doesn't fit |
|---|---|
| Certified court transcripts (filed records) | US courts require AAERT-certified or licensed court reporter transcripts for filed records. AI is not a court reporter. |
| Sworn depositions for filed records | Depositions used as filed evidence must be transcribed by a certified court reporter who can swear to verbatim accuracy. Hybrid AI+human services (Verbit Pro, Rev Pro) bridge this. |
| Evidence filed in court proceedings | Audio evidence transcribed for court filing requires chain-of-custody documentation and certified transcription. AI alone does not meet the standard. |
| High-privilege strategy discussions | Internal counsel work product, attorney-only strategy calls. Uploading these to cloud AI raises questions about privilege waiver and vendor access. Use offline tools (self-hosted Whisper, Voibe) or don't transcribe at all. |
| Sworn statements requiring certification | Any statement filed with a sworn certification of accuracy needs a human certifying official, not an AI output. |
| Jurisdictions explicitly prohibiting AI | Some jurisdictions and specific courts have local rules prohibiting AI transcription of certain proceedings. Check local rules before relying on AI output for anything filed. |
The general principle: if the transcript will be filed, sworn to, or used as the official record of a proceeding, you need certified human transcription — often a specifically-certified court reporter. If the transcript is for your own use, prep, triage, or internal workflow, AI is appropriate. The middle ground (hybrid AI+human services like Verbit and Rev Pro) covers cases where you want AI economics but legal-grade output.
AI vs human cost-quality comparison
Honest pricing snapshot as of June 2026. The 100-400× cost gap between AI cloud and certified court reporter services is the reason AI transcription has changed legal practice for non-court workflows — and the reason certified human services remain the right answer for filed records.
| Service | $/min | $/hour | Quality | When to use |
|---|---|---|---|---|
| VexaScribe (AI cloud) | ~$0.005 | ~$0.30 | 92-96% on clean legal English; review for legal-specific terms | Non-court workflows: prep, discovery, intake, internal |
| Sonix / Trint / Otter (AI cloud) | $0.10-$0.17 | $6-$10 | Similar AI accuracy; sentence-level timestamps | Similar non-court use cases; vendor preference |
| Rev AI (AI tier) | $0.25 | $15 | AI accuracy; sentence-level export defaults | When you're already in the Rev ecosystem for human tier |
| Verbit (AI+human hybrid) | $0.50-$1.00 | $30-$60 | AI first pass + certified human review; 99%+ accuracy | Legal-grade transcripts at lower cost than full human |
| Rev Pro (human verbatim) | $1.99 | $119 | Certified human; 99%+ accuracy; available for legal use | Sworn statements, depositions where AAERT not required |
| Ditto / GMR / Languages Unlimited (legal specialist) | $2-$4 | $120-$240 | Legal-specialized human transcription; certified | Filed evidence, jurisdictions requiring legal-specialty certification |
| AAERT-certified court reporter | $3-$5+ | $180-$300+ | Court-admissible verbatim; certified for filing | Depositions for filed records, trial transcripts, court proceedings |
Example: 100 hours of audio. AI cloud ≈ $30-$60. Hybrid AI+human (Verbit) ≈ $3,000-$6,000. Certified human (Rev Pro) ≈ $12,000. Court reporter ≈ $18,000-$30,000+. For discovery review at this volume, the choice between AI ($30) and court reporter ($30,000) decides whether the review happens at all on most matters.
See our general cost of transcription guide for non-legal context and a fuller breakdown of pricing models.
Attorney-client privilege and AI transcription
A real consideration that deserves honest treatment, not marketing-speak. When you upload audio to a cloud AI transcription service, the audio leaves your direct control — it's processed on the vendor's servers, encrypted in transit and at rest, and stored according to their retention policy. For non-privileged material, this is generally fine. For privileged material, the threshold is higher.
Non-privileged: cloud AI is generally appropriate
Client intake interviews recorded with explicit consent. Opposing party depositions or witness statements you obtained legitimately during discovery. Public-record audio. Recorded calls with proper consent where the content is factual, not strategic. For these categories, cloud AI services with strong privacy posture (encryption, no training on user audio, configurable retention) are an accepted workflow for many firms.
