Nabla AI: Clinical Workflow, Pricing, and Evaluation Checklist

Nabla AI turns clinical conversations into draft notes that a clinician reviews before using them in the medical record. Its current catalog also includes dictation and coding assistance. For buyers, the useful question is not simply whether Nabla generates a readable note: it is whether the chosen configuration preserves the right information through editing, regeneration and EHR delivery.

The individual subscription is $119 USD per month after a 30-day trial, according to Nabla’s current help-center entry. Group pricing is available by inquiry.

Our assessment: Nabla merits evaluation when configurable note structure and a defined EHR delivery route address a real documentation bottleneck. Make the acceptance decision on the reviewed, saved record—not the first draft, a vendor-wide accuracy claim or the subscription price alone.

Sources checked October 22, 2026. Published by Vero, a competing clinical-documentation software provider.

Nabla pricing and access: what is established

The subscription article, updated October 4, 2026, establishes the following access routes. It does not publish a complete feature-by-plan entitlement grid.

On smaller screens, swipe each table sideways or focus it and use the arrow keys to view every column.

Cancellation is documented under Settings → Account → Manage in the subscription field, before the next payment. Refunds, annual discounts, a permanent free allowance and post-cancellation access are not established by that subscription entry. Record them from the offer that applies to your organization.

For desktop use, Nabla specifies Chrome or Edge; it also offers iOS and Android apps. The extension’s documented telehealth route uses a browser tab, not a separate desktop Zoom, Teams or WebEx application. Mobile calls can interrupt microphone access.

Nabla Echo is a different video-audio route: it runs locally and needs the browser’s Local Network Access permission. A blocked permission can prevent capture. Confirm the supported device and conferencing configuration with the organization’s administrator instead of assuming that “telehealth supported” covers every audio path.

Does Nabla fit my practice?

Start with the encounter language, required output and work you want to perform. The capabilities below follow Nabla’s published documentation, checked October 22, 2026.

Encounter language is not note-output language

Nabla’s supported-language list, updated May 7, 2026, includes English and UK English; Spanish and Spanish (Spain); French; Arabic variants for Egypt, Lebanon, Morocco and Saudi Arabia; Armenian, Bengali, Cantonese, Croatian, Filipino, German, Greek, Gujarati, Haitian Creole, Hebrew, Hindi, Italian, Japanese, Khmer, Korean, Mandarin, Persian, Polish, Portuguese, Punjabi, Russian, Serbian, Tamil, Telugu, Thai, Urdu and Vietnamese.

The settings guide, updated June 29, 2026, separates the main and secondary encounter languages from the interface/note language. It lists English, English (UK) and French for note output. A Spanish-language visit therefore does not establish Spanish-language note output. Translated patient instructions are a separate setting, not the same output as the clinical note. Have a qualified reviewer check the source, note and translated instructions for the chosen language pair, and arrange an interpreter when needed.

Four workflows to distinguish

• Ambient documentation: captures a clinician–patient conversation and produces a structured draft for review. This fits encounter summarization rather than verbatim entry into a chosen field.
• A dictated note within Nabla: select Start dictated note beside the new-encounter control, speak the note, then edit it in Nabla. The help guide describes automatic or explicitly spoken punctuation and pause/resume controls. This does not itself insert text into an EHR field.
• Nabla Dictation at the cursor: the separate dictation workflow enters speech into the selected destination, such as an EHR note or message. Nabla documents voice commands, reusable phrases and anchored dictation while consulting other chart content. Verify the desktop setup, destination and supported commands before relying on them.
• Coding assistance: Nabla describes ICD-10 suggestions and E/M suggestions based on medical decision-making, with supporting rationale and clinician confirmation. This is a review workflow for proposed codes—not evidence that a claim has been submitted or will be paid.

Plan scope: the $119 subscription entry leaves module entitlements unspecified. Obtain the included dictation, coding and integration features, language/output options, fees and setup requirements in writing.

