Rev vs Verbit: how they compare in 2026
Rev and Verbit both turn recorded legal proceedings into transcripts with their own speech recognition engines, and both sell human review too. Verbit sits in the top two bands on fourteen of fifteen axes and Rev on eight of fifteen. The widest gap is privilege. Verbit's terms carry a clause for privileged materials: access limited to staff who need it, no use beyond the service, return or destruction on request, and a statement that nothing in the agreement waives privilege. Law firm deposition material stays out of its general speech recognition training unless the firm opts in, and no customer content trains a generative model. Rev's terms say customer content may be used for continuous training of Rev's own speech recognition and other AI models, with no opt out located, while excluding generative model training. Verbit also marks the AI only output on its entry plan as not court admissible. Rev's counterweight is reach and price: its AI analyzes up to 500 case files at once with every finding cited to a timestamp or page and line, and it publishes per seat prices for every plan but its top one.
At a glance
All 15 axes, side by side
The same grid applied to every vendor in the index, graded from public sources. Hover a grade to see what the letter means on that axis.
AI Centrality
How much of the product is actually AI. Whether the machine learning is the mechanism the buyer is paying for or a feature layered onto conventional software, and whether the vendor is specific about which is which.
The models are the product. Rev's own speech recognition models transcribe every file, and its AI analyses whole case files at once, drafts summaries, chronologies, affidavits and briefs in an agentic editor, and searches case photos. Human transcription, captions and deposition summaries are sold beside the platform, but the platform itself is AI throughout. Verified 22 September 2026.
The models are the engine of the product and the business survives without them, which is B. Captivate is the vendor's own automatic speech recognition engine, trained on legal terminology, adapting to customer vocabulary, speakers and acoustic conditions, and Gen.V is a generative layer producing summaries, chapters, keyword tagging and inconsistency detection. Every legal plan is priced and described around that engine. What holds it off A is that Verbit also sells the same output produced by people: human review, certified transcripts and the VITAC, Take Note and AST brands are human transcription operations, and the Terms describe the Services as processing live and pre-recorded media using ASR or human options. Remove the models and a transcription business remains, which is what the court reporting incumbents sell. Verified 12 September 2026.
Citation Accuracy and Hallucination Disclosure
Whether the vendor publishes measured accuracy on citations and assertions, grounds output to primary sources, and says plainly what its system does when it does not know. Legal has a documented public record of fabricated citations reaching filed briefs, so an untested claim of accuracy is not evidence.
Grounding is real and described, short of measurement of the analysis. Rev says its AI draws findings only from the files uploaded to a case rather than the internet, and that every finding links to a citation with the exact timestamp and source file, with page and line citations for depositions, so a user can check it in seconds. It publishes transcription accuracy as 96% or better for AI transcripts and 99% or better, guaranteed, for human transcripts, without the test behind the AI figure. No error rate is published for multi-file analysis or drafting. Verified 22 September 2026.
R15 governs: the product cites no legal authority, so the authority-grounding and citator limbs do not apply. What bites is grounding and measured accuracy. Grounding is unusually strong in the product's own idiom: output is time-coded and speaker-identified against the source recording, so every word resolves to the audio a reader can replay, and a published Post-Production Corrections Policy exists for fixing output. The hallucination position is stated contractually rather than in marketing: Terms section 4(d) states that AI Analysis Features use probabilistic models that may produce inaccurate, incomplete or misleading results, do not constitute professional advice of any kind, are not a substitute for qualified human review, and that the customer is solely responsible for verifying all output before relying on it. What keeps this off A is measurement: no accuracy figure, test set or error rate is published on any surface read, and the accuracy commitments in section 7(f) are targets conditioned on reasonable audio quality as determined by the vendor. Verified 12 September 2026.
Autonomy and Oversight Model
What the system decides on its own, what a lawyer must approve, and whether the vendor documents where the review point sits. A tool that drafts under review and a tool that files without one are different products and different risks.
Cited findings and a written verification duty, short of stated limits. Every finding is linked to its source for review, drafts open in an editor, and human-verified transcription is offered where accuracy matters most. The Terms of Service make the customer solely responsible for verifying the accuracy and completeness of all work product before acting on it. The agentic editor drafts documents directly from an instruction, and nothing states what it may not be used for without review. Verified 22 September 2026.
Modes, the constraint, the review surfaces and the route back to human judgement are all published, which is the A band, and the constraint is stated more plainly here than anywhere else in this corpus. The modes are the product tiers themselves: AI-only transcription, AI with optional human review, and certified human-completed transcripts, each described and priced separately. The constraint is categorical rather than a numeric threshold and it is printed on the plan: the self-serve Standard legal plan states in its own feature list **Not court admissible**, so a buyer is told before purchase what the unsupervised output may not be used for, with court-admissible transcripts available by request on the higher tiers and an upgrade path to certified transcripts named on each. The review surfaces are the editing workspace, immediate rough drafts against final transcripts, and the Post-Production Corrections Policy incorporated into the Terms. The route back to human judgement is contractual as well as commercial: Terms section 4(d) places verification of all AI output on the customer and states the features are not a substitute for qualified human review. R15 applies to the threshold limb: on a transcription product the boundary is a document-class boundary rather than a confidence score, and the vendor draws it explicitly. Verified 12 September 2026.