Privileged: higher threshold, three options
Strategy discussions between attorneys, internal counsel work product, attorney-client conversations you're transcribing for your own notes. Ranked by privacy strength:
- Self-hosted Whisper or on-device tools (Voibe). Audio never leaves the device. Strongest privacy posture for high-privilege material. Requires technical setup (Whisper) or Mac (Voibe).
- Cloud AI with strong privacy posture (VexaScribe). AWS eu-west-2 (London) hosting; no training on user audio; encryption in transit (TLS 1.2+) and at rest (AES-256); configurable retention. Acceptable for many privileged use cases where convenience trade-offs are documented and disclosed; consult firm compliance counsel for specific matters.
- Standard cloud AI (Otter, Sonix). Verify each vendor's training and retention policies before privileged use. Some vendors have used user data for model training historically; check current policy explicitly.
Practical rule: when in doubt about privilege, don't upload to any cloud service. Self-host or use on-device tools for high-privilege material. The cost difference between cloud AI ($0.30/hour) and self-hosted Whisper (free) plus a one-time hardware investment makes self-hosting a reasonable option for firms transcribing significant privileged material.
Vendor questions to ask
- ● Where is audio processed? US, EU, somewhere else? Implications for CLOUD Act, GDPR, and cross-border data transfer.
- ● Is user audio used to train AI models? If yes, walk away for privileged material. (Most current vendors say no, but verify.)
- ● What's the retention policy? Configurable retention or fixed? Can audio be deleted on request?
- ● Encryption at rest and in transit? TLS 1.2+ in transit, AES-256 at rest is the modern baseline.
- ● Vendor employee access? Can vendor employees access user audio? Under what circumstances?
- ● SOC 2 / ISO 27001 / similar certifications? Independent attestation of security controls.
- ● BAAs for HIPAA? Required for medical-legal work involving PHI.
8 use cases for AI transcription in legal practice
Concrete workflows where AI transcription is appropriate. All are non-court-filing — the AI output is the attorney's working tool, not the filed record.
1. Deposition preparation
You're preparing your witness for a deposition next week. You have recordings of your prep sessions (with client consent) and want a transcript to review with the witness, mark up, and use as the basis for further prep. AI transcription gets you the transcript in 10 minutes for under $1 of compute; you review it for legal-specific term accuracy, then use it for the actual prep work. The deposition itself will be recorded by a certified court reporter — that's the filed record.
2. Discovery review at scale
Opposing counsel produces 200 hours of recorded calls in discovery. Reviewing audio in real-time would take 200 hours of attorney time at $400/hour ($80,000). AI transcription converts the audio to searchable text in a few hours for under $100, lets your team grep for keywords and triage which calls need human review, and produces a working index. The relevant subset (maybe 10 hours of actual important audio) gets careful human review and, where filed, certified transcription. The 95% reduction in attorney triage time is the real economic win.
3. Client intake interviews
Initial consultation with a new client, recorded with explicit consent for case file purposes. AI transcription gives you searchable notes within minutes of the meeting ending; the file becomes part of your client matter system. Not for court filing — just for your own case notes and matter management.
4. Internal team calls
Weekly partner meetings, case status calls, strategy sessions where the partners are deciding next steps. Recorded internally, transcribed for your own records and follow-up. Pure internal use; AI is the obvious fit.
5. Recorded witness interviews
Pre-deposition or pre-trial witness interviews recorded with consent. The interviews aren't filed as evidence — they're your prep material. AI transcription lets you search, reference, and revisit witness statements during case prep. Note: these are not sworn statements and won't be filed; if they need to be filed later, a certified human transcription is what gets submitted.
6. CLE and legal podcast notes
Continuing legal education recordings, ABA conference sessions, legal podcasts. You want searchable notes and quotable excerpts. No privilege issues, no court filing — AI is the right tool for the job.
7. Background research interviews
Subject matter expert calls, retained expert preparation, scholarly research. Standard qualitative interview workflow, same pattern as academic research. AI transcription for the first pass; manual review for accuracy on technical terms.
8. Multilingual witness statement drafts
Spanish-speaking witness recorded statement. You need a working English draft before deciding whether to engage a certified translator for filed submission. AI transcribes in Spanish, AI translates to English for your review. If the statement is filed, certified human translation replaces the AI version; if not, the AI draft is your working file.
Decision framework: AI vs hybrid vs human
Three questions to decide which tier you need.
Question 1: Will the transcript be filed with a court, sworn to, or used as the official record?
