Arabic–English AI scribe: documenting bilingual consultations

An Arabic AI medical scribe has to do more than recognize Arabic. In a typical GCC consultation, the doctor greets the patient in Arabic, takes the history in a dialect, uses English for drug names and findings, then dictates the plan in English. A useful scribe follows every switch and still produces one clean clinical note. Verto was built for exactly this kind of visit.

This article covers why bilingual documentation is hard, what a good design does about it, and how to test a scribe with your own patients. If you are new to ambient scribes, start with our AI medical scribe guide.

How GCC consultations really sound

Most scribe marketing assumes one language per visit. Clinics in the UAE, Saudi Arabia, Egypt and the wider region rarely work that way. A single morning list can include:

• An Emirati or Saudi patient speaking Gulf Arabic, with an Arabic-speaking doctor who dictates in English.
• An Egyptian or Levantine doctor talking to a Gulf patient, each in their own dialect.
• An expatriate patient speaking English, with Arabic used only between staff.
• A family member who answers for the patient, sometimes in a different language from the patient.

Clinical vocabulary adds another layer. Doctors trained in English often keep drug names, doses, investigations and diagnoses in English, even mid-sentence in Arabic. That habit is natural, but it is the hardest thing for speech recognition to handle.

Four technical problems to understand

1. Switching languages between sentences

The scribe must detect the language of each segment and transcribe it correctly. Systems that lock one language for the whole session produce nonsense when the visit switches. Look for automatic, per-segment detection. You should not have to press a language button in front of the patient.

2. Mixing languages inside a sentence

This is harder still. When a sentence is mostly Arabic with one English drug name in the middle, the minority-language words can be transcribed imperfectly. Be wary of any vendor who says this is fully solved. Ask them to show you mixed sentences, not only clean ones.

3. Dialects

Arabic is not one spoken language. Speech recognition that works well for one dialect may struggle with another. Ask which dialect a scribe is tuned for, and test it with the dialects your patients actually speak.

4. The language of the record

Most GCC clinical records, codes and claims are kept in English. Patients, on the other hand, need instructions they can read. A good bilingual scribe keeps a faithful transcript in the language spoken, writes the clinical note in the language of the record, and gives the patient instructions in Arabic.

How Verto handles bilingual visits

Verto transcribes each spoken segment in the language it was spoken, automatically, with no setting to change. The rest of the design follows from that:

• Egyptian Arabic by default. Verto listens for Egyptian Arabic alongside US English. Other dialects, including Gulf Arabic, are usually understood, but recognition is tuned best for Egyptian Arabic. We say this plainly, so you can test it in your own clinic.
• Honest about code-mixing. Verto switches languages between sentences, not word by word inside one sentence. Doctors get the cleanest transcript when they finish a thought in one language before switching.
• Facts you can prove. Every fact in the English note links back to the Arabic or English utterance it came from, with audio replay when retention is consented. You can check a translated fact against the patient's own words.
• Arabic patient instructions with a dose check. The plan, medications and follow-ups can be turned into Arabic for the patient. An automatic check confirms that every number and dose unit (mg, ml, IU) from the English survives in the Arabic, and flags any line where it does not. Clinicians can edit any line, and the printed handout shows the English and Arabic side by side.

Speaker identification matters more in bilingual rooms. When a companion answers for the patient, or an interpreter repeats the doctor's words, the scribe must still know whose history it is. Verto uses a short voice enrollment so it can label the clinician reliably. Everyone else starts as the patient until you correct it, and each fact card shows who said it, so you can catch a companion's answer that was recorded as the patient's own.

The translation runs on Helix's own AI host. Patient text is not sent to an outside translation service. For clinics in Egypt, where Arabic-first practice is the norm, our clinic software for Egypt page explains how the scribe works with an Arabic front office.

Habits that help any bilingual scribe

Small changes in how you speak make a big difference to transcript quality, whichever scribe you use:

• One language per sentence. Finish the thought, pause, then switch.
• Say doses and numbers clearly. Keep a drug name, its dose and its frequency in the same language as the sentence around them.
• Summarize aloud. A short English summary at the end (“So, the plan is…”) gives the scribe a clean anchor for the assessment and plan.
• Watch the room. Close the door, keep phones and televisions off, and place the microphone between you and the patient.

How to test an Arabic AI scribe

Several scribes now offer Arabic support, including Heidi and Sahl AI. The right choice depends on your patients and your workflow, so test before you buy. A fair pilot looks like this:

1. Pick real visits that reflect your mix: Gulf, Egyptian, Levantine and English-speaking patients, plus visits with a companion.
2. Include hard cases: mixed sentences, long medication lists, laterality and numbers.
3. For each note, check three things: missing facts, invented facts and wrong numbers. Count how often you had to go back to the recording.
4. Check that you can trace a translated fact back to the original Arabic words.
5. Check where the note ends up. Does it write into your chart, codes and orders, or do you copy and paste?
6. Ask where the audio and transcript are processed and stored, and whether they are used for training.

The traceability question matters most when you cannot read the source language easily yourself. Our article on AI scribe accuracy and traceability goes deeper. Verto also works across specialty templates, from ophthalmology to OB/GYN, so test on the templates you would use every day.