Do AI Scribes Code Your Notes? ICD-10 Suggestions, Honestly
By Patient Square Team · · 5 min read
Short answer: a responsible AI scribe suggests ICD-10 codes; it doesn't code your notes. That distinction is the whole point of this page, and it's where honest vendors part ways from the rest. A suggestion is a likely diagnosis code, drawn from the visit, that you confirm. A coding engine claims to assign the final, billable code itself. We do the first. We don't do the second, and we won't pretend to. Here's the difference, why it matters clinically and financially, and what to ask any vendor that blurs the line.
Key takeaways
- An AI scribe should suggest ICD-10 codes, not assign or bill them. A person confirms the code against the documentation.
- A coding engine claims to finalize codes, often with E&M levels and modifiers; that takes on clinical and financial responsibility a scribe shouldn't.
- We offer ICD-10 diagnosis suggestions only. No CPT, no E&M leveling, no HCC, no claim submission.
- Suggestions still save time, because recalling the right code is the slow part; confirming a shortlist is fast.
ICD-10-CM diagnosis codes; recalling the right one is the slow part a suggestion speeds
person who confirms every code: you, against the documentation
CPT, E&M, or HCC codes we assign; suggestions are ICD-10 diagnosis only
Do AI scribes code your notes, or just suggest codes?
They should suggest, and the honest ones say so. Here's the line. An AI scribe listens to the visit and can surface the diagnosis codes that fit what it heard. That's a suggestion: a shortlist, a starting point. You look at it, check it against what's documented, and confirm or change it. The responsibility stays with you.
A coding engine is a different claim. It says: here's the final, billable code, assigned and ready. Often it goes further, attaching an E&M visit level and modifiers. That's not a documentation tool anymore; it's a tool asserting the answer to a question that carries audit and denial risk.
ICD-10-CM has more than 70,000 diagnosis codes, per the CDC's classification. The hard part for a clinician usually isn't deciding the diagnosis, it's finding the exact code among thousands of near-neighbors. A suggestion solves that without taking the decision away from you. For the full set of questions to ask a vendor, including this one, our 9-question evaluation scorecard is the companion guide.
Suggestion versus coding engine: the honesty table
This is the table to keep in mind when a vendor demo gets vague about "AI coding."
| ICD-10 suggestion (what we do) | Coding engine (what we don't) | |
|---|---|---|
| What it produces | Likely diagnosis codes from the visit | Final, assigned, billable codes |
| Who decides | You confirm against the documentation | The tool claims to decide |
| E&M / visit level | Not produced | Often claimed |
| CPT / procedure codes | Not produced | Often claimed |
| HCC / risk codes | Not produced | Sometimes claimed |
| Claim submission | Never | Sometimes implied |
| Who carries the risk | You, as always | Ambiguous, which is the problem |
The right-hand column isn't inherently wrong to exist; dedicated coding and billing systems are a real category. The problem is a scribe quietly implying it does that work when it doesn't, or doing it without the audit trail a billing system needs. Ask which column a vendor is actually in.
Why doesn't the scribe just finish the coding?
Because finishing the coding is a clinical-and-financial decision, not a transcription task. The code drives the claim. The documentation has to support the code. Get it wrong and you've got a denial, or worse, an audit exposure. That's a responsibility that belongs to a clinician or a certified coder looking at the whole chart, not to a draft an LLM produced from the conversation.
There's a simple integrity test here. A tool that drafts a note can be wrong and you'll catch it on review, because you're reading the note anyway. A tool that silently finalizes a billable code is asking you to trust an output you might not scrutinize the same way. We'd rather hand you a suggestion you'll obviously confirm than a "final code" that invites you to skip the check. The same logic governs our prescription drafts, which is why those pass a safety screen, covered in our prescription draft safety explainer.
What AI Scribe by Patient Square actually does with codes
Plainly: ICD-10 diagnosis suggestions, and nothing past that. AI Scribe by Patient Square is an ambient AI medical scribe that listens during the visit and hands back a structured SOAP note, ICD-10 suggestions, and a prescription draft, ready to review and sign about two minutes after the visit. The ICD-10 part is suggestions you confirm.
What we don't do, said clearly so there's no ambiguity in your evaluation:
- We don't assign CPT procedure codes.
- We don't compute E&M visit levels.
- We don't produce HCC risk-adjustment codes.
- We don't submit claims or touch your billing.
If your practice needs an automated coding or billing engine, that's a different product category, and the honest answer is that we're not it. What we are is a scribe that drafts a clean note and narrows the diagnosis-code search so you spend less time hunting through 70,000 codes and more time confirming the right one.
When is the coding honesty the deciding factor?
For some buyers it's the whole decision. If you're being pitched "AI that codes your visits," slow down and find out whether the vendor means suggestions you confirm or codes they assign. The first is a time-saver. The second is a liability question dressed up as a feature.
We think the honest version is the one that survives an audit: the clinician or coder confirmed every code, the scribe just made the confirmation faster. That's the position we hold, and we'd rather lose a deal to a "full coding engine" pitch than overclaim our way into one. The full price ladder, with no coding-engine asterisks, is on our pricing page.
If suggestions-you-confirm is the right fit, book a demo and watch how the ICD-10 suggestions appear against a real visit, then see the prescription draft safety explainer for the same honesty applied to Rx. The product is a scribe that respects where the decision belongs: with you.
Common questions
Do AI scribes actually code your notes?
A responsible AI scribe suggests ICD-10 codes from the visit; it does not assign or bill them. The difference matters. A suggestion is a starting point a coder or clinician confirms. A coding engine claims to produce the final, billable code. Most AI scribes, ours included, do the first, not the second.
What is the difference between ICD-10 suggestions and a coding engine?
Suggestions surface likely diagnosis codes from the conversation, which you review and confirm. A coding engine claims to assign the final code, often with E&M level and modifiers, ready to bill. The first speeds your work and keeps you in control; the second takes on clinical and financial responsibility a scribe should not.
Why does an AI scribe not just finish the coding for me?
Because coding is a human-confirmed decision with clinical and financial consequences. The code drives the bill, the documentation must support it, and errors create audit and denial risk. A scribe that suggests a code lets you confirm against the chart. A tool claiming to finalize codes is taking on a job it should not.
Does AI Scribe by Patient Square do CPT, E&M, or HCC coding?
No. We offer ICD-10 diagnosis suggestions you review and confirm. We do not produce CPT procedure codes, E&M visit levels, or HCC risk codes, and we do not submit claims. If your evaluation needs an automated coding or billing engine, that is a different category of product, and we will tell you so plainly.
Are ICD-10 suggestions still useful if I have to confirm them?
Yes, because the slow part is often recalling the right code, not confirming it. A suggestion drawn from the visit conversation narrows the search to a few likely candidates you check against the documentation. You stay the decision-maker, you just start from a shortlist instead of a blank field.