Operational and Outcome Evidence
Named, dated evidence that the product works in production at real firms or legal departments. Case studies with figures and identified customers count. Unattributed testimonials and launch announcements do not.
Named law firms with figures, on the customers' own account. Case studies describe Doubek, Pyfer & Storrar cutting legal drafting time by 97%, Greening Law Group's review of body-camera footage that used to take 10 to 30 hours, and Anderson Law cutting evidence review time by 94%. The figures are reported results rather than measurements with a stated method, and the studies are not dated on the pages read. Rev says it serves more than 1 million users and 125,000 organisations. Verified 22 September 2026.
Figures without the named customers, which is the B band's stated shape. The vendor publishes that it serves more than 3,000 businesses and institutions, operates the VITAC, Take Note and AST brands as its own, and states a 99.0% monthly platform availability target in Terms section 7(b) with defined exclusions, which is a measurable commitment rather than a claim. Turnaround times are committed in the Order or on the site and calculated in US Eastern business days. What is absent is attribution: no named law firm, court or court reporting agency appears with a figure for what changed on any surface read, and no dated deployment case study for the legal line was located. Analyst and award placement is not treated as deployment evidence. Verified 12 September 2026.
Privilege and Confidentiality Posture
How client confidences are handled: attorney client privilege and work product treatment, segregation of one client matter from another, whether client data trains any model, and what the vendor commits to in writing rather than in marketing.
Confidentiality is addressed in the agreement, alongside a right to train Rev's own models. The Terms of Service bind both parties to confidentiality with prompt written notice of any subpoena, court order or government demand, human transcriptionists are ID-verified and sign NDAs, and the security page says third-party models are never trained on customer data. The same terms say customer content is analysed by Rev's speech recognition and other AI models and may be used for continuous training of those models, which Rev keeps in-house, while excluding it from generative AI training; no opt-out was located. Privilege, the retention period and the identity of the generative models used are not addressed. Verified 22 September 2026.
The first record in this corpus to satisfy the privilege limb R33 requires, and it does so contractually rather than in marketing. Terms section 5(c), headed Privileged Materials, applies where the customer submits files containing attorney-client privileged or work product materials and commits the vendor to four things: limit access to personnel with a direct need to know for providing the Services; not disclose to any third party except as required to provide the Services and then under equivalent confidentiality protections; not use them for any purpose other than providing the Services and customer-specific optimisation; and return or destroy them on termination or on request. It adds that nothing in the agreement constitutes a waiver of any privilege held by the customer or its clients, and places responsibility for deciding whether submission is consistent with professional conduct rules and privilege law on the customer. Work product is named expressly, which is the limb MyCase and Clio were held at B for missing. The other A limbs are met: no generative AI training on customer content at all under section 4(c), and no general ASR training on law firm materials without written opt-in under 4(e); retention and deletion in DPA section 8, with delete or return within 30 days of termination at the customer's choice; access control documented control by control on the Trust Page; and an explicit position on third-party model providers, who may only be used if contractually prohibited from training on customer data and must not retain data beyond the minimum their own terms require. Verified 12 September 2026.
UPL and Professional Responsibility Posture
Whether the vendor is clear that it supplies a tool rather than legal advice, who its audience is, and how it addresses unauthorized practice of law, competence and supervision duties, and jurisdiction limits. ABA Formal Opinion 512 is the reference point. Where the advice line is not the duty a product raises, the axis is read through the nearest professional duty it does raise: judicial conduct rules and the reviewing duty for products sold only to courts, and the duty to bill for time actually spent for products that draft time entries.
A verification clause against drafting marketed for legal use. The Terms of Service say the services use AI, speech recognition or human work and make the customer solely responsible for verifying all work product before acting on it. Nothing states that Rev gives no legal advice or addresses supervision, while the platform is sold to draft legal briefs, affidavits and investigative summaries for filing or sending. Verified 22 September 2026.
A real published position on the advice line, short of the supervision and competence dimension. Terms section 4(d) states that AI Analysis Features do not constitute professional advice of any kind, are not a substitute for qualified human review, and that the customer is solely responsible for verifying all output before relying on it. Section 5(c) goes further than most by naming the customer's rules of professional conduct and privilege law as the customer's responsibility to assess. The product-class equivalent of the advice line is the admissibility line, and the vendor draws it: the AI-only plan is marked not court admissible, certified transcripts are a named upgrade, and jurisdictional formatting and cover pages are described as what makes a transcript admissible. What is missing for A: nothing addresses the supervising lawyer's competence duties when relying on machine transcription of testimony, nothing states who may use the service, and nothing addresses the state-by-state licensure position for digital court reporting, which is the live professional-responsibility question in this product class. Verified 12 September 2026.
