LegalClerk.ai

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 engine of what a firm buys, and a working telephony platform sits underneath them. The paid plan is sold as unlimited AI-handled calls, the AI answers and conducts every call, and no human receptionist is offered on any tier except an optional live-agent escalation at enterprise level; a firm buying this is buying the automation. What keeps it off the top band is that the product beneath is a phone system that functions without any model. The published agreement describes the service as an online-based phone solution and call handling, with number allocation, number porting, per-minute rates, answer supervision and codec requirements, and never mentions artificial intelligence at any point. The vendor also ships a Chrome extension named for parallel dialling and click-to-call, which is manual outbound dialling rather than an AI feature, and its own privacy notice describes the company as an AI-driven dialling solution for businesses globally rather than a legal intake product. Remove the models and a telephony service remains. Checked 4 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.

Accuracy is asserted repeatedly and never measured. The claims are specific and go directly to output quality: the system never misses a required field or form detail, follows the intake script every single time with zero room for human error, produces no forgotten fields and no misheard details, and eliminates intake errors. Not one is accompanied by a figure, a test set, an evaluation, a transcription error rate or any statement of what happens when the system mishears a caller. Nothing addresses hallucination in either direction, and no accuracy or benchmark page exists on the estate. The limbs about grounding to primary authority and citation status do not apply to this product class and are named here rather than counted against it: the system asks scripted questions and records answers, it does not retrieve or cite law. The floor band does not fit, because its words require that nothing be published on accuracy, and accuracy is the central marketing claim. It matters because the same pages describe the system detecting accident types, timestamping incidents against limitation periods and classifying injury severity, all from a caller speaking on a phone line.

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.

The system runs an entire client conversation unattended, and the controls around it are real and published. A firm sets when the AI picks up, writes the script it follows, defines the criteria for screening, and configures which calls are warm-transferred or escalated as urgent; serious matters are flagged for priority attorney callback, and the enterprise tier adds optional escalation to a live agent. The review surface is substantial: every call is recorded and transcribed, a written summary and structured lead profile are delivered, and call analytics and lead reports are available. What holds this below the top band is that no lawyer is in the loop at the time the system speaks to a prospective client, and nothing is published about what happens when it is wrong: no error handling, no correction path, no abstention behaviour, and no statement of what the system does with a caller it cannot understand. The disclosure decision belongs here rather than anywhere else, and it is the vendor's own words: the FAQ states that most callers cannot tell it is AI, and that this is by design. Concealing the automation boundary from the person on the line is an oversight choice, published plainly, and it is recorded as one.

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.

No production evidence was located, and the social proof on the home page does not survive inspection. The page carries the line that the product is loved by lawyers all across the country above three user avatars; those images resolve to i.pravatar.cc, a placeholder avatar generator, requested by index number. They are not customers. Beyond that the estate claims the product is trusted by top-tier law firms and used by leading legal teams, with no firm named anywhere, no case study, no named individual and no dated deployment. The numbers published are industry statistics and cost comparisons rather than outcomes: that 39 per cent of calls to law firms go unanswered, that a traditional receptionist costs 48,000 to 65,000 dollars a year against roughly 4,800 for the service, and a resulting 90 per cent cost reduction. None of that is evidence of what happened at a firm that deployed it. The band above does not fit, because its words describe customer logos and unattributed testimonials standing in for evidence, and neither a logo wall nor an attributed testimonial was located. Searched the home page, the practice-area pages, the pricing page, the New York intake page and the blog index on 4 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 asserted in general terms in the marketing, and the agreement points the other way on the one clause that reaches call content. The product records and transcribes conversations with prospective clients, which is material a firm would treat as confidential from the first call. The marketing offers HIPAA-grade encryption, secure storage, detailed access logging, secure storage plus redaction, and intake staff trained in client confidentiality. The agreement does not carry any of that across. Its confidentiality section is a mutual non-disclosure clause covering proprietary information the parties disclose to each other, which is business information rather than caller material, and Customer Data, expressly defined to include call recordings, voicemails and call metadata, is instead the subject of section 4.4: the customer grants a licence to use, copy, transmit, distribute and store that data to the extent necessary or desirable to provide and improve the services. No privilege or work product treatment appears anywhere, nothing describes tenancy or separation between firms, and there is no published data processing agreement, although the terms state one forms an inseparable part of the contract. Nothing addresses the prospective-client relationship the product exists to create.