If yes → certified human transcription. For depositions and trial transcripts, AAERT-certified court reporter. For other filed transcripts, legal-specialist services (Ditto, GMR, Languages Unlimited) or Rev Pro depending on the matter. AI is not appropriate as the sole source for filed records.
If no → proceed to Question 2.
Question 2: Is the content highly privileged (attorney-only strategy, internal work product)?
If yes → offline or self-hosted AI. Self-hosted Whisper, Voibe on-device, or a cloud AI tool only after vetting privacy posture with firm compliance counsel. Default to offline when in doubt.
If no → proceed to Question 3.
Question 3: Do you need legal-grade accuracy (99%+) or is “good enough for triage” acceptable?
Legal-grade required (e.g., transcripts to be referenced in motions, statements that will inform sworn filings) → hybrid AI+human services (Verbit, Rev Pro). AI first pass, certified human review, signed-off transcript.
Triage / prep / internal use acceptable → pure AI cloud (VexaScribe, Sonix, Otter, Rev AI). The 92-96% accuracy on clean legal English is adequate for prep, discovery triage, intake notes, and internal records. Plan for human review of legal-specific terminology.
The framework is conservative on filed records and privilege, permissive on internal workflows. That's the honest pattern in 2026 legal practice.
Honest comparison of legal transcription options
Snapshot of the landscape as of June 2026. Strengths and weaknesses described honestly — including ours. Verify pricing and policies with each vendor before committing.
Certified court reporter (AAERT)
Certified humanStrength: Court-admissible for depositions and trial transcripts; chain-of-custody documentation; expert testimony on accuracy if challenged.
Weakness: $3-$5/min; scheduling-dependent; the only option for filed court records.
Ditto Transcripts
Legal-specialist humanStrength: Specialized in legal transcription including depositions, court hearings, witness statements; certified for legal use; supports legal terminology and formatting standards.
Weakness: Premium pricing for legal specialty; turnaround times longer than AI tools.
GMR Transcription
Legal-specialist humanStrength: Established legal transcription with certified human transcriptionists; HIPAA-compliant options for medical-legal work; legal formatting expertise.
Weakness: Standard human pricing ($1.50-$3/min depending on turnaround and certification level).
Languages Unlimited
Legal-specialist humanStrength: Multilingual legal transcription specialty; certified translations for non-English depositions and statements; one-stop shop for multilingual legal work.
Weakness: Premium pricing; turnaround scales with language and certification requirements.
Rev (Human tier)
Certified humanStrength: $1.99/min for human transcription; legal-friendly turnaround; widely used by law firms for client intake transcripts and non-deposition use cases.
Weakness: Not court-reporter-certified by default; for depositions and trial transcripts, hire a court reporter directly.
Verbit
AI+human hybridStrength: AI first pass + certified human review; legal vertical specialty; cost-effective alternative to full human transcription for legal-grade output; HIPAA-ready workflows.
Weakness: Higher cost than pure AI; enterprise focus with sales-led pricing.
Sonix
AI cloudStrength: Solid AI transcription with structured editing; legal terminology handling adequate for prep and intake; sentence-level timestamps standard.
Weakness: $0.10-$0.17/min pricing higher than VexaScribe; sentence-level by default in many exports.
VexaScribe
AI cloudStrength: Whisper Large-v3 with word-level timestamps; AWS eu-west-2 (London) hosting with GDPR-aligned posture; no training on user audio; $2-$20/month subscription; well-suited to solo and small-firm economics for non-court workflows.
Weakness: Cloud-hosted (audio leaves your device); not appropriate for high-privilege strategy material or court-filed records; not HIPAA-BAA-signed in 2026.
Voibe (on-device)
Self-hosted AIStrength: On-device transcription; audio never leaves the user's machine; ideal for high-privilege material where cloud exposure is a concern; one-time purchase model.
Weakness: Mac-only; on-device processing slower than cloud for long recordings; less editor sophistication than cloud platforms.
Self-hosted Whisper / WhisperX
Open-source AIStrength: Free; runs entirely on your hardware; full control over data; same Whisper Large-v3 model that powers most cloud tools; ideal for highest-privilege material.
Weakness: Requires technical setup; no built-in editor or collaboration; you're responsible for IT, storage, and updates.