AI Governance and Bias Disclosure
Published governance over model behavior: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.
Checked the home, security, About, law firm and platform pages, the Terms of Service and the Privacy Policy on 22 September 2026. No AI governance framework, accountable owner, testing before release, or finding on how transcription or analysis performs across accents, languages or audio quality was located. Verified 22 September 2026.
A governance framework with real substance, short of testing results and a named owner, which is B. What is unusual is that the AI governance is contractual rather than a principles page: Terms section 4 is a dedicated Artificial Intelligence section covering customer-specific optimisation, general ASR training with de-identification requirements, a prohibition on generative AI training, generative AI features, legal-specific provisions and FERPA, and section 4(g) commits the vendor to technical and organisational measures to prevent re-identification of training data, restrict access to identifiable data, and ensure data in AI Analysis Features is encrypted in transit and not retained by third-party providers. The DPA adds a commitment to ensure automated processing does not produce legal effects concerning or significantly affect data subjects without appropriate safeguards. Security governance is documented separately and in detail: an ISMS led by a named CISO role, risk assessments reviewed by senior leadership at least annually, policies reviewed annually, and annual plus quarterly internal audit programmes. What is absent: no accountable owner for AI is named, no evaluation or testing results are published, no model card exists, and nothing addresses uneven output across accents, dialects or speaker groups, which is the bias question that matters for a product transcribing testimony. Verified 12 September 2026.
AI Safety and Data Stewardship
Retention, deletion, access control, and what happens to prompts and documents after they are processed. Whether the vendor states its subprocessors and its incident practice, or leaves the buyer to assume.
Security controls are published; retention, subprocessors and incidents are not. The security page describes SSO, multi-factor authentication and domain claiming, encryption in transit and at rest on Amazon S3, backups several times a day, file-level sharing permissions, and ID checks and NDAs for human transcriptionists. The Terms of Service let Rev set maximum retention periods without stating them, and the Privacy Policy keeps personal data as long as necessary, with backups kept longer. No subprocessor list or incident notification commitment was located. Verified 22 September 2026.
Four of the five A limbs are contractual and specific, and the fifth is the gap. Retention and deletion: DPA section 8 requires deletion or return of all personal data within 30 days of termination at the customer's choice, indicated through the Services or in writing, with one copy retained solely for the establishment or defence of legal claims, and the Trust Page states a customer can have data removed at any time through its account manager. Access control is documented in unusual detail: encryption in transit by enforced HTTPS and at rest in AWS S3 using AES-256, two-factor authentication, customer-assignable roles, least privilege internally, regular access reviews, and joiner-mover-leaver rules. Incident practice is contractual: notification without undue delay after becoming aware of a data incident, with sufficient information for the customer to meet its own notification obligations, backed by a CISO-led incident response function and a 24/7 security operations centre. The fifth limb holds it at B: **no subprocessor list is published**. The DPA commits the vendor to notify before authorising a new subprocessor, to make the then-current list available on request, and to a five business day objection window, but a buyer cannot see who processes its recordings before contracting. Verified 12 September 2026.
AI Liability and Recourse
What the vendor stands behind contractually when its output is wrong. Indemnities, caps, carve outs, insurance, and whether any of it is published or only reachable through a negotiated agreement.
Addressed only through a standard limitation clause. The Terms of Service exclude consequential and indirect damages and limit Rev's liability, disclaim warranties for AI, speech recognition and human work, and require the customer to indemnify Rev for claims arising from its content and use. Nothing stands behind a transcript or analysis that proves wrong. Verified 22 September 2026.
Specific caps and a real indemnity, short of the exposure the product creates. Terms section 8 caps each party's liability at the lesser of USD 1,000,000 or fees actually paid or accrued in the 12 months preceding the claim, excludes indirect and consequential damages, and carves out intellectual property violations and the customer's payment obligations from the cap. Section 6(b) is a three-limb supplier indemnity covering third-party IP infringement arising from authorised use, unauthorised disclosure of confidential information or personal data to the extent caused by the vendor's negligence or failure to adhere to its published Security Standards, and violation of law caused by the vendor's provision of the Services; the second of these is a genuine indemnity tied to the Trust Page as the standard. Service levels are committed at 99.0% monthly availability and a Post-Production Corrections Policy is incorporated for defective output. What keeps this off A is the exposure that matters here: nothing allocates the loss when a transcript of testimony is wrong. Section 4(d) does the opposite, placing verification of all AI output on the customer, and section 7(g) disclaims all warranties including fitness for purpose. No insurance position is published. Verified 12 September 2026.
Practice Systems Integration Depth
How deeply the product reaches into the systems legal work already lives in: document management such as iManage and NetDocuments, Word and Outlook, contract lifecycle management, matter management, e-billing, and court filing systems.
Named connections for getting files in, short of documented depth. Files can be pulled from Clio, Dropbox, Box, Google Drive or OneDrive, the Pro plan adds a Clio Manage integration, and a speech-to-text API is documented separately at rev.ai. What syncs back to practice systems and how each connection is configured is not described on the pages read. Verified 22 September 2026.
