Head-to-Head Architecture Comparison
When internists evaluate documentation systems, the bottleneck is rarely the quality of the prose. The harder problem is the distance between a finished draft and a signed, codable chart in athenaOne. Nabla and Merry AI both produce competent ambient notes. They differ in where the text lands, what gets checked before signing, and what the annual invoice looks like.
Nabla's published athenaOne integration uses an 'Export to Athena' button that refreshes the chart and updates HPI, PE, A/P, and social history. That is a real improvement over clipboard workflows, and it deserves credit. Merry AI takes a narrower, field-level approach. A browser extension writes discrete values into mapped exam, procedure, and assessment fields, so E/M content and 20610 procedure content stay in separate sections.
| Feature | Merry AI | Nabla |
|---|---|---|
| Real pricing (list) | Pro Annual $54/mo per clinician | Pro $119/mo per clinician; enterprise quoted individually |
| Annual cost, 10 internists | $6,480 | $14,280 |
| DOM injection speed | Field-level write at signing; under 1 s per mapped field in internal tests | One-click export; chart refresh updates HPI, PE, A/P, social history |
| Group note-splitting | Diarized sessions split into per-patient drafts | Not described in public athenaOne materials |
| CPT G2211 capture | Suggested on longitudinal visits; suppressed on modifier 25 days outside CMS exceptions | Modifier-aware suppression not described publicly |
| Attestation logging | Clinical Logic Bridge block with field-level provenance and timestamps | Clinician review before export; field-level provenance not described publicly |
| Data retention | Audio shredded from RAM at signing | Audio not stored by default per vendor; confirm transcript window in BAA |
Pricing figures reflect published list prices at the time of writing. Nabla enterprise contracts are quoted individually, so confirm your own terms. At list price, Merry AI Pro Annual costs $648 per clinician per year, against $1,428 for Nabla Pro. That is a difference of $780, or 54.6%. For a 10-internist group, it comes to $7,800 a year before any coding effect.
Mapping athenaOne Exam, Procedure, and Assessment Fields
athenaOne does not publish DOM selectors, and its encounter view changes with each release. Any extension that writes into it must treat the layout as a moving target. We say this plainly because fragile injection is a patient-safety problem, not a cosmetic one.
Selector Strategy and Regression Testing
Merry AI maintains a versioned selector map for three targets: the physical exam section, the procedure documentation section, and the assessment and plan section. Selectors resolve against semantic anchors, such as section headings and accessible labels, rather than brittle positional paths. When a widget loads inside an iframe, the extension attaches directly to that frame's document. Each athenaOne release triggers a regression suite. If a target fails to resolve, injection halts and the draft stays in the review pane rather than landing in the wrong section.
What Gets Written, and Where
Quantified musculoskeletal findings go to the exam fields. These include knee flexion and extension in degrees, WOMAC subscale and total scores, effusion, and stability. Dated failed therapies go to the assessment. Examples are an NSAID trial with start and stop dates, a completed physical therapy course, and prior injections. The 20610 note goes only to the procedure section. It covers site, approach, medication, dose, consent, and immediate response. Numeric values require an explicit verbal statement during the encounter. Inferred numbers appear as suggestions and are never written silently.
Modifier 25, 20610, and G2211 Logic
Modifier 25 requires a significant, separately identifiable E/M service beyond the pre-procedure evaluation for 20610. Merry AI will not suggest modifier 25 on a same-day E/M unless the E/M documentation contains independent elements. These are measured ROM degrees, at least one dated failed therapy, and a problem managed beyond the injection decision. If any are absent, the suggestion is withheld and the gap is listed for the clinician. Modifier 59 is never offered on E/M lines, consistent with NCCI guidance. Confirm code definitions against the AMA at https://www.ama-assn.org/practice-management/cpt.
The G2211 Suppression Rule
CMS generally denies G2211 when the base E/M carries modifier 25 on the same date. Since CY 2025, the exceptions are same-day annual wellness visits, vaccine administration, and Medicare Part B preventive services. Merry AI therefore suppresses G2211 on a 20610 day unless one of those exceptions is documented. Recovery comes from the larger volume of longitudinal visits without procedures, which is where G2211 belongs. Nabla's public materials do not describe equivalent modifier-aware suppression, so ask for it in writing during evaluation.
Retention, Attestation, and Next Steps
Merry AI shreds encounter audio from RAM at signing. Only the signed note and an operation log persist. Nabla states that it does not store audio by default, so confirm its transcript and note retention windows in your BAA. Neither approach leaves an audio record for a later dispute, which is why the attestation trail matters.
Every Merry AI note carries a Clinical Logic Bridge attestation block. It lists which fields were AI-suggested, which were edited, and which coding suggestions were accepted or withheld, each with a timestamp. In California, patient-facing text derived from these notes carries the AB 3030 disclaimer and instructions for reaching a human clinician.
If you want to see this on your own templates, Schedule a 15-Minute Workflow Audit. Bring two de-identified 20610 encounters, and we will walk through the chart together, field by field.