Pattern: certified human services (court reporters, Ditto, GMR, Languages Unlimited) dominate filed records. Hybrid services (Verbit, Rev Pro) bridge legal-grade accuracy with AI economics. Pure AI cloud (VexaScribe, Sonix, Otter, Rev AI) is the right tool for non-filed legal workflows. Self-hosted (Voibe, Whisper) is the right answer for high-privilege material.
Where VexaScribe genuinely fits — honestly
We're an AI cloud transcription tool. We are not, and do not claim to be, a court reporter service or a hybrid AI+human service. Where we fit in legal workflows:
VexaScribe is a good fit for:
- ● Solo practitioners and small firms where $1.99/min human transcription isn't economically sustainable for routine workflows. $2-$20/month covers a meaningful volume of non-court audio.
- ● Discovery review at scale. Hundreds of hours of recorded calls, voicemails, body-cam audio. AI transcription makes them searchable for triage at a cost that makes the review feasible.
- ● Client intake interviews recorded with explicit consent. Searchable case file notes, not court filings.
- ● Internal team calls. Partner meetings, case status, strategy discussions where the content is not high-enough privilege to require offline tools.
- ● Multilingual witness statement drafts. 99 supported languages via Whisper Large-v3; AI transcription + AI translation for a working draft before certified human translation for filing.
- ● CLE and legal podcast notes — no privilege, no filing.
- ● Research interviews — same workflow as academic qualitative research.
VexaScribe is NOT a fit for:
- ● Court-filed transcripts. Use an AAERT-certified court reporter or legal-specialist service (Ditto, GMR, Languages Unlimited).
- ● Sworn depositions for filed records. Court reporter or hybrid service (Verbit Pro, Rev Pro).
- ● High-privilege strategy material where cloud exposure raises privilege concerns. Self-host Whisper or use Voibe on-device.
- ● Medical-legal work involving PHI. We do not currently sign BAAs under HIPAA. Use Verbit (HIPAA-ready) or specialized medical-legal services.
- ● Jurisdictions explicitly prohibiting AI for the relevant proceeding. Check local rules.
Privacy and security posture
- ● Hosting: AWS eu-west-2 (London, UK)
- ● Encryption: TLS 1.2+ in transit, AES-256 at rest
- ● Training: User audio is not used to train AI models
- ● Retention: Configurable; default retention disclosed in our privacy policy
- ● Model: Whisper Large-v3, word-level timestamps
- ● Languages: 99 supported via Whisper
- ● Pricing: $2-$20/month subscriptions; no per-minute billing surprises
If your matter is non-filed, non-sworn, and either non-privileged or appropriately consented for cloud processing, we're a good fit. If not, the options above (court reporter, Verbit, Ditto, GMR, Voibe, self-hosted Whisper) are the right place to look.
Specific use case: discovery review at scale
The use case where AI transcription most changes legal practice in 2026. The pattern:
- Receive discovery audio. Opposing counsel produces 100-500 hours of recorded calls, voicemails, body-cam audio, deposition recordings (which may already be transcribed), and meeting recordings.
- Bulk transcribe with AI. Upload all audio to a cloud AI service. See our bulk transcription guide for the practical mechanics. Cost: ~$30-$150 for 100-500 hours.
- Make transcripts searchable. Export to a format your case management system or document review tool can index. Most discovery platforms (Relativity, Disco, Everlaw) accept TXT or DOCX with metadata.
- Triage by keyword search. Search for case-relevant terms, names, dates, locations, products. Identify the subset of audio that's actually relevant — typically 5-20% of the total volume.
- Human review the relevant subset. Attorneys (or contract reviewers) listen to and verify the relevant audio. The AI transcript is a navigation tool — they jump to specific moments using timestamps.
- Certified transcription only where needed. For audio that will be cited in motions or filed as exhibits, engage certified human transcription on the specific clips that matter. Not the full 100-500 hours — just the 5-20 relevant clips that will actually be filed.
The economic argument. Pre-AI workflow: attorneys listen to 500 hours of audio at $400/hour = $200,000 in attorney triage time, before any actual case work. Post-AI workflow: $150 of AI transcription + ~50 hours of focused attorney review on the relevant subset = $20,000 in attorney time. The ~90% reduction in triage cost is what makes thorough discovery review of audio-heavy productions feasible.
See related: interview transcription (similar workflow, different vertical), transcription timestamps (for jumping to specific moments in audio during review), speaker labeling (for multi-party calls).
Frequently asked questions
Is AI transcription court-admissible?