Real integrations, documented, short of depth across the legal stack. A public API with developer documentation is published at verbit.readme.io, which is more than most records in this corpus offer, and API access is named as an inclusion on the Custom plans. Named integrations include Relativity on the Legal Visor plan, which is the one that reaches an eDiscovery platform where litigation work already lives, plus Zoom and Teams for live capture, cloud storage destinations, and a mobile application for recording on the move. The legal transcription material states API support to connect transcripts directly to case management and document systems. What is not established: the case management and document systems are not named individually, the integrations page was not opened, and the recurring phrase dozens of video platforms, video conferencing and cloud storage integrations describes breadth without a list an implementer could work from. Verified 12 September 2026.
Deployment Model and Data Residency
Where the software runs and where the data sits. Multi tenant cloud, single tenant, private deployment, on premises, and whether region of residence is a published option or an enterprise conversation.
The hosting provider is named; region and tenancy are not. The security page says files are encrypted at rest with Amazon S3 server-side encryption. No region, tenancy model, processing location or deployment option other than the hosted service is stated. Verified 22 September 2026.
Real residency detail, short of a full picture, which is B. Delivery is cloud on AWS, stated on the Trust Page, with platforms replicated across multiple availability zones and regions behind the AWS uptime SLA. The privacy policy names the processing locations: the United States, Europe, the United Kingdom, Israel and other locations as necessary to deliver the Services, and states that customer data may only be processed in the locations permitted in the data processing addendum and other commercial agreements, so residency is a contractable term rather than a fixed one. Cross-border transfers are handled in DPA Schedule 2 with the EU Standard Contractual Clauses under Modules Two and Three governed by Irish law, the UK Addendum, and Swiss adjustments, plus Part 4 additional safeguards including a commitment to resist bulk surveillance requests. What is absent for A: the tenancy model is never stated, no region menu or single-region option is published, and the split between where data is processed and where it is stored is not described. Verified 12 September 2026.
Security Certifications and Trust Center
Independent attestation a buyer can pull without a sales call: SOC 2, ISO 27001, penetration test summaries, a trust center with current reports and named scope rather than a badge image.
Named standards, short of a report a buyer can reach. The security page states SOC 2 Type II compliance, CJIS and HIPAA compliance, and that a SOC 3 report is available for review. No auditor, report period or link to the SOC 3 report was found on the pages read. Verified 22 September 2026.
Certifications are real, named with versions and scoped, and the evidence is gated, which is B. The Trust Page states that ISO 27001:2022 certification and SOC 2 attestation are available for certain services and configurations, and directs a buyer to an account manager or support for details; the scoping caveat is itself a disclosure and is recorded rather than penalised. Around it sits the most detailed published security programme of any record in this pull: infrastructure monitoring through GuardDuty, CloudWatch and CloudTrail, cloud security posture management with infrastructure-as-code scanning, a SIEM, a CISO-led incident response team with a 24/7 external security operations centre, published vulnerability remediation targets of 48 hours for critical and 14 days for high severity, a documented secure development lifecycle with code review and OWASP Top 10 alignment, and an independent external penetration test commissioned at least annually. The DPA adds a contractual audit right on 14 days' notice once every 12 months, including by the customer's own third-party auditor. What holds it off A: the certificate and report are obtainable only through a sales or support conversation, no auditor is named, no report period or certificate number is published, and there is no self-serve trust portal. Verified 12 September 2026.
Model Supply Chain Disclosure
Which models sit underneath, whose they are, where they run, and whether the vendor commits to telling customers when that changes. A legal buyer inherits every dependency it cannot see.
Rev's own models are described; the generative ones are not identified. The Terms of Service say Rev's speech recognition and other AI models are proprietary and maintained in-house, and that its AI services also rely on generative AI models, without naming those models or their providers. No change notification is published. Verified 22 September 2026.
The vendor's own engine is named and what sits underneath the generative layer is not, which lands on C. Captivate is named as the proprietary automatic speech recognition engine and its behaviour is described, including domain adaptation and customer-specific optimisation. Beyond that the disclosure stops at a category: the data policy states that Verbit integrates large language models to enhance transcription and analysis, and Gen.V is named as the generative suite, but no model and no model provider is identified anywhere on the estate. What partially offsets this, and is recorded rather than credited to the band, is that the position is contractual: Terms section 4(c) permits only third-party AI model providers that are prohibited from training on customer data, and the DPA requires any such provider to be listed as a sub-processor with notification before it begins processing and a five business day objection window. So a customer can learn who the providers are and object to them, but only after asking, and the sub-processor list is not published. Verified 12 September 2026.
Commercial Transparency
Whether a buyer can learn what this costs without entering a sales process: published rates, the unit being charged, what sits behind an enterprise tier, and what implementation adds.