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.

Nothing published addresses the advice line, on a product that speaks directly to members of the public about their legal problems. There is no statement that the company is not a law firm, no disclaimer that it does not give legal advice, no description of what the AI may and may not say to a caller, no jurisdiction limit, and nothing on the firm's supervision or competence obligations when an automated system conducts first contact. The agreement is silent on the subject across every section, and the pricing page's claim that the product is HIPAA and ABA-compliant is a compliance assertion with no authority named rather than a position on advice. The exposure is specific and comes from the vendor's own descriptions: the system timestamps an incident to flag potential statute of limitations issues automatically, filters callers on whether another party was at fault, classifies injury severity to gauge case strength, and prompts callers about what they have said to an insurer, all while, on the vendor's own account, most callers cannot tell they are speaking to software. The band above does not fit, because it describes a boilerplate disclaimer sitting in the terms, and no such disclaimer exists. Searched the home page, the feature pages, the practice-area pages, the pricing page, the terms and the privacy notice on 4 September 2026.

AI Governance and Bias Disclosure

Published governance over model behaviour: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.

No governance position of any kind was located. There is no responsible AI page, no principles statement, no accountable owner named, no pre-release testing regime, and no management system. Nothing addresses uneven performance across the populations this product actually serves, which is the gap worth naming rather than a generic one: the system is sold for multilingual intake and for immigration practice, it is a speech system taking calls from distressed callers, and speech recognition performance across accents, dialects and languages is the obvious place for uneven output in a voice product. Nothing published tests or discloses it. The nearest statements are security assertions about encryption and access logging, which are a different subject and are graded on the stewardship row rather than counted twice here. The platform navigation and footer were inventoried on 4 September 2026 and the only policy documents published are the terms and conditions and the privacy notice.

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.

Access and encryption are addressed in marketing terms and the privacy notice does not reach the recordings. Published: AES-256 encryption shown as a site badge, encryption in transit and at rest, access restricted to authorised personnel, detailed access logging, regular security audits, and secure storage with redaction listed as a plan feature. What is absent is the set that governs the material this product creates. No retention period is stated for call recordings, transcripts, summaries or lead records; no deletion or return-of-data commitment covers them; no subprocessor list exists; and no breach notification commitment was located anywhere. The privacy notice, last updated 15 September 2025, describes the company as an AI-driven dialling solution and enumerates the personal data it collects as contact, account, usage and technical data, with the substance of the document given over to cookies, social plugins, Google OAuth tokens and a Chrome extension addendum. Call content is not mentioned in it at all, so nothing published governs the recordings of prospective clients that the service exists to make. Its deletion and rights routes direct the reader to support addresses at a different company's domain.

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.

The allocation of loss is published and every part of it runs one way. Section 11 caps cumulative liability at the fees paid in the preceding twelve months, or one hundred euros during a free trial, and excludes consequential, indirect, incidental, reputational and punitive damages including loss of data and lost business opportunity, with any claim required to be served by registered post within one year. Section 9.3 provides the service as is and as available and disclaims all implied warranties including fitness for purpose, warranting only a commercially reasonable level of care. Section 10 runs the indemnity from the customer to the vendor, covering breach, violations of law and claims relating to Customer Data; there is no vendor-side indemnity anywhere. No warranty on output, no service credit, no insurance position and nothing addressing a missed call, a misheard intake field or a caller mishandled by the system was located. This is the middle band because the exposure is addressed and readable before signing, through a standard limitation clause that disclaims precisely the risk the product creates, rather than left unaddressed.

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 into the systems a firm already runs, with what moves stated and no depth behind it. Clio, MyCase, Lawmatics, PracticePanther, Google Calendar and Calendly are named, against a claim of thirty-five or more case management integrations, and dedicated pages exist per integration. What moves is stated at workflow level rather than left to inference: new leads, booked appointments and intake summaries sync automatically, and client data from intake reaches the case management system as a CRM-ready lead profile. That is more than a logo wall, which is what lifts it off the band below. What is missing is everything an implementer would need: no field mapping, no direction of sync, no trigger conditions, no error or conflict handling, no configuration guide and no API documentation. The agreement, at section 4.6, disclaims integrations entirely, stating they are not the vendor's products, that use is solely between the customer and the third party, and that the customer warrants it has obtained all necessary consents to transfer data in both directions. The dedicated integrations index was not opened in this pass and is named here as the limit; the grade rests on the connections and sync behaviour described on the pages that were read.