Generally no, not as the official record. US courts require certified verbatim transcripts from court reporters or AAERT-certified electronic reporters for filed records, sworn depositions, and trial transcripts. AI transcription is widely used as a draft — for prep, internal review, and discovery triage — but the certified record still requires a human in 2026. Some jurisdictions have started accepting AI-assisted transcripts when reviewed and certified by a qualified human; check your local rules. The practical workflow: AI does the first pass cheaply and fast; a human court reporter certifies the final filed transcript. Treating AI transcripts as 'court-ready' without human certification creates real evidence-admissibility risk.
Can I use AI transcription for a deposition?
For prep yes, for the official record no. Pre-deposition work (witness prep recordings, your own prep notes, transcribing background interviews) is exactly where AI transcription fits — fast, cheap, internal-use. The deposition itself must be recorded and transcribed by a certified court reporter who can swear to verbatim accuracy; AAERT certification and state-specific court reporter licensing are typical requirements. Some hybrid services (Verbit, Rev Pro) offer AI-assisted human-certified deposition transcripts where the AI handles the first pass and a certified reporter reviews and signs off — that's the credible path if you want AI involvement in actual deposition records.
What about attorney-client privilege when using AI transcription?
Real consideration. When you upload audio to a cloud AI transcription service, the audio leaves your control — it's processed on the vendor's servers, encrypted in transit and at rest, and stored according to their retention policy. For non-privileged conversations (recorded client interviews with explicit consent, public records, opposing party depositions you obtained legitimately), cloud AI is generally acceptable. For privileged communications (your strategy discussions, internal counsel work product, attorney-client conversations you're transcribing for your own notes), the threshold is higher. Three options ranked by privacy strength: (1) Self-hosted Whisper or Voibe — audio never leaves the device, ideal for highest-sensitivity privileged material. (2) Cloud AI with strong privacy posture (VexaScribe: AWS eu-west-2 hosting, no training on user audio, configurable retention, encryption everywhere). (3) Standard cloud AI (Otter, Sonix) — fine for non-privileged, verify their training and retention policies before privileged material. When in doubt, don't upload privileged content to any cloud service.
How accurate is AI transcription for legal terminology?
Whisper Large-v3 (the model VexaScribe and most modern AI tools use) was trained on a broad audio corpus that includes legal proceedings, news, and professional speech. Accuracy on clean legal audio runs 92-96% for general English speakers and US legal terminology. Specialized terms (case names, statute citations, foreign-language witness statements) require review — Whisper is good at common legal vocabulary (deposition, voir dire, prima facie) but less reliable on case-specific terminology (specific party names, novel statutory references). Plan for 15-30 minutes of human review per hour of audio for legal use cases. For Latin terms, foreign-language testimony, or jurisdiction-specific procedural language, the review time may exceed 30 minutes. Honest framing: AI is excellent for the first 90% of the work, and the remaining 10% (legal-specific terms, case citations) is where careful human review matters.
How does AI handle multiple speakers in legal recordings?
Speaker diarization (identifying who's speaking when) is included in modern AI transcription tools, with realistic accuracy varying by audio quality. With separate microphones and minimal speaker overlap, diarization is 90-95% accurate at labeling speakers correctly. With a shared room microphone, mid-call interruptions, or background noise typical of legal recorded calls (witness testimony, recorded phone calls, deposition audio captured by court reporters), accuracy drops to 75-85%. The AI doesn't know who anyone is by name — it labels speakers as 'Speaker 1', 'Speaker 2', etc. You rename them in the editor (Attorney Smith, Witness Jones, Opposing Counsel) after the first pass. For high-stakes legal work, verify speaker attribution against the audio before relying on it; the most common diarization error is misattributing brief interruptions ('objection' from one attorney attributed to the witness, for example).
Can AI transcribe witness statements in other languages?
Yes — VexaScribe and other Whisper-based tools support 99 languages including major legal-relevant ones (Spanish, Mandarin, Arabic, French, Portuguese, Russian, Vietnamese, Korean, Hindi, German). For witness statements in non-English languages, the workflow is: (1) Transcribe in the original language with a competent AI tool. (2) Either get a certified human translation if the transcript will be filed, or use machine translation for prep/internal use. (3) For court-admissible filings, you need a certified human translator — AI translation is not legally certifiable in 2026. For prep work, Whisper transcription plus AI translation gets you 90% of the way to a useful witness statement summary for $5 of compute instead of $200 of human translation. Plan for verification of names, dates, and key statements against the original audio.
