EverHealth Scribe

Graded on the platform not algorithm precedent applied to ModMed, Elation Health, Oracle Health, NextGen and OmniMD, and reinforced by a second finding confirmed on 23 July: the intelligence is not EverHealth's. EverCommerce announced a strategic AI partnership with CarePilot to power native embedded AI across DrChrono, and the operating figures published for this product at launch are explicitly attributed to CarePilot data.

That places EverHealth in the position this index graded C for Veradigm, distributing a third party model under its own brand. Credit where due, as with Veradigm: the partner is named publicly rather than obscured. See the separate record for CarePilot. The grade describes the mechanism, not the quality of the output.

The review gate is stated explicitly in the launch material: notes are generated within the EHR ready for provider review and approval, so approval is described as a precondition rather than a follow up. That is the safer of the two patterns this index tracks, the same order RevMaxx describes and a contrast with products that file into the chart pending review. Held at B because no acceptance rate, edit burden figure or confidence threshold is published, and because nothing was located describing what happens to a note the provider does not approve.

No accuracy figure, model card, named models or evaluation methodology located. The one useful disclosure is indirect: the parent has publicly named CarePilot as a strategic AI partner for embedded capabilities across the platform, which is more than most vendors reveal about their supply chain, but it is not stated whether Scribe itself runs on that partnership or on something else. That ambiguity is the specific question to close.

A supplier is named publicly and the link to this product is not confirmed, which is the specific gap on this record. The parent has publicly identified CarePilot as a strategic artificial intelligence partner for embedded capabilities across its platform, which is more than most record system vendors reveal about their supply chain.

Nothing states whether this scribe runs on that partnership or on something else, so a buyer has a name without knowing whether it applies to the product they are purchasing. A named partner that may or may not sit behind your product is not a supply chain disclosure until the link is confirmed, and confirming it is a single question.

The training position is stated in outline and is a genuine disclosure: no secondary use of identifiable patient data unrelated to care, with de identified data available to improve system performance. That sits in the middle of the three formulations this index now sees, above silence and below either a flat commitment or a permission gate, and it turns on a word the material does not define, since de identification has a specific meaning with two recognised methods and the statement gives the outcome claimed rather than the method used.

Nothing else is named: no model, no hosting arrangement, no sub processor list. Ask whether the scribe runs on the named partnership, which de identification method is used, and whether the supplier is bound by the same terms.

More concrete than most launch announcements and still not evidence. Two figures are reported: an average of eight minutes of documentation time saved per visit, and a 32 percent increase in same day claim submissions. Both are granular enough for a practice to test against its own data, and same day submission rate in particular is independently measurable inside a practice management system rather than being a subjective impression. Held at C because both are user reported at launch with no denominator, cohort, time period or methodology, and no study, controlled comparison or third party assessment was located.

The transparency note supplies part of what this axis measures and leaves the rest open.

What is stated: health privacy compliant handling with encryption and role based access, no secondary use of identifiable patient data unrelated to care, and de identified data may be used to improve system performance.

That is a real training position rather than silence, and it sits in the middle of the three formulations this index now sees. It turns on a word the note does not define. De identification has a specific meaning under the privacy rule with two recognised methods, and the note states the outcome claimed rather than the method used. Ask which.

A second statement in the product material sharpens it. Clinician corrections are described as refining the model over time. So the review step is also a training signal: a clinician editing a draft is contributing to model improvement. That is a reasonable design and it should be understood by the people doing it, and it interacts with the de identification question, since a correction is meaningful only in the context of the text it corrects.

What is absent is retention. Nothing states how long audio persists, whether it is retained after the note is produced, or how long transcripts are held. That is the most commonly published commitment in this category and it is missing here while more difficult disclosures are present.

The partner relationship remains the pressing part, and the earlier assessment framed it correctly. Establish whether patient audio reaches the supplier, what that supplier retains, and whether the commitments in this note bind it.

Ask for the retention schedule, the de identification method, and whether the supplier is bound by the same terms.

No product specific business associate agreement posture was located.

An agreement plainly exists, since the platform is a record system holding protected health information for its customers and the ambient module sits inside it. What is absent is any published statement of what it covers, and the module raises a question the underlying record system agreement may predate.

That is the substance here. A practice that signed with this record system vendor before the ambient product launched agreed terms describing a records platform. The module introduces audio capture of clinical encounters and transmission to an external supplier, neither of which the original agreement was written to contemplate. Establish whether adopting the module amends the agreement, whether a separate addendum exists, and whether enabling it constitutes acceptance of new terms.

The supplier chain is the second question and it is now concrete rather than hypothetical, because the supplier is publicly named. Under the rule a business associate must have written agreements with the subcontractors it uses to handle protected health information. So ask the vendor to confirm that the chain reaches the supplier, what that agreement permits it to do with encounter audio, and what happens to a practice's data if the partnership ends. A supplier relationship between two companies can be terminated on terms the practice never negotiated.

The vendor's transparency note addresses oversight, evaluation and data use and does not address the contractual position at all, which is a gap it could close easily given the rest of the document.

Ask for the agreement, whether module adoption amends it, and confirmation of the subcontractor terms.

No named or dated attestation and no trust centre were located for this module. The platform describes health privacy compliant cloud infrastructure with encryption, access control and secure storage, which is a description of design rather than an examination of it.

One route to evidence is available here that is not available for most vendors in this lane, and it is worth naming because a buyer can pursue it without the vendor's cooperation. The parent is a listed company on a United States exchange, and listed companies are required to describe their cybersecurity risk management, strategy and governance in their annual report, including their processes for assessing risks from third party providers and whether any incident has materially affected them. That disclosure is filed, dated and made under liability, which is a different order of accountability from a marketing page.