Clinical workflow and note-editing controls

A useful trial follows the work the clinician actually performs: open the correct encounter, establish permission to capture, select the intended template, capture the conversation, review the draft, make corrections, then verify the destination. Agree on a non-recording alternative for patients who decline and a fallback for technical failure.

Nabla’s help-center documentation describes the following editing controls.

For a clinician, these differences matter more than a long template count. A full rewrite may be convenient early in drafting; a section-level change may be better once the rest of the note has been carefully edited. Neither choice removes the need to compare against the source.

Do not use a template to fill missing examination findings. If the encounter contains a patient report but no observed measurement, the resulting note should retain that distinction. A reusable preference such as concise prose is different from an encounter fact such as an examination result. The clinical notes software guide covers broader signing and version-control requirements.

EHR delivery: choose the route before the trial

Separate manual transfer, a custom Nabla Connect implementation and the named EHR integrations below. Their setup and receiving fields are not interchangeable.

Copy and paste: a clinician reviews the note, copies selected sections or the entire document, and pastes it into the appropriate EHR field. The clinic owns patient selection, field selection and the saved-copy check. Count that effort when measuring time; a quick draft can still create a slow handoff.

Nabla Connect: the documented flow launches an encounter from the EHR, passes patient context into Nabla, and returns content when the provider chooses Export Note. It requires EHR-side implementation, backend authentication and callback handling. Arrange site activation with the EHR implementation team.

The callback documentation gives technical evaluators something concrete to inspect. It describes signed callbacks, replay protection, duplicate handling through a request identifier, and a response that controls the success message shown to the clinician. Note sections use structured content; assessment-and-plan subsections can remain distinct instead of being flattened. Transcript export is optional and must be enabled.

Documented Epic and athenahealth workflows

Nabla’s integration pages describe these workflows as of October 22, 2026. The final column lists the site-specific setup and purchasing decisions.

Available description versus roadmap: the Epic page presents schedule synchronization and chart suggestions as current capabilities, but describes Smart progress notes as forthcoming. It also identifies orders and nursing documentation as upcoming Epic capabilities. Require a dated availability confirmation before including roadmap items in the purchase scope.

Ask the implementer to demonstrate a failed write, not just a successful one. If a callback reaches an integration service but the EHR save fails, which status does the clinician see? What evidence ties a retry to the same patient, encounter and intended note version? A transport acknowledgement should not be mistaken for a durable saved record.

The acceptance evidence should include the source encounter identifier, exported version, receiving note identifier, reopened text and any remaining tasks. Keep this evidence in an approved evaluation workspace. The EHR integration guide explains why narrative delivery, structured-list updates and task creation need separate checks.

What the published Nabla trial shows

A pragmatic randomized trial at UCLA assigned 238 outpatient physicians across 14 specialties to Nabla, DAX Copilot or usual care. The intervention ran November 4, 2024–January 3, 2025; the paper was published online November 26, 2025.

Nabla was used in 29.5% of 23,653 visits in its study arm. Its change in time-in-note relative to control was −9.5%, with a 95% confidence interval of −17.2% to −1.8%. This is a relative change in an EHR documentation measure, not 9.5 minutes saved per visit or a note-accuracy percentage.

The study reported occasional clinically significant inaccuracies. Some secondary wellbeing estimates were uncertain, and the authors called for larger multicenter confirmation. The design, selected setting, incomplete use across visits and earlier product period limit extrapolation. It does not validate every current Nabla feature or establish better patient outcomes. Lukac et al., NEJM AI

Use the published efficiency findings alongside a measured local evaluation of error handling. Before a pilot, define which errors require stopping the workflow, who adjudicates ambiguous findings and what happens to unsuccessful attempts.

NB-01: review a note without losing the correction

Fictional teaching case. Source, draft and correction are manually authored.

