May 26, 2026
Written By Elizabeth Cifers
AI has settled in quite comfortably at the retina clinic. It listens to visits, drafts notes, suggests codes, answers phones, and keeps the schedule moving. In many ways, it is the most efficient “team member” you’ve added in years.
But unlike your staff, AI does not raise its hand when it is unsure. It does not say, “this part doesn’t make sense” or “you may want to double-check that.” It simply produces output—and does so confidently.
In retina, where documentation, testing, procedures, and coding must all align, AI without oversight is risky. That is why the conversation cannot be just about adopting AI. It has to be about governing it.
If any of this sounds even slightly familiar, it should. Retina has been here before. When electronic health records were introduced, they promised efficiency, consistency, and better documentation. To be fair, they delivered on some of that promise. Templates made documentation faster, and notes became more structured. The clinic ran more efficiently (well, after the complaining stopped).
But something else happened subtly along the way.
The individual note, you know the one that reflected the specific patient, the specific encounter, and the actual clinical thinking, started to disappear. In its place came something more uniform. More repeatable. More efficient. And sometimes, less accurate.
We learned (over time) that just because a note looked complete didn’t mean it was supportive. That cloned documentation reads well until someone reviews more than one note. That consistency, without intention, can raise more questions than it answers.
AI is following a very similar path. Just faster.
From Templates to Automation—Same Problem, Better Packaging
Templates required active participation. Someone had to select and modify them. When used in the exam lane, they had to at least glance at what was being pulled in. AI further streamlines that process.
Now the note doesn’t just populate. It is created. It flows. It reads well. It often sounds like you on a very organized day. And, just as with templates, it introduces a subtle shift: documentation begins to reflect what is typical rather than necessarily what is true for that patient on that day.
The difference is that AI does this with a level of polish that makes it much harder to challenge.
It doesn’t look cloned.
It doesn’t look templated.
It looks, well, right.
The goal was never to lose the individual note. It happened gradually, in the name of efficiency, and AI accelerates that trade-off. A detail that sounds clinically appropriate but wasn’t explicitly evaluated. A test interpretation that aligns with expectations but not necessarily with the image. A coding suggestion that fits the pattern but not the nuance of the decision-making.
None of these is a dramatic error. Humans make the same errors, but if we have learned anything, it is that if the documentation is close enough to pass muster, we let it slide. Once that happens with humans or AI, the errors tend to repeat. Because AI is not just efficient, it learns, so it remains consistently efficient, even if it is incorrect.
With EHR templates, drift was easier to spot. Notes looked similar. Language repeated. Patterns became obvious over time. AI drift is more sophisticated. The wording varies. The structure feels natural. The note reads as if it were thoughtfully written, even when parts of it were not.
The same fundamental issue remains, but it is far less visible. Documentation begins to reflect a polished, coherent version of care, yet occasionally disconnects from what actually occurred. Even when the physician reviews the note, they may let it stand because it is “close enough” and not enough to feel wrong in the moment.
But enough to matter later.
This is where most conversations take a predictable turn toward “best practices” and “implementation strategies.” But before any of that, there is a more immediate reality. Without governance, AI will run exactly as designed: efficiently, consistently, and without hesitation.
It will generate notes.
It will suggest codes.
It will fill in gaps you didn’t even realize were there.
And it will do all of that without ever questioning whether it should.
Governance is what creates that pause. Not to slow things down unnecessarily, but to ensure that efficiency does not replace accuracy. The note still reflects the patient. The coding still reflects the decision-making. The documentation still holds up when someone else reads it without context. Without structure, AI doesn’t just assist the workflow; it begins to shape it.
This is where the risk becomes less theoretical. Documentation is not written for the moment it is created. It is written for every moment it is reviewed afterward. And those reviewers, whether payers, auditors, or consultants, are not experiencing the visit. They are reading the record.
If that record reflects patterns rather than patients, consistency rather than clinical variation, or suggestion rather than decision-making, it raises questions. The same questions that happened with templated documentation:
AI does not answer those questions. You do.
The caution here is not hypothetical. Retina has already experienced what happens when efficiency outpaces oversight. Templates were adopted quickly, and governance followed later, usually in response to denials, audits, or uncomfortable reviews or audits.
AI has the potential to follow the same path. The difference lies in the speed.
What took years to evolve with EHRs can happen much faster with AI. The drift, the normalization, and the quiet acceptance of “this looks fine”—all of it accelerates. That means governance cannot be an afterthought.
AI is not the problem. It is a powerful tool that, in many ways, improves how clinics function. But it is also very good at producing content that looks complete, sounds confident, and moves forward without resistance or afterthought.
And retina has already learned (sometimes the hard way) that those are not the same as accurate, individualized, and defensible. At some point, the question is no longer how efficient the clinic has become. It’s whether the documentation still reflects the care that was actually provided.
And if that answer isn’t clear, AI won’t be the one to blame.
AI will keep making your documentation look right, but that doesn’t mean it will hold up. If your notes were reviewed across multiple encounters, would they still make sense?
If you’re not sure, have ECC review them with you before someone else does.
Clear documentation. Accurate coding. Defensible claims. Mastering Retina Documentation & Coding: From Exam to Claim
👉 https://course.elizabethcconsulting.com/
.png)