Nearly every tier is priced; Unlimited is quoted. The plan table publishes a free tier, Essentials at $25.49 per seat per month billed annually or $29.99 monthly, and Pro at $47.99 billed annually or $59.99 monthly, with file and minute limits; human transcription is $1.99 a minute. The Unlimited plan, with CJIS and HIPAA controls and SSO, is priced by a specialist. Verified 22 September 2026.
Real pricing published for part of the range, with the structure published across all of it. The legal Standard plan is priced at $24 per month for 100 hours, being 50 live and 50 post-production, with a self-serve registration path rather than a sales conversation, and the plan page sets out four legal tiers side by side with a line-by-line feature comparison covering hours, languages, integrations, user counts, security inclusions, admissibility and upgrade paths. The same structure is published for the media, corporate and education verticals. Commercial terms sit in the Terms rather than in a sales call: payment in advance unless an Order says otherwise, 30 days' prior written notice of any price change taking effect only at the next renewal, a 60 day billing dispute window, interest on overdue amounts at the lower of 1.5% monthly or the legal maximum, suspension on 10 days' notice for non-payment, and termination for convenience on 60 days' notice. What holds it off A is that three of the four legal tiers, including both products an attorney or agency would actually buy, carry no figure and route to a demo. Verified 12 September 2026.
Firm and Practice Coverage
Who the product is actually built for. AmLaw, midlaw, small firm and solo, in house departments, government and courts, and which practice areas are supported rather than merely claimed.
Segments are named with substance, short of limits. The site addresses criminal prosecution, criminal defense, law enforcement, civil law firms and court reporting agencies, and lists transcription in 37 or more languages on the Pro plan. What case types, file formats or jurisdictions it does not handle is not stated. Verified 22 September 2026.
Who the product serves is documented precisely and the limits are stated, which is the A band. Dedicated pages address law firms, courtrooms, law enforcement and government alongside corporate, education and media, and within the legal line the buyer types are named separately and served by different products: attorneys and litigation teams by Legal Visor, court reporting agencies and digital reporters by Legal Capture, courts and agencies by the courtroom material, and corporate legal departments in their own material. The proceeding types supported are enumerated rather than gestured at: depositions, trials, bench hearings, jury trials, arbitrations, witness interviews, dictation, and digital evidence including body-camera footage and 911 calls. The limits are the part that earns the grade, because they are published at the point of sale rather than buried: the entry plan is marked not court admissible, capped at one user, English transcription only, with basic export and integrations, and the higher tiers state where Spanish is added, where certified transcripts are an upgrade, and where court-admissible output is by request. R15 applies to the practice-area limb, which does not bite on a product whose unit of work is a proceeding rather than a practice area. Verified 12 September 2026.
The 12 legal signals, side by side
Recorded rather than graded. These are the questions a practitioner has to answer before a tool touches a client matter, and the answers are taken from public material only.
Client Data in Training
Can material a lawyer puts into this product be used to train a model?
The Terms of Service say customer content is analyzed by Rev's speech recognition and other AI models and may be used for continuous training of those models, which Rev says are proprietary, kept in-house and not shared. Customer content will not be used to train generative AI models, and the security page says third-party models are never trained on it. No opt-out was located.
For the legal buyer the position is opt-in, and it is set out in the agreement rather than a policy page. Terms section 4(e) provides that files and output in connection with deposition transcription, Legal Visor, or other services provided directly to law firm customers shall not be used for general ASR training unless an authorized representative gives written opt-in. Around that sit three further terms a reader should weigh together.
Section 4(c) prohibits the vendor, and any of its providers, from using customer files or output, identifiable or de-identified, to train, fine-tune or incorporate into the weights of any generative AI model or large language model whose output is exposed to a third party, and permits only third-party model providers that are contractually barred from training on customer data. Section 4(b) permits general ASR training for customers outside the law firm carve-out on an opt-out basis, after segmenting audio into clips of a few seconds insufficient to convey substantive content and disassociating them from the account, with no re-identification permitted; opt-out is prospective and does not require retraining.
Section 4(a) reserves customer-specific optimization using files in identifiable form to tune recognition for that customer, which is not covered by any opt-out and is the term a buyer should read most carefully.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
The Terms of Service let Rev set general practices and limits, including the maximum period it keeps customer content, without stating that period. The Privacy Policy keeps personal data as long as necessary and keeps backups longer. No retention period or customer deletion setting is published.
Retention is under the customer's control, with the mechanism stated in the agreement and the period tied to termination rather than to a default in-service schedule. DPA section 8 requires the vendor, within 30 days of termination and at the customer's choice indicated through the Services or in written notice, to delete or return all personal data processed on the customer's behalf and to delete existing copies, subject to legal retention requirements, with one copy retained solely for the establishment, exercise or defense of legal claims.
Terms section 5(c) adds that privileged materials are returned or destroyed on termination or on request, and section 9(c) gives 30 days of post-termination export assistance. The Trust Page states data can be removed at any time by contacting the account manager. What is not published is a default retention period for material held during the subscription, and Terms section 1(b) reserves the right to retain customer data after account deletion as reasonably necessary to comply with law.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
Access is set in Rev's own permission model, maintained by the customer: sharing permissions set per file or per workspace, invitation or link access only, and domain claiming to control which users belong to an account. Nothing describes inheriting permissions from a case or document management system.