Fixed source packet

• Encounter: NB-01, October 22, 2026 at 10:00, adult outpatient review. All identifiers are synthetic.
• Historical chart context: a September 18 entry lists “Medication A, one tablet daily.” The current dose and strength have not been reconciled.
• Current patient statement: “I stopped taking it about two weeks ago. I cannot remember the tablet strength.” The patient reports an outside blood test on October 20 and says they have not received the result.
• Clinician-supplied instruction: record the medication discrepancy as unresolved. Do not document a new medication decision. Ask the records coordinator to obtain the outside report; the clinician will review it when available.
• Post-capture addition: at 10:18, the clinician manually records that coordinator Lee accepted the request task and will check its status at 15:00 today, escalating an unavailable report to the covering clinician. The request has not yet been sent.
• Boundaries: no examination, current weight, laboratory value or completed clinical review is supplied.

The distinction between the source packet and the note matters. A historical list supplies context; the patient’s current statement supplies a different, attributed account. Neither authorizes the software to decide which medication instruction is clinically correct.

Deliberately flawed draft

Medication A continued daily. Blood test reviewed. Records request sent; follow-up complete.

The wording makes four unsupported state changes: historical use becomes current use, a reported test becomes a reviewed result, an accepted task becomes a sent request, and unfinished work becomes completed follow-up.

Show the corrected note and revision checks

Corrected note: The September 18 chart entry lists Medication A, one tablet daily. The patient reports stopping it approximately two weeks ago and cannot recall the tablet strength. Current use, dose and strength remain unreconciled; no new medication decision is documented. The patient reports an outside blood test on October 20; the report is not yet available for clinician review.

Clinician addition, 10:18: Coordinator Lee accepted the task to obtain the report. Request not yet sent. Lee will check status at 15:00 today and escalate an unavailable report to the covering clinician. Clinical review remains pending.

Review result: The corrected text preserves the history/current-account conflict, the missing report, and the named owner and timing. The 15:00 check belongs to this fictional case; use the clinician’s specified timing in a real workflow.

Apply Nabla’s documented controls to the case

Start by preserving the accepted text as the comparison version in the approved test workspace. Then use the documented Regenerate section control on the history section. The proposed acceptance condition is that the 10:18 addition in the plan remains unchanged, while the revised history still attributes the patient’s statement and retains the unresolved discrepancy.

Next, test full-note regeneration separately. The original transcript does not contain the post-capture addition. Compare the entire new draft against the accepted version, use highlighted edits to aid review, and use Undo if the replacement is unsuitable. Compare each manual edit and record the observed behavior of the tested configuration.

Finally, export the accepted version and reopen the EHR note. If an earlier version was already copied, it remains a separate object to reconcile. A correct Nabla editor and an outdated EHR note can coexist. Keep any report-request task open until the actual work and required review are complete.

A reproducible Nabla evaluation checklist

Proposed test pack: eight cases, two attempts each, sixteen attempts in total. Status: not run. Use authorized synthetic inputs and one named configuration before introducing patient information. A clinician should approve the source facts and judge clinical meaning; a health-IT evaluator should verify the receiving system.

Record the test date, app/build or vendor release identifier, operating system, browser, microphone, audio route, language, template, custom instructions, retention settings, user role and EHR version. If the product does not expose a model version, record that as unavailable instead of guessing.

Read the same source script for both attempts and keep the recording only where the approved test configuration permits it. Record interruptions and failed starts as attempts. Preserve the raw draft before editing, the accepted correction, active review time, total elapsed time, saved destination and the reason for a pass, fail or not-run finding.

Classify defects separately: unsupported claim, omitted fact, attribution error, changed uncertainty, lost manual edit, wrong destination or false completion. For each, record whether the error is present in the first draft, survives review, or appears during transfer. A single “accuracy” score hides these different failures.