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.

Cloud delivery is implied throughout and neither limb is stated. Nothing published says whether the platform is single or multi-tenant, no dedicated or isolated option appears on any tier, and no region is named for storage or for processing. The only geographic facts published are contractual rather than architectural, and they point in an unexpected direction for a product sold with HIPAA, PHIPA and PIPEDA badges: the agreement is governed by the laws of Slovakia with exclusive jurisdiction in the Slovak courts, notices are addressed to a Bratislava street address, fees may be denominated in euros, the free-trial liability floor is expressed in euros, and daily call thresholds are measured from midnight to midnight Central European Time. None of that states where customer data sits, and it is recorded rather than credited as residency. Where the speech processing happens is a separate silence, since no model provider is identified. A buyer cannot establish from any published source which country holds the recording of a call with their prospective client.

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.

Four badges, no scope, no date, no report and no auditor. The home page displays compliance marks for HIPAA, PHIPA and PIPEDA alongside an AES-256 encryption badge, each as an image with no accompanying statement of what was assessed, by whom, over what period, or against which controls. The pricing page repeats the claim as HIPAA and ABA-compliant. HIPAA, PHIPA and PIPEDA are statutory regimes a company self-attests to rather than independent attestations, and no Business Associate Agreement is offered or mentioned anywhere despite the HIPAA claim. The site further describes itself as proven, audited and secure and refers to regular security audits, naming no assessor and publishing no summary. No SOC 2 of either type, no ISO certification, no penetration test summary and no trust centre or portal exists on the estate, so there is nothing gated to request and this is an absence rather than a retrieval limit. This is the band its words describe exactly: badges on the site with no scope, no date, and no report available.

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.

The vendor gestures at the technology and identifies nothing underneath it. The estate refers to advanced natural language processing, human-like voice technology, NLP tagging used to classify accident types, and AI trained on real legal scenarios. No model is named, no provider is named, no version is given, no hosting location is stated, and no commitment to notify customers of a change was located. Speech recognition, language model and text-to-speech are three separate suppliers in a product of this kind and none of the three is identified. One third party is named in the agreement and is not counted here: customers must agree to the Twilio Voice Services Policy and Acceptable Use Policy, which identifies the telephony carrier layer. Naming who carries the call says nothing about whose model conducts the conversation, and infrastructure is graded on its own axis rather than credited twice. This is the middle band rather than the floor because the architecture is described in general terms; a buyer simply cannot learn whose system is listening.

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.

A real figure with its unit, published without a sales conversation, against an agreement that contradicts the terms attached to it. The pricing page states 400 dollars per seat per month with a 0 dollar setup fee for unlimited AI-handled calls, alongside a free trial and an enterprise tier at custom pricing, with the feature split between them published. That is a rate, a unit and an implementation cost, which is why this sits above the band where only the shape is visible. Two things keep it off the top. The enterprise tier is withheld entirely. And the published commercial terms conflict with the agreement a buyer signs: the page says cancel anytime, while section 7.8 states the service is billed in advance on an annual basis and is non-refundable with no exceptions, section 12.1 renews the term automatically unless notice is given thirty days out, and section 7.7 sets the initial free trial period at three days against the seven advertised on the page. Usage fees for outbound and toll-free calls sit outside the flat rate, and the per-minute rates that determine them are not published, obtainable only by emailing the vendor. Unlimited is also qualified in the agreement by a fair use ceiling of 3,000 minutes per user per month.

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.

Practice coverage is documented with real substance and the boundary is never drawn. Seven practice areas carry dedicated material with intake logic specific to each: personal injury, family law, criminal defence, immigration, real estate, business and corporate, and estate planning, and the treatment is operational rather than a list, describing accident-type detection and liability filtering for injury work, calm structured intake for family matters, and multilingual handling for immigration callers. Firm size is addressed from solo practitioners to high-volume and multi-location firms, with multi-location team management reserved to the enterprise tier. Jurisdictional pages exist at state and city level. What is absent is any statement of where the product stops: no practice area is identified as unsuitable, nothing addresses in-house or government legal work, no call type is described as one the system should not take, and nothing states which languages the multilingual intake actually covers.