Primary_care · AI symptom assessment and urgency triage (consumer app + Ada Assess enterprise)
Ada
Ada Health GmbH
Ada, from Berlin's Ada Health, is one of the most widely used AI symptom checkers, and its editorial picture is a study in reading evidence carefully. The enterprise product, Ada Assess, holds a verified CE Class IIa certificate under the EU Medical Device Regulation, issued by TÜV SÜD in December 2022 — a real, notified-body-assessed medical-device authorisation. But the consumer app's regulatory class is described inconsistently across Ada's own pages, which variously call it Class I under the old directive and Class IIa under the regulation, and there is no FDA authorisation: in the United States Ada is positioned as non-diagnostic. The clinical evidence is mixed and must be read honestly. Ada's own vignette study put its top-three condition suggestions below GPs, and the broader independent literature finds symptom checkers as a class are modest — primary-diagnosis accuracy often well under half — while newer, company-partnered real-world studies in strong venues report meaningful shifts in appropriate care-seeking. A genuine notified-body certificate and a substantial, peer-reviewed evidence base earn one mark; the mixed and largely category-level accuracy data, the absence of any FDA authorisation, the consumer-app class ambiguity, and the company-reported nature of its scale figures hold it to one.
Performance Metrics
Clinical Evidence
Ada's evidence base has three distinct layers, and conflating them would misrepresent the product. The first is Ada's own vignette study (Gilbert et al., BMJ Open 2020; PMID 33328258), a clinical-vignettes comparison of digital symptom-assessment apps against GPs across 200 primary-care cases. Ada's top-three condition-suggestion accuracy was reported at 70.5% against GPs at roughly 82%, with high condition coverage and safe-urgency advice broadly comparable to clinicians. It is a manufacturer-authored, simulated-case study — useful, but not real-world and not independent. The second layer is the independent symptom-checker literature, which evaluates Ada among many tools rather than in isolation and is distinctly sobering. A systematic review (npj Digital Medicine 2022; PMID 35977992) found primary-diagnosis accuracy across digital symptom checkers ranged from about 19% to 38%. A twelve-checker evaluation (PLoS One 2021; PMID 34265845) found the correct diagnosis somewhere in the top five suggestions a mean of 51% of the time. A five-year follow-up of triage accuracy (JMIR 2022; PMID 35536633) found a median triage accuracy near 56% with no improvement over five years and more than 40% of emergencies missed across the apps studied. These are category-level findings, not Ada-specific verdicts, but they are the honest backdrop against which any single-tool claim should be read. The third and newest layer is company-partnered real-world outcome research in strong venues, which measures behaviour rather than diagnostic accuracy. A 2026 NEJM AI study conducted with CUF in Portugal reported that appropriate care-seeking roughly doubled (about 30% to 64%, physician-reviewed) among users, and a 2026 Nature Health study on South Africa's MomConnect reported that more than half of pregnant women and mothers who received treatment had not planned to seek care before using Ada. Both are meaningful and prospectively designed, and both are company-partnered — a real evidentiary step up from vignettes, to be weighed with their sponsorship in mind. The through-line: Ada's demonstrated value is care navigation and appropriate triage, not diagnosis.
| Study | Design | n | Sensitivity | Specificity | AUC | Published |
|---|---|---|---|---|---|---|
| Gilbert S, et al. (Ada Health; clinical-vignettes comparison to GPs) | ExpertExpert review | 200 | — | — | Ada top-3 condition suggestion 70.5% vs GPs ~82.1%; safe-urgency advice broadly comparable | BMJ Open, 2020; PMID 33328258 (200 primary-care vignettes; manufacturer-authored) |
Clinical Pulse
Verified clinician sentiment
Verified clinician reviews launching soon · Apply to contribute →
Reviews are signed by clinicians verified against their regulator's public register.
Inside the algorithm
Editorial featureHow Ada turns described symptoms into an urgency recommendation.
Five stages — from raw input to verdict — drawn from manufacturer documentation and the public regulatory record.
- INGEST
- NORMALISE
- DETECT
- LOCALISE
- VERDICT
Stage 01 · INGEST
A person describes their symptoms in an adaptive interview.
Ada begins with a chief complaint and then asks an adaptive series of structured questions — each answer shaping the next — rather than taking a free-text description alone. There is no imaging or laboratory input; the raw material is the person's own symptom report.
For the consumer app this interview is self-service, with no clinician in the loop; Ada Assess runs the same style of interview in a partner- or clinician-facing context.
Input
Symptom Q&A
Inference
Per-assessment
Inside the Auris+ Listing
Five more sections complete this device’s Auris+ Listing.
Decision Ledger
ProPro unlocks a private, cross-vendor log of every case you read on this device — what the AI called, what you concluded, and a one-line reason if you overrode it. Ready for the EU AI Act's deployer logging obligations when they land in 2028.
Clinical Evidence Deep Dive
ProPro unlocks the structured clinical-evidence summary — study count, target patient population, and a tabular accuracy-metrics view drawn from peer-reviewed sources.
Peer-Reviewed Publications
ProPro unlocks the curated peer-reviewed publication list with PubMed cross-links — the citation backbone of every editorial verdict.
Post-Market & Regulatory Conditions
ProPro unlocks the post-market surveillance summary, recall record, and the conditions of approval that bound real-world use.
AI Algorithm Version History
ProPro unlocks the chronological record of algorithm version changes — what changed when, drawn from manufacturer changelogs and regulatory filings.
Regulatory Approvals
Safety Record
No Ada Health recall, FDA warning or MHRA field safety notice was identified in public sources as of July 2026; because Ada holds no FDA-regulated diagnostic device in the US there is no corresponding MAUDE record, and the FDA and MHRA databases could not be queried directly from this environment, so this is "none found in public reporting" rather than an exhaustive audit. Ada is assessment and triage support, not a diagnosis, which bounds its harm surface, but the category carries a real and documented safety consideration: independent evaluations of symptom checkers as a group have found substantial fractions of emergency scenarios under-triaged, with one five-year study reporting more than 40% of emergencies missed across the apps tested. Ada's own vignette study reported safe-urgency advice broadly comparable to GPs, but the category evidence means a reassuring or non-urgent Ada result must never substitute for seeking care when a person is worried, and the outputs remain decision support to be confirmed by a clinician.
Intended Use & Indications
Ada is a symptom-assessment platform intended to help people and care partners navigate to the right level of care. The user describes a chief complaint and answers an adaptive, structured question set; Ada's probabilistic reasoning engine, working from a physician-curated medical knowledge base, returns a list of possible conditions and an urgency recommendation across a graded scale (broadly, self-care to emergency). The consumer app is self-service with no clinician interpreting the result; Ada Assess, the enterprise product, produces partner- and clinician-facing assessment reports and is the version that holds formal EU medical-device certification. Ada is not a diagnostic device: it suggests conditions and urgency rather than diagnosing, and in the United States the company positions it as non-diagnostic and does not hold FDA authorisation. Because it collects symptom input rather than imaging or laboratory data, it has no target anatomy; its outputs are decision support to be confirmed by a clinician or by seeking care.