Report critical failures per all attempted cases, completion rate per attempt, and correction effort for completed cases alongside the count of failures. Where the task is applicable, include reviewer disagreement and its resolution. Report tested languages and audio routes separately. The sixteen-attempt pack is an acceptance rehearsal.

Set the decision rule before running the pack. A wrong-patient write or an undetected invented clinical action should stop rollout pending correction and retest. A formatting preference can be handled differently. Keep remediation owner, deadline and recheck evidence in the decision record.

Privacy: follow the audio, note and feedback separately

The useful privacy question is where each copy goes under the actual configuration. “No default audio storage” is not the same as “no local audio,” “no feedback retention” or “no cloud processing.”

Resolve the difference between the older Storage Policy’s statement that user data is not stored on Nabla’s servers and the Trust Center’s description of hosted clinical notes and U.S./EU cloud regions. Obtain a written, configuration-specific data map covering local copies, cloud processing and retention.

For U.S. HIPAA-covered workflows, the published terms include a BAA. Verify that the correct entity and service are covered by the executed agreement. HHS also requires an appropriate cloud-service agreement and the organization’s own risk analysis. HHS cloud guidance

Nabla’s Trust Center lists security reports available by request and describes U.S. hosting for U.S. clients and EU hosting for other clients. Review each report’s scope, date and exceptions. Outside the United States, apply the relevant local privacy, professional and recording requirements.

Also decide what support staff may receive. A screenshot of a failed draft, exported transcript or diagnostic log can create another data copy. Use synthetic material for the proposed tests and an approved support route for any later clinical incident. The PHI guide explains why identifiable context can remain sensitive even without a patient’s name.

Cost the complete workflow, not only the license

Use the same visit volume and work scope for the manual and assisted routes. Include reviewing, correcting, transferring, recovering failures and handling extra administrative work—not just the time until the first draft appears.

Hypothetical monthly example: one clinician completes 300 eligible notes. Manual documentation takes 3 minutes per note. Clinician time is valued at $120/hour. Assisted review, correction and transfer together take 1.5 minutes per note. Administration costs $10/month; $240 of setup effort is spread across six months.

• Manual route: 300 × 3 minutes ÷ 60 × $120 = $1,800 of valued time.
• Assisted route: $119 subscription + $900 clinician time + $10 administration + $40 setup allocation = $1,069/month.
• Difference: $731/month in modeled capacity value, not cash savings automatically realized.
• Sensitivity: if review, correction and transfer take 3 minutes per note, the assisted total becomes $1,969, or $169 more than manual documentation.

Replace the assumptions with your quote and measured end-to-end effort. Add integration fees, tax and hardware to complete the budget. Count appointments where the patient declines capture and failed sessions in the broader staffing plan. Our scribe pricing guide develops the complete-workflow cost approach.

Who should shortlist Nabla?

An individual clinician with a stable note workflow: use the self-service route to establish whether the available templates and review controls fit your documentation. Complete the trial by checking accepted notes in the receiving record under the approved configuration.

A practice or health system seeking connected documentation: evaluate Nabla with the EHR administrator and privacy team. Specify launch context, receiving fields, error handling, support responsibility and ownership of ongoing configuration. Budget implementation separately from seats.

A clinic requiring fully disconnected work or automatic clinical completion: do not assume the reviewed offering meets that requirement. Documented backup after a connection loss is narrower than fully offline launch and generation. A copied order suggestion is narrower than an authorized and transmitted order.

When considering Vero as another drafting route, compare the same source, required output and receiving-record check rather than different demonstrations. See Vero’s scribe workflow for product scope.

For a wider enterprise purchase, the Ambience Healthcare guide provides another workflow and procurement route to investigate. Keep the shortlist small enough to run identical cases.

A defensible decision records the chosen configuration, accepted use cases, excluded actions, unresolved issues, clinical and technical approvers, and recheck triggers. Repeat affected tests after a template change, model release, new audio route or integration change. The goal is a documentation workflow whose limitations are visible and whose final record can be trusted after human review.