Independent comparison 2026
AI documentation tools, compared
What distinguishes Nixi AI from Doctolib, Tandem, and international vendors. A sober comparison with sources.
The differences between AI documentation tools live below the surface: in peer-reviewed clinical evidence, data residency, retention, and specialty depth. Nixi AI is the only tool with a peer-reviewed clinical study in a German specialty (EULAR Rheumatology Open 2025). Nixi AI deletes audio recordings after 72 hours, and texts after the retention period the practice sets (1–90 days).
Comparison table
Superscript digits link to public sources (footer). Competitor claims are cited against their own publications.
- Peer-reviewed evidence (German specialty)
- EULAR Rheumatology Open, December 2025 ¹
- German medical ASR
- <5% WER (internal benchmark, real-world data)
- Multi-speaker diarisation
- Built for complex histories
- Multilingual consultation in one session
- Doctor + patient in different languages
- German billing codes (GOÄ / EBM)
- Native code suggestions
- Data residency
- EU and Switzerland, stored in Frankfurt
- Audio retention
- 72 hours, then deleted ³
- Transcript retention
- Practice-defined (1–90 days)
- Model training on patient data
- Anonymised only (DPA § 12)
- On-premise / federated (hospitals)
- On request
- § 203 StGB-compliant contracts
- Explicit, per customer
- Peer-reviewed evidence (German specialty)
- Unclear (no DPA reviewed)
- German medical ASR
- EU-hosted, no specialty-trained ASR
- Multi-speaker diarisation
- Standard
- Multilingual consultation in one session
- One language per session
- German billing codes (GOÄ / EBM)
- AI billing assistant
- Data residency
- EU (per vendor)
- Audio retention
- Not stored
- Transcript retention
- 48 hours ⁴
- Model training on patient data
- Unclear (no DPA reviewed)
- On-premise / federated (hospitals)
- No
- § 203 StGB-compliant contracts
- Standard DPA
- Peer-reviewed evidence (German specialty)
- Capio study (preprint 2025, not peer-reviewed)
- German medical ASR
- English-first, German added later
- Multi-speaker diarisation
- Standard
- Multilingual consultation in one session
- One language per session
- German billing codes (GOÄ / EBM)
- SNOMED / ICD-10, no native EBM
- Data residency
- EU
- Audio retention
- Not stored
- Transcript retention
- Configurable
- Model training on patient data
- De-identified, for product improvement (per DPA)
- On-premise / federated (hospitals)
- No
- § 203 StGB-compliant contracts
- Standard DPA
- Peer-reviewed evidence (German specialty)
- None for Germany
- German medical ASR
- English first
- Multi-speaker diarisation
- Standard
- Multilingual consultation in one session
- One language per session
- German billing codes (GOÄ / EBM)
- Not supported
- Data residency
- Varies (Heidi EU region; Abridge/Suki USA)
- Audio retention
- Variable
- Transcript retention
- 30 days (DAX)
- Model training on patient data
- Variable
- On-premise / federated (hospitals)
- No
- § 203 StGB-compliant contracts
- Often unavailable
Where the vendors substantively diverge
Three differences that matter day-to-day
1. Privacy that crosses other vendors' business-model boundaries
★Zero audio retention, practice-defined transcript retention. § 203 StGB per practice.
Strengths
- Audio is deleted automatically after 72 hours
- Transcript retention set by the practice (1–90 days, then automatic deletion)
- DPA closed explicitly per practice or hospital
- Stored in Frankfurt, no data centres in the USA
Limitations
- Doctolib retains transcripts for 48 hours ⁴ and uses pseudonymised patient data for AI training (opt-out available) ⁵
- Microsoft Dragon Copilot retains audio + transcript for 30 days by default
- Tandem and Nabla don't store audio but keep transcripts in the cloud
- Price
- Basic €0 · Practice Pro €89/month, or €69/month billed annually
- Best for
- Practices with strict data-protection requirements (KV, supervisory authority, hospital network)
2. Peer-reviewed clinical evidence in a German specialty
EULAR Rheumatology Open, December 2025. 88 consultations. 0 serious errors.
Strengths
- Consultation time reduced in 82% of cases (72 of 88)
- 0 serious errors; all reported errors classified as minor
- 100% patient participation rate, no refusals
- 70% of patients rated the AI assistance positive or very positive
Limitations
- Voize has a comparable study, but only for nursing care (Charité, October 2025)
- Doctolib cites an internal study from its own organisation ²
- Tandem relies on the Capio report from Scandinavia, not peer-reviewed
- Price
- Study open access. DOI 10.1016/j.ero.2025.12.002
- Best for
- Practices that need durable evidence before rollout
3. Specialist-built, not a generalist tool
DAS28, UPDRS, pathology macros, GOÄ + EBM as standard.
Strengths
- Built by pharmacists with clinical experience
- German billing codes (GOÄ, EBM) as standard feature
- Specialty-specific templates, not one-size-fits-all
Limitations
- Generalist tools (Doctolib, Heidi, jameda Noa) are cheaper if you don't need specialty-specific templates
- For general practice, paediatrics, or gynaecology without specialty-specific code logic, a generalist tool is sufficient
- Price
- Basic €0 · Practice Pro €89/month, or €69/month billed annually
- Best for
- Specialist practices in rheumatology, neurology, pathology, and adjacent fields
Frequently asked questions
What to know before switching
- Which AI documentation tool has peer-reviewed evidence in a German specialty?
- Nixi AI. The Knitza & Aries study was published in EULAR Rheumatology Open in December 2025 (DOI 10.1016/j.ero.2025.12.002). Setting: German rheumatology clinic, 88 consultations, 108 patients. Voize has a comparable study but for nursing care (Charité). Doctolib cites an internal study from its own organisation.
- Who stores transcripts and audio, and for how long?
- Nixi AI: neither audio nor transcripts beyond the session. Doctolib: transcripts for 48 hours (source: Doctolib Help Center article 40589528427156). Microsoft Dragon Copilot: 30 days by default for audio + transcript. Tandem and Nabla don't store audio but keep transcripts in the cloud.
- Which vendors train AI models on patient data?
- Doctolib uses pseudonymised patient data for AI training by default (opt-out available per practice; source: Doctolib privacy policy, February 2025). With Nixi AI, only anonymised data may be used to improve the service; personal data is not used for this (DPA § 12).
- Which vendors offer § 203 StGB-compliant contracts?
- § 203 StGB compliance requires an explicit data-secrecy obligation on the processor. Nixi AI includes this clause explicitly per customer in every contract. Doctolib and Tandem use standard DPAs without § 203-specific language. International vendors (Heidi, Abridge, Suki) often don't offer § 203-compliant contracting at all.
- Which vendors support German billing (GOÄ / EBM)?
- Nixi AI: native code suggestions for GOÄ and EBM as a standard feature. Doctolib: AI billing assistant in the all-in-one bundle. Tandem: SNOMED- and ICD-10-based, no native EBM. Heidi / Abridge / Suki: no German billing logic.
- Which vendors operate exclusively in the EU?
- Nixi AI: data centres in the EU and Switzerland (adequacy decision), stored in Frankfurt. Doctolib: HDS-certified in the EU. Tandem: EU. International vendors (Heidi, Abridge, Suki): USA or global, with data flows outside the EU.
Shorter documentation time in 82% of consultations
Start your free trial and see why.