A separation model is documented, at the customer and material level rather than the matter level. Terms section 5(c) requires access to privileged materials to be limited to personnel with a direct need to know for providing the Services, and bars disclosure to third parties except as needed to deliver them under equivalent confidentiality. DPA section 2(f) prohibits combining personal information received from one customer with personal information from any other person, except for the de-identified segments used in general ASR training.
The vendor's data policy states that models cannot access or learn from other customers' data, and the Trust Page documents customer-assignable roles so no user has more access than their part of the work requires, alongside internal least-privilege access and regular reviews. What is not addressed is segregation between matters inside a single customer's account: nothing published describes a wall between one firm's cases, which is the question a firm with adverse-party depositions on the same platform would ask.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
The Terms of Service say that if either party receives a subpoena, document request, court order or government demand for the other's confidential information, it will give prompt written notice so the request can be challenged or limited. No transparency report was located.
Notice is committed and a reporting mechanism exists, both with qualifications the reader should see in the vendor's own words. The Trust Page states that governmental authorities are not permitted access unless required under applicable law and on legitimate grounds, that any disclosure is limited to what is strictly necessary and is reviewed by the vendor's legal and privacy teams, and that the vendor will do its best to notify the customer before disclosure unless notification is prohibited by law.
DPA Part 4 goes further and is contractual: unless legally prohibited the vendor will redirect the authority to the customer and inform it that the vendor is not authorized to disclose, will use commercially reasonable legal mechanisms to challenge the demand, and where urgent access prevents a challenge will notify the customer as soon as possible afterwards. It also commits the vendor, once every 12 months on the customer's written request, to report the types of binding legal demands it has received, including any under section 702 of FISA.
Terms section 5(a) separately requires prompt prior notice of any compelled disclosure of confidential information to allow a protective order to be sought. The report is annual, on request, and to the customer rather than published, and the notice commitment is best efforts.
Primary Law Corpus Provenance
Where does the law in this product come from, and does the vendor have the right to use it?
Checked the home, platform and law firm pages on 22 September 2026. Rev works only on the files a customer uploads and does not answer from a body of law, so no legal corpus is identified.
The product answers from the customer's own recording rather than from a corpus, so the provenance question lands on the training corpus, and there the source is named without a separate licensing basis. Terms section 4(b) identifies the corpus explicitly: data derived from customer files and output, segmented into clips of a few seconds insufficient to convey substantive content, disassociated from the account, and used solely to correlate acoustic speech patterns with text sequences across a large aggregate training corpus.
Law firm materials are excluded unless the customer opts in. No third-party or licensed speech corpus is identified, and no statement covers what else the general engine was trained on before or alongside customer-derived data. The permission basis is the agreement itself rather than a separate license, and customer warranties in section 3(a) place responsibility for lawful acquisition of the files on the customer.
Good Law Verification
Does the product tell you when the authority it just cited has been overruled?
Checked the home, document editor and legal AI templates pages on 22 September 2026. Citations point to timestamps and pages in the customer's own files; nothing addresses checking legal authority for later treatment.
The product cites no legal authority and produces no legal analysis, so nothing addresses checking subsequent history and nothing would be expected to. The currency mechanisms it does publish concern the accuracy of the record rather than the standing of law: a Post-Production Corrections Policy incorporated into the Terms, human review and certification tiers, and jurisdictional formatting for admissible transcripts. Recorded so the row states the position rather than leaving a reader to infer it.
Refusal and Uncertainty Behavior
What does the product do when the answer is not in the corpus?
Checked the home, multi-file analysis and document editor pages on 22 September 2026. Rev says findings come only from uploaded files rather than a model's best guess, but nothing describes what the product does when the files do not answer a question.
The hand-off to a person is documented, in the agreement and in the product structure, short of anything demonstrable. Terms section 4(d) states that AI Analysis Features use probabilistic models that may produce inaccurate, incomplete or misleading results, are not a substitute for qualified human review, and that the customer is solely responsible for verifying output before relying on it. The product expresses the same boundary structurally: AI-only output is marked not court admissible, human review is an upgrade on every plan, and certified transcripts are the tier for evidentiary use.
Section 7(f) adds a genuine refusal behavior at the input end, since the vendor may reject files that do not meet reasonable audio quality levels rather than transcribe them badly. What is not published is anything measurable: no confidence scoring exposed to the user, no uncertainty marking within a transcript, no rate at which files are rejected or escalated, and no evaluation demonstrating the behavior.
Fabricated Citation Record
Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?
Searched the AI Hallucination Cases database maintained by Damien Charlotin on 22 September 2026 for Rev.com and for Rev AI; no recorded case naming the product was returned. No court order, opinion or disciplinary record naming the product was located. This is a statement about the public record rather than a finding about the product.