It will describe the enterprise rather than this module, so it does not substitute for a scoped attestation. It does establish whether a security programme exists, who owns it at board level, and how third party risk is handled, which is directly relevant given this product depends on an external supplier.

The supplier question is the substance of this axis. Ask what security assessment was performed on the artificial intelligence partner before embedding it, whether the parent's own attestations or assessments cover the supplier's processing, and whether the result can be shared. A company embedding a third party into a clinical workflow at this scale will have done that diligence, and its outcome is the artefact a practice actually needs.

Ask also whether any attestation exists scoped to the health business rather than the group.

No clearance claimed and none required for ambient documentation, and this axis reads better than the category norm because the vendor publishes an explicit statement of intended purpose.

Its transparency note states that the product is not intended to diagnose, treat, or recommend clinical care decisions, and describes the scope as outpatient documentation support for licensed clinicians in real time or post encounter workflows. It further states that the system does not autonomously enter documentation into the record and that all output is presented for clinician review and editing before explicit submission.

That combination is the right one for this question. Whether software falls inside the device framework turns on intended purpose as claimed by the manufacturer, not on what the technology could do, and the two things a regulator would look for are a scope statement and evidence that a professional stands between the output and the patient. Both are published. Almost nothing else in this lane does this, and several vendors use marketing language that pulls against their own regulatory position.

One capability sits at the boundary and should be confirmed as assessed. The product recommends diagnosis codes, which is ordinarily administrative rather than clinical, and is compatible with the stated intended use.

The transparency note also states that the product listens with consent, which is the obligation that actually governs ambient documentation in most jurisdictions. Consent requirements vary by state and remain the practice's responsibility, so a buyer should establish what the product does to support obtaining and recording it rather than assuming the reference covers them.

Ask whether the intended use statement is reflected in contract terms.

The earlier assessment found nothing. That is overturned, and this record now carries the strongest disclosure on this axis anywhere in the category.

The vendor publishes a structured AI transparency note for this product. It states that evaluation includes accent based assessment alongside provider feedback, note completeness, volume and usage derived measures; that monitoring is continuous; and that on detection of disparities the response includes adjustments to prompts, pipeline, tooling and model fine tuning. Separately it publishes a limitations section naming audio quality, overlapping speech, accents and jargon as factors affecting output.

No other vendor across this category publishes a named evaluation activity addressing speaker variation together with a stated remediation path. That is worth saying plainly, because this index has recorded the same absence ninety times and this is the exception.

What is committed is a process rather than a result. No findings are published: no populations examined, no accuracy figures by accent, no thresholds defining a disparity, and no account of what has been detected or changed. A commitment to look is not the same as evidence of what was seen, and the useful next step is small: publish what the assessments have found.

The precision recorded in the earlier assessment stands and is worth preserving. The same day claims figure is a speed metric rather than a coding intensity one, so this product does not sit on the revenue optimisation gradient this index tracks elsewhere. Faster submission of the same claim is a different thing from a differently coded claim.

Ask for the evaluation results and the disparity threshold.

Two passes located no accuracy or error figure, no published limitations and no warranty, indemnity or remediation commitment. One disclosure on this record deserves attention on this axis rather than only on stewardship, because it changes what a clinician is doing when they correct a draft. Clinician corrections are described as refining the model over time, so the review step is simultaneously the quality control and a training signal.

That is a reasonable design and it should be understood by the people performing it. A clinician editing a note is contributing to a model that will serve other customers, and a correction is meaningful for training only in the context of the text it corrects, which interacts directly with the undefined de identification claim made elsewhere in the same material. Nothing published tells the clinician any of that at the point of editing. The supply arrangement compounds the position.

Where the capability may be supplied by a partner, the party that would answer for a wrong note is not necessarily the party whose name is on the product, and the material does not establish which company controls model behaviour. Ask whether corrections train a shared model, what a clinician is told, and which entity carries the performance representations.

Native to DrChrono and unavailable outside it, the same shape as the other EHR native scribes here. Notes are generated within the record rather than transferred into it, which is the meaningful part, and the vendor frames the product as fitting inside the workflow clinicians already use rather than adding a second application. No integration with any other EHR is offered or implied.

No hosting region, residency option or model provider was located.

One part of this axis is answered better than almost anywhere in this category and deserves crediting properly. The vendor names its artificial intelligence supplier publicly, in a launch announcement, and carries that supplier's name in the product title of its own mobile application. So a practice can determine who actually processes their encounters without asking, which is the single hardest fact to obtain in this lane and the one most vendors leave entirely opaque. Several products elsewhere in this index are built on suppliers a buyer has no way to identify.

That naming is the beginning of the answer rather than the whole of it. Identifying the supplier tells a buyer who to ask; it does not tell them what that supplier retains, where it processes, or whether the commitments in the vendor's transparency note bind it contractually. A peer in this category names its model provider and states that an agreement is in place with them, which is the fuller version of the same disclosure.

So the questions run one level deeper here than usual. Where does audio go once it leaves the record system platform, is it processed in the same jurisdiction, which foundation model does the supplier use, and what does that provider retain.

The corporate chain adds a wrinkle worth knowing. The mobile application is published by a third differently named legal entity again, so three names attach to one product.

Ask for the processing location, the full subprocessor list including the supplier's own, and the model provider.

No published rate card and no statement on whether the module is included in the DrChrono subscription or licensed separately, which is the first question for an existing customer. Notably the launch material does not take the position athenahealth took with athenaAmbient, where inclusion at no incremental charge was stated up front.

Ambulatory and bounded by the DrChrono base, which serves independent practices, with the product explicitly positioned under a brand promise aimed at that segment. No specialty count, specialty tuning claim, note format list or language coverage was located, so breadth cannot be compared against competitors that enumerate them. Launched March 2026, so the record should be refreshed once coverage claims mature.