What's the cost difference between AI and certified human legal transcription?
Roughly 10-50x at base prices. AI cloud (VexaScribe Basic: 1,000 minutes = ~16.7 hours for $5/month): effectively $0.30/hour for transcription. Hybrid AI+human review (Verbit, Rev Pro): $0.50-$1.00/min ($30-$60/hour). Certified human transcription (Rev human): $1.99/min ($119/hour). Court reporter services (AAERT-certified, deposition-grade): $3-$5/min ($180-$300/hour), often with rush surcharges, expedited delivery, certified copies, and per-page litigation costs. For 100 hours of audio: AI = ~$30-$60, hybrid = ~$3,000-$6,000, certified human = ~$12,000-$30,000. The cost difference is why AI legitimately changes legal practice for discovery review, internal prep, and non-court workflows; the same cost gap is why certified human services remain the right choice for filed records and depositions.
Should solo lawyers and small firms use AI transcription?
For non-court workflows, almost always yes. Solo practitioners and small firms are exactly the audience for whom AI transcription is most economically meaningful — paying $1.99/min for human transcription of client intake interviews or internal team calls is unsustainable for a five-person firm with a hundred recorded conversations a year. AI at $0.30/hour effective covers the workflow at a sustainable cost. The honest constraint: for any recording that might be filed as evidence, used as the basis for a sworn statement, or referenced in a deposition transcript, plan to engage a certified human service or court reporter for the official record. For everything else (prep, drafts, research, internal calls, client intake, discovery audio triage), AI is the right answer for solo and small-firm economics.
Is VexaScribe HIPAA-compliant for medical-legal cases?
Not currently, and we're honest about that. VexaScribe is hosted on AWS eu-west-2 (London, UK) with GDPR-aligned security posture, encryption in transit (TLS 1.2+) and at rest (AES-256), and no training of AI models on user audio. We do not currently sign Business Associate Agreements (BAAs) under HIPAA. For medical-legal work involving Protected Health Information (PHI) — medical record summarization, healthcare provider depositions, personal injury matters involving medical records — you need a HIPAA-compliant transcription service with a signed BAA. Specialized providers (DAX Copilot, Nuance Dragon Medical, Augnito) and some general transcription services (Rev offers HIPAA-compliant options on enterprise plans, Verbit has HIPAA-ready workflows) cover this. For non-PHI legal work, our standard security posture is generally acceptable; consult your firm's compliance counsel for specific matters.
Methodology & disclosure
Sources: Court reporter certification requirements verified against the AAERT (American Association of Electronic Reporters and Transcribers) standards and state-by-state court reporter licensing boards. Pricing verified against each vendor's public pricing pages as of June 2026 (where pricing is published) or industry-standard rates (where pricing is sales-led). Whisper Large-v3 accuracy on legal English benchmarked against publicly-available legal audio (CourtListener archives, public domain deposition recordings) and internal VexaScribe testing. Attorney-client privilege considerations informed by ABA Formal Opinion 477R on confidentiality and technology and state bar opinions on cloud computing.
Disclosure: This page is published by VexaScribe. We have a commercial interest in legal use of AI transcription for non-court workflows. We do not have a commercial interest in framing AI as a replacement for court reporters — we've been explicit throughout this page that it isn't. The hybrid services (Verbit, Rev Pro), legal specialists (Ditto, GMR, Languages Unlimited), and self-hosted alternatives (Voibe, Whisper) are described honestly because they're genuinely the right answer for different use cases.
Not legal advice: This page describes general industry patterns. Specific matters require specific advice from compliance counsel familiar with your jurisdiction, your firm's policies, and the particular case. AI transcription policy varies by jurisdiction and is evolving in 2026; verify local rules before relying on AI output for any filed work.
Editorial standards: See our editorial standards.
Related guides
Bulk transcription
Mechanics of transcribing high volumes — the discovery review pattern
Interview transcription
Witness interviews, client intake, recorded calls
Transcription timestamps
Word-level timing for jumping to specific moments during review
Speaker labeling
Diarization for multi-party calls and witness statements
Cost of transcription
General pricing context across AI, hybrid, and human services
AI vs human transcription
When to choose which — broader treatment beyond legal