Searched on 12 September 2026, on both the product name and the company name, against published trackers of AI hallucination decisions including coverage of the Charlotin AI Hallucination Cases database, for court records addressing fabricated or hallucinated legal citations in this product's output. None located. This is a statement about the public record on that one subject as of that date, and this signal does not record other litigation.
Bar Guidance Alignment
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
Checked the home, law firm, security and About pages, the Terms of Service and the Privacy Policy on 22 September 2026. No material engages with lawyers' professional or ethical obligations or names any ethics opinion.
Professional obligations are referenced generically and no guidance is named. Terms section 4(e) makes the customer solely responsible for determining whether use of AI Analysis Features on legal materials is consistent with its privilege, confidentiality and professional conduct obligations, and section 5(c) repeats the point for privileged materials, adding that submission must be assessed against applicable rules of professional conduct and privilege law.
That is a real acknowledgment that professional conduct rules govern the customer's use, and it is more than most records in this corpus carry, but it names no authority. No bar opinion, no ABA Formal Opinion 512, no state guidance and no court rule on AI-assisted transcription is cited anywhere on the estate, and nothing maps the product's use to any jurisdiction's requirements. The nearest thing to jurisdictional engagement is transcript formatting and certification described as meeting jurisdictional standards, which is court practice rather than professional responsibility guidance.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
The site claims large time savings, such as up to 30 hours of case preparation saved and drafting time cut by 97% in one case study. Nothing addresses how AI-assisted work is reflected in what a client is billed.
Nothing located addresses what happens to the bill, and unlike most records that take this value the product sits squarely inside a fee relationship rather than outside one. Deposition and hearing transcripts bought by a law firm are conventionally charged on to the client as a disbursement, and the vendor's own pricing is built on units a firm could pass through: a monthly plan with an hours allowance, per-proceeding capture, and certified transcripts as a priced upgrade.
Despite that, no published material addresses disclosure of the charge to the client, no guidance covers how a firm should treat the cost of AI-generated versus human-certified output, and nothing addresses the compression question this signal exists for, being what happens to the bill when a transcript that took a reporter days is produced in minutes. The absence is recorded as a gap rather than as an absence of relevance.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
Checked the security page, the Terms of Service and the Privacy Policy on 22 September 2026. A SOC 3 report is said to be available for review; no subprocessor or model provider list and no client-facing disclosure material was located.
The pack a firm would forward is substantially published, but the one artifact a client AI clause usually demands by name is not. Published and forwardable: a full DPA with the EU Standard Contractual Clauses under Modules Two and Three, the UK Addendum and Swiss adjustments; a Trust Page that the DPA designates as the Security Documentation and Annex II; a contractual prohibition on generative AI training and on any third-party model provider that trains on customer data; an opt-in requirement before law firm materials enter ASR training; a named data protection officer; and a contractual audit right.
What is missing is the list itself. DPA section 5(b) commits the vendor to make its then-current sub-processor list available on request, to notify before a new sub-processor begins processing, and to a five business day objection window, and third-party AI model providers must be listed there, so the identities exist and are obtainable, behind an email. A firm asked to name who processes its deposition audio cannot answer from the public record.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
Every finding links to the exact timestamp and source file, with page and line citations for depositions, and human-verified legal transcripts are available. No record of which model produced an analysis or draft, or export of AI use for disclosure to a court, is described.
Substantial elements of a record exist and they are built for transcript authentication rather than for disclosing AI use. Published: time-coded and speaker-identified transcripts tied to the source recording; support for chain-of-custody requirements on Legal Capture; certified transcripts with cover pages and jurisdictional formatting for evidentiary use; the distinction between rough drafts, AI output and certified final transcripts; and a Post-Production Corrections Policy incorporated into the Terms.
Together those let a party show where a transcript came from and who completed it. What is absent is the AI-specific disclosure a certifying lawyer would need: nothing states that the record identifies which engine or model produced a passage, which portions were machine-generated against human-corrected, or who reviewed them, and no disclosure template or court-facing guidance was located. The most useful published fact for this purpose is negative and sits on the plan page, being that AI-only output is marked not court admissible.
The questions both sides leave open
Derived from the records above rather than written, so it cannot favor either vendor. Take these into both conversations and ask each side the same question.
- Good Law Verification
Which one fits
Choose Rev if
- You want AI that works across a whole case file, not only the transcript. Rev analyzes up to 500 files at once across audio, video, documents and images, cites every finding to the exact timestamp or page and line, drafts summaries, chronologies and affidavits in an editor, and searches case photos.
- You want to start without a sales call. Rev publishes a free plan, Essentials at $25.49 per seat per month billed annually and Pro at $47.99, with human transcription at $1.99 a minute and 99 percent accuracy guaranteed; only the Unlimited plan is quoted.
- Your practice is criminal defense, prosecution or a smaller civil firm on Clio. Rev addresses prosecution, defense, law enforcement and civil firms, states CJIS and HIPAA compliance, pulls files from Clio, Dropbox, Box, Google Drive and OneDrive, and adds a Clio Manage integration on its Pro plan.
