Nextvisit AI

A multi agent system is the product: models that reference prior patient records, track clinical change and build patient timelines, alongside the capture and generation layer. Physician founded with no services or platform business underneath, and the founding psychiatrist is described as still using the product in his own practice.

Conventional draft and review with the gate clearly placed: notes are produced ready to sign, the clinician finalises them, and the mobile companion supports recording, editing and reviewing before pushing updates to the main workspace. Held at B because no acceptance rate, edit burden figure or confidence threshold is published, and because the marketing emphasis on minimising edits sits in tension with wanting a clinician to read carefully.

Architecture is described at a useful level, with multiple models collaborating and agents assigned to reference prior records and track change, but nothing is measured against it. The headline claim of over 98 percent accuracy carries no methodology, reference standard or denominator, and joins the other unverifiable accuracy percentages this index tracks. No model card or named models located.

The training commitment is the plainest this index has recorded in this segment and it deserves stating in the vendor's own terms: it does not train on customer protected health information, full stop. That formulation leaves none of the room this index has had to unpick elsewhere. There is no shared models carve out, no third party qualifier, no use limitation that reaches training only by construction, and no permission gate that shifts the decision to a form nobody reads.

Recordings are stated to be ephemeral by default and a customer can export or delete notes at any time. Coming from a vendor recording psychiatry, therapy and addiction medicine sessions, that is close to the best answer available on the boundary half of this axis. Two things hold it in the middle band. Nothing is enumerated: no model or model family, no foundation model provider, no hosting arrangement and no sub processor list was located. And the commitment is prospective.

The vendor states it built and tuned its model on psychiatry, therapy and addiction medicine encounters from the start, so a corpus of exactly this content already exists and the commitment covers a customer's data going forward while saying nothing about what came before. Ask what that founding corpus was and on what consent basis, and ask what by default implies about the non default.

Genuine third party review presence on G2 with vendor responses, which is a venue the company does not control, plus a claim of thousands of clinical notes processed and documentation time reduced by over 30 percent without denominator or method.

One accolade should be read carefully rather than credited: the company cites being named Best AI Medical Scribe for Behavioral Health in New York of 2025, but no awarding body, criteria or entrant pool was located, and regional superlative awards of that shape are frequently directory placements rather than assessments. No study or controlled evaluation exists.

The earlier assessment called this the most consequential gap in the record and noted that the segment benchmark was an explicit does not train and does not retain commitment. That benchmark is now met on the training half, in the plainest language this index has recorded anywhere in the category.

The vendor states that it does not train on customer protected health information, full stop. That formulation leaves no room for the qualifiers this index has had to unpick elsewhere: no shared models carve out, no use limitation that reaches training only by construction, no permission gate. It also states that recordings are ephemeral by default and that a customer can export or delete their notes at any time, and it points to a trust centre for the detail.

That combination is close to the best available answer on this axis, and it is being made by a vendor recording psychiatric and addiction medicine sessions, where the stakes are highest.

Three things hold it short of the top grade, and they are small. By default implies a non default, so establish whether retention can be enabled and what happens then. No retention period is stated for notes as distinct from recordings, and export and delete on request is a control rather than a schedule. And no de identification practice is described.

One further question follows from the product. The vendor states it built and tuned its model on psychiatry, therapy and addiction medicine encounters from the start. Establish what that corpus was and on what basis it was used, since the commitment covers a customer's data going forward and says nothing about what came before.

Privacy compliance is asserted consistently across the vendor's material, and no business associate agreement posture, template or scope statement was located. The earlier assessment called that a notable omission for a product recording psychiatric sessions and it remains the right characterisation.

The distinction matters more here than almost anywhere in this lane. Compliance is a posture a vendor asserts about its own practices. The agreement is the contract that makes it accountable to the covered entity, defines permitted uses and creates breach obligations. A clinician recording psychotherapy sessions without one has committed a violation whatever the product's engineering.

One feature of the product makes the scope question concrete rather than formal. The vendor ships a note type for medication assisted treatment, which places substance use disorder records in scope. That content is governed by a separate federal confidentiality regime whose consent and redisclosure rules differ from the general privacy rule, and a standard business associate agreement does not by itself satisfy it. Establish whether the vendor executes the additional agreement that regime contemplates, and what its position is on redisclosure.

A second follows from the buyer profile. The vendor serves solo psychiatrists and small practices alongside larger groups. Those buyers have no procurement function, so whether an agreement is offered by default at signup or only on request materially determines whether it exists at all. A peer in this segment includes it with every account.

Ask whether an agreement is executed at signup, whether it covers substance use records specifically, and what the redisclosure terms are.

No named or dated attestation was confirmed, and the grade records what could be verified rather than a judgement about controls.

The position here is better than a bare absence and should be described accurately. The vendor references a trust centre and directs readers to it for detail on retention and data handling. Its contents were not retrieved in this pass, so nothing is graded on them, but a vendor that maintains a trust centre and points to it is in a materially different position from one that publishes nothing.

The specific next step is therefore narrow: retrieve the trust centre, confirm which report is held, its type and period, and whether the scope covers the mobile companion application as well as the platform.

One piece of context makes the answer likely to be favourable and worth checking rather than assumed. A co founder has published evaluation guidance for behavioural health buyers listing six factors, among them security certifications and specifically SOC 2 Type II, and stating that mental health data warrants standards beyond baseline privacy compliance. That is the correct advice for this segment. A vendor whose leadership names Type II as the buying criterion is more likely than not to hold it, and this index does not grade on likelihood.

The sensitivity of the content raises the bar rather than the company's size lowering it. Recorded psychiatric and addiction medicine sessions are among the most sensitive material any vendor in this index handles.

