Start with what the system should NOT do
Before asking what the tool does, ask what it refuses to do. A good AI medical record system should never fill in a diagnostic hypothesis, ICD-10 code or treatment plan on its own — those fields need to be reserved for the doctor by default, not an option hidden in settings.
Questions about the usage model
- Does it work for both in-person visits and telehealth, or only one of the two?
- Can you attach audio already recorded on another device, or does it only record live?
- Does it need an internet connection the whole time, or can you record offline and process later?
- How many computers does the license cover — and what happens if the clinic grows?
Questions about privacy and sensitive data
- Where is the team's voice stored, the one used to identify who's speaking?
- Is there a central patient registry, or does identification happen only through the content of the conversation?
- Is one clinic's database isolated from other clinics using the same system?
Questions about the final record format
- Does the generated text come out in an editable format (Word, for example), or only in a closed screen within the system itself?
- Can the record's fields be customized to match how the doctor already works, or is the template fixed?
- Do medications appear by the scientific name, converted from what the patient said, or as-is?
The right system is the one you can audit
In the end, the most important criterion isn't how "smart" the system seems — it's how easy it is to audit what it did. An AI-generated record you can check line by line against what was actually said in the visit is more trustworthy than one that feels magical but is opaque.
Want to see this working in your clinic?
Talk to Marcelo