Choose Verbit if
- You are sending privileged depositions and need the privilege protected in the contract. Verbit's terms limit access to privileged materials to staff with a direct need to know, bar any use beyond the service and customer specific optimization, require return or destruction on request, and state that nothing in the agreement waives privilege.
- You do not want your testimony training a shared model. Verbit's terms keep deposition and other law firm material out of its general speech recognition training unless an authorized representative gives written opt in, and bar any customer content from training a generative model, by Verbit or its providers.
- You need to know which output can go before a court. Verbit marks its AI only Standard plan as not court admissible and offers human reviewed and certified transcripts with jurisdiction specific formatting and cover pages for evidentiary use, backed by a published corrections policy.
In summary
Rev
Rev, from Rev.com, Inc. of Austin, Texas, is an investigative intelligence and legal transcription platform for criminal prosecution and defense teams, civil law firms, law enforcement and court reporting agencies. Its own speech recognition models transcribe audio and video, and its AI analyzes up to 500 case files at once with findings cited back to a timestamp or page and line, and drafts summaries, chronologies, affidavits and briefs. Human transcription and deposition summaries from SmartDepo, which Rev acquired, are sold alongside. The AI Legal Index grades it in the top two bands on eight of fifteen capability axes, with an A on AI centrality. Plans are published from free to $59.99 per seat per month. As of 22 September 2026 the index located no retention period, subprocessor list or AI governance position.
Verbit
Verbit, headquartered in New York, is a transcription and captioning platform whose legal line produces the record of depositions, hearings, trials, arbitrations and interviews, using its own speech recognition engine, Captivate, trained on legal terminology, with a generative layer for summaries and inconsistency detection. It sells Legal Capture to court reporting agencies, Legal Visor to attorneys and a self serve Standard plan. The AI Legal Index grades it in the top two bands on fourteen of fifteen capability axes, with A grades on autonomy and oversight, privilege posture and practice coverage. Its terms protect privileged materials by name, keep law firm material out of general training without opt in, and mark AI only output on the entry plan as not court admissible. As of 12 September 2026 the index located no published subprocessor list and no named model provider.
Questions buyers ask
Rev vs Verbit: which is better for deposition transcription?
On published evidence Verbit sits in the top two bands on fourteen of fifteen AI Legal Index capability axes and Rev on eight of fifteen, mostly because Verbit's terms protect privileged material, exclude law firm recordings from general training without opt in, and mark which output is court admissible. Rev goes further with AI over a whole case file and publishes prices for most plans. Firms sending privileged testimony have more to read from Verbit.
Does Rev train AI on my recordings?
Rev's Terms of Service say customer content is analyzed by Rev's speech recognition and other AI models and may be used for continuous training of those models, which Rev describes as proprietary and kept in house. The same terms say customer content will not be used to train generative AI models, and Rev's security page says third party models are never trained on it. No opt out was located. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 25, 2026. No vendor pays for placement.
Are Verbit's AI transcripts court admissible?
Not on the entry plan. Verbit's self serve Standard legal plan, at $24 a month for 100 hours, is marked not court admissible in its own feature list. Court admissible transcripts are available by request on the higher tiers, and certified human completed transcripts with cover pages and jurisdiction specific formatting are sold for evidentiary use. Rev offers human verified legal transcripts alongside AI transcription it states is 96 percent accurate or better. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 25, 2026. No vendor pays for placement.
How much do Rev and Verbit cost?
Rev publishes a free plan, Essentials at $25.49 per seat per month billed annually or $29.99 monthly, Pro at $47.99 or $59.99, and human transcription at $1.99 a minute, with its Unlimited plan quoted. Verbit publishes one legal figure, $24 a month for its Standard plan with 100 hours, and routes Legal Capture, Legal Visor and its Custom plan to a demo. Verbit's terms give 30 days' notice of price changes, effective at renewal. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 25, 2026. No vendor pays for placement.
What do Rev and Verbit both leave unpublished?
The generative models. Both name their own speech recognition engines, and neither names the generative models or providers behind its summaries and analysis. Neither publishes the test behind its transcription accuracy, a measurement of whether accuracy holds across accents and audio quality, or a record showing which passages a machine produced and who edited them. Neither addresses how a firm should pass transcript costs to a client when the transcript came from AI. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 25, 2026. No vendor pays for placement.
Two readings to weigh. Rev's right to train its own speech recognition and AI models on customer content is a term Rev publishes, and it excludes generative model training; a firm should read it before uploading privileged recordings. Verbit's terms also reserve customer specific optimization using a customer's files in identifiable form, which no opt out covers, and its subprocessor list, including any third party model providers, is available only on request. Rev acquired SmartDepo, which the index grades separately. Rev was verified on 22 September 2026 and Verbit on 12 September 2026. Neither vendor reviewed this page.
Neither vendor paid for inclusion, placement or a grade, and neither reviewed this page before it published. Everything above comes from public material on the dates shown. How the index grades.