When AI in the doctor’s office makes mistakes: why Notat writes from facts first

NRK has reported concern about AI in doctor offices. The concern is real: when a clinical AI system writes text that sounds correct but contains deviations, the error becomes harder to catch. In healthcare, this is not a polish problem. It is a patient-safety and accountability problem.

We are not using the article as an endorsement of Notat. We are using it as a clear reminder that clinical AI architecture matters.

The risk with transcript-direct notes

Many AI scribes start with a transcript and ask a language model to write a note directly from that conversation. That can look impressive in a demo. Real consultations are messier: patients correct themselves, clinicians consider options before making a final plan, medications are discussed twice, and important negative findings may appear in a single sentence.

When a model writes straight from a raw transcript, it can smooth over uncertainty, make tentative decisions sound final, or merge an early option with the actual plan. The note can still sound professional.

FactsContext™: facts first, note second

Notat is built differently. Our patent-pending FactsContext™ engine extracts structured clinical facts first:

• symptoms and duration
• relevant negative findings
• examination findings
• medications and changes
• assessment and plan
• ICD-10-relevant details
• safety-netting and follow-up

Then the note is generated from those facts. Not straight from the raw transcript.

The most important part: clinicians see the raw facts

Notat shows the raw medical facts to the clinician. That means the clinician can read the note and inspect the context behind it. The question “where did this sentence come from?” should have an answer.

That is the core of FactsContext:

• The AI understands the encounter before it writes.
• The note is generated from facts, not guesswork.
• The facts are visible and reusable for coding, referrals, patient summaries, or the chart.
• The clinician always reviews and signs.

What we do not claim

No serious clinical AI should promise zero errors. Notat does not replace clinical judgment, and clinicians must always review notes before use. That is why we avoid polished accuracy percentages without methodology.

What we can say clearly is this: an AI scribe that shows the medical facts behind the note is easier to verify than one that only returns polished prose.

Why FactsContext is a better safety model

When the consultation becomes structured facts before it becomes a note, the clinician gets a review checkpoint before the record is used:

• Are the facts right?
• Is anything important missing?
• Is the assessment supported by what was actually said?
• Is the code supported by the documentation?

That is where clinical AI should move: facts first, transparency always, clinician control at the end.

Read more about FactsContext™, our clinical AI evaluation method, or compare Notat with other tools on the comparison hub.