Ask for the report named in the trust centre, its type and period, and the scope.

No clearance claimed and no device pathway attaches to documentation. The flag from the earlier assessment stands, and the second pass adds a regulatory dimension the record did not previously carry.

The risk tracking concern first. The platform tracks risk factors across visits and the vendor states it gets risk language right on the first pass. In behavioural health, risk language means suicide and self harm, which is the highest stakes inference in the specialty. A system that generates risk language from a session is producing the sentence a later reviewer will treat as the clinician's assessment. Establish what the system infers, what it surfaces, what it does when a session contains ambiguous risk content, and whether generated risk language is visually distinguished for confirmation rather than presented as drafted text.

The new dimension is the addiction medicine product. The vendor publishes a note type built for medication assisted treatment, and documentation of substance use disorder treatment falls under a separate federal confidentiality regime with its own consent requirements, stricter than the general privacy rule and with specific restrictions on redisclosure. A peer in this lane engages with that regime explicitly and publishes consent language for it.

Nothing addressing it was located here, and a vendor shipping a purpose built medication assisted treatment note is squarely inside its scope. That is the question to put before any addiction practice adopts the product.

Ask what the system infers about risk, and what the vendor's position is under the substance use confidentiality regime.

No fairness statement, subgroup analysis, accent or dialect performance disclosure or language coverage claim was located, and the earlier assessment's reasoning about why that matters in this specialty stands.

The second pass adds a quality claim that sharpens the gap rather than closing it. The vendor publishes an overall note quality figure above ninety eight per cent, and states plainly that this is internal scoring measured across active clinicians over a stated period. Labelling it as internal is honest and better than the unattributed accuracy figures common in this category.

It is also a single aggregate number, which is the specific thing this axis exists to look behind. An overall score is dominated by ordinary sessions with clear audio and familiar speech patterns, and conceals whatever variation exists underneath. The vendor is well placed to look, because it already has a scoring apparatus running continuously across active clinicians. Breaking that existing measurement out by speaker characteristics would require no new infrastructure.

The specialty argument makes this more pressing than average. Behavioural health assessment already carries documented disparities in how symptoms are recorded across patient populations, particularly in the language used to describe affect, insight and cooperation. A system tuned on psychiatric encounters will have learned the conventions present in its training material, including any patterned differences, and a platform tracking data points across visits propagates them forward rather than leaving them in one note.

Ask what the existing quality scoring shows when broken out by patient and clinician characteristics, and what the evaluation set contained.

Two passes located no published limitations, no evaluation methodology and no warranty, indemnity or remediation commitment. The one published figure, better than 98 percent accuracy, carries no methodology, reference standard or denominator, so it cannot be tested and does not function as a commitment however precise it looks.

The architecture is described at a useful level, with multiple models collaborating and agents assigned to reference prior records and track change, which tells a buyer what is happening without telling them how often it is right, and an agent that references prior records introduces a specific risk this index has recorded elsewhere: content carried forward from an earlier note can appear in a new one with the authority of fresh clinical observation, and nothing published states whether the draft distinguishes what was heard in this session from what was read from a previous one.

The segment raises the consequences. A psychotherapy or addiction medicine progress note supports a clinical formulation, is submitted for reimbursement, and is discoverable in custody, criminal and civil proceedings, and risk content within it carries immediate clinical and legal weight. Ask for an accuracy figure with a defined denominator, for how carried forward content is marked, and for how the system handles disclosed risk content.

Integration is asserted rather than enumerated. The vendor positions the product as making the EHR work for the clinician, and a third party reviewer describes it integrating seamlessly with their EMR, but no named system, integration architecture or write back mechanism was located. The mobile companion pushes updates to the vendor's own workspace rather than to a record system.

No hosting region, residency option, subprocessor list or model provider was located. Delivery is cloud across desktop and web with a named iOS companion application.

The vendor's trust centre is the obvious place these answers would sit, and it is referenced rather than reproduced in the material retrieved. So this axis should be revisited against that document rather than treated as settled.

Two product facts shape what to look for there.

The companion application is built for capture away from the desk: home visits, voice notes between sessions, and multi location practices. Mobile capture in uncontrolled settings raises the questions this index has asked of other field products. What is held on the device before it syncs, is it encrypted at rest there, what happens if the handset is lost with unsynced sessions on it, and does capture work without a connection.

And the stated ephemerality of recordings interacts with residency in a useful way. If audio genuinely does not come to rest, the residency question narrows to the processing window and whatever the note store holds. That is a smaller surface than most peers present, and it makes the unanswered part more specific: where does processing occur, and which model service performs it.

The last question matters here because the vendor states it tuned its own model on specialty encounters, which suggests proprietary work sitting on top of, or instead of, a commercial foundation model. Establish which.

Ask for the region, the subprocessor list, the model provider, and the on device retention model.

Partially published. An entry tier is described concretely at up to 15 SOAP notes a month with basic analytics and a single provider setup, which tells a solo clinician what they can evaluate without paying. No full rate card, tier ladder or per clinician figure was located, so what the product costs above that entry point cannot be determined without contact.

Deep within one specialty and properly enumerated on clinician type: psychiatrists, psychiatric mental health nurse practitioners, psychologists, therapists, licensed clinical social workers and physician assistants, across outpatient psychiatry, therapy clinics, community mental health centres and telehealth. Note types cover progress notes, SOAP and intake assessments.

Its distinguishing depth is treatment specific tagging for Spravato, TMS and lithium monitoring, the last being genuinely specialised since lithium has a narrow therapeutic index and requires serial serum monitoring that a general tool would not track. Held at B rather than A because JotPsych covers the same specialty with more enumerated note structures and DSM-5-TR coding, and no language coverage is published here.