My AI Scribe Journey: What I Learned in the First Two Weeks

Estimated read time: 10 minutes 

I recently made a pretty significant change in my practice.

After using a live scribe for about three years, I decided to try an AI scribe.

I went into this thinking it could potentially do 80–90% of what my live scribe was doing while costing dramatically less.  

What I didn't anticipate was how frustrating the transition would be.

There were days when I loved it.

There were days when I wanted to break my iPad.

There were multiple times when I almost gave up and went straight back to my live scribe.

But after about two weeks, something changed.

I started to understand how to use it.

And now I actually think the AI scribe may create better and more detailed documentation than I was getting before.  

Here's what I learned.

Why I Used a Scribe in the First Place

My original reason for hiring a live scribe wasn't primarily to save time.

I wanted better notes.

I felt rushed trying to document everything myself while still spending the amount of time I wanted with patients. My templates weren't good enough to solve that problem on their own.

Having a live scribe improved my documentation considerably.

But it also created another problem.

I became dependent on him.

When we moved to ModMed, my scribe learned where everything was and handled most of it for me. I knew how to review and correct the note, but I didn't necessarily know how to build everything myself inside the EMR.  

The AI scribe exposed that immediately.

Days 1–3: This Was Harder Than I Expected

On day one, my live scribe was basically sitting in the background fixing the AI's work.

Not exactly the efficiency breakthrough I was hoping for.

I quickly learned that I needed to be more deliberate.

I needed to verbalize the chief complaint.

I needed to describe my physical examination.

I needed to clearly state diagnoses and plans.

And more complex encounters—especially ulcers, prescriptions and procedures—were harder than straightforward visits.  

By day three, I actually gave up for half the day and went back to my live scribe.

That became a recurring theme during the first couple of weeks:

Try it → get frustrated → consider quitting → try again.

The Unexpected Problem: I Had to Learn My EMR

This was probably my biggest personal obstacle.

My live scribe had been doing so much inside ModMed that there were basic things I hadn't needed to learn.

Even something as simple as getting the correct fracture-management information into a note became a learning experience.

I initially blamed some of these frustrations on the AI.

In reality, some of them were my problems.

I needed to become a better ModMed user.  

That's an important distinction if you're considering an AI scribe.

The transition isn't necessarily just:

Human scribe → AI scribe.

It may also be:

Someone else operating your EMR → you operating your EMR.

Those are two separate learning curves.

My Breakthrough: The "Running Commentary"

One thing has made a major difference.

I tell patients that I'm going to do something I call a running commentary.

Essentially, I verbalize:

• What I'm seeing
• What I'm thinking
• What I'm doing

That gives the AI the information it needs to create the documentation.

But I discovered an unexpected benefit.

Patients hear my thought process too.

There are things we see during an examination that we may be thinking internally but never actually explain to the patient.

The running commentary forces me to verbalize those observations.

That helps the AI.

It helps the documentation.

And I think it helps the patient understand what's happening during the examination.  

AI Doesn't Mean You Should Throw Away Your Templates

Initially, I wasn't sure whether I should use AI or my existing protocols.

Eventually I realized:

Use both.

For common visits, I'm still using templates and protocols.

For example:

• Ingrown toenails
• Nail fungus
• Plantar fasciitis
• Achilles visits
• Shockwave follow-ups
• Wound-care documentation

My live scribe actually used some of his extra time to build more protocols for me.

The AI can then add information on top of those protocols.

For repetitive follow-ups, I can also use "new from old" and let the AI capture whatever has changed.

I'm finding that the best workflow isn't necessarily AI instead of templates.

It's:

AI + protocols + templates + dictation.  

Where the AI Scribe Started to Shine

After roughly two weeks, I started noticing something surprising.

The AI was capturing things that my live scribe—or I—might previously have missed.

If a patient came in for one problem but also discussed swelling, neuropathy, balance, hammertoes or another issue, the AI captured that discussion.

My documentation became more comprehensive.

I also really liked the HPI and physical exam documentation.

Instead of filling the note with fluff, it was capturing more of the actual conversation and examination.

By this point, I started feeling like I had what I jokingly called a new superpower.  

Where AI Still Struggled

It wasn't perfect.

Complex notes could still slow me down significantly.

Some examples included:

• Wound templates
• DME plans
• Laboratory orders
• Specialist referrals
• Certain self-pay procedures
• X-ray combinations
• Selecting the correct plans within ModMed

Sometimes the AI would pull in the wrong template or plan.

Then I had to delete it and figure out which one I actually needed.

Trying to solve those problems between patients was one of the biggest sources of frustration.  

Don't Try to Fix Every Complex Note Between Patients

This is something I would change if I started again.

At first, I wanted every note finished immediately.

That destroyed my flow.

When a complex note didn't work properly, I would sit there trying to fix it before seeing the next patient.

That created unnecessary stress.

I eventually realized that complex notes can wait.

Save them.

Fix them during a break, lunch, or another block when you have more time.

Simple notes can often be completed immediately.

Complex notes don't necessarily need to be.

Use the AI Processing Time

There is a short period after I finish recording while the AI processes the information.

Rather than sitting there waiting, I started using that time.

I'll enter:

• Follow-up timing
• Lab instructions
• Products the patient needs
• Orthotic instructions
• Shockwave timing
• Other instructions my staff needs

That turns what initially felt like wasted processing time into productive workflow time.  

If I Were Starting Over, I'd Do These Five Things

After going through the first couple of weeks, here's what I'd recommend.

1. Give It 2–4 Weeks

Don't judge the AI scribe after two days.

I wanted to quit multiple times.

By around two weeks, things started clicking.

Interestingly, getting comfortable with the AI took substantially less time for me than training my live scribe had taken.  

2. Learn Your EMR First

If you're going to use an iPad, become comfortable with the iPad version before adding another layer of complexity.

Know where your common plans, orders and templates live.

3. Practice on Your Most Common Cases

You probably have four or five conditions that make up a significant percentage of your day.

Practice those first.

Don't start by trying to perfect your most complicated wound-care patient.

4. Build Protocols for Common Orders

Make common tasks easy to pull into the note.

For me, that includes things such as LFT orders and nail samples.

5. Make Sure Your Internet Is Good

AI adds another technology layer.

If connectivity is poor, everything slows down.

And when you're already learning a new workflow, those delays become incredibly frustrating.  

AI Scribe vs. Live Scribe: Where I Landed

I'm still early in this process.

But my opinion has changed dramatically from the first few days.

Initially, I wasn't sure I could tolerate the AI workflow.

Now I'm finding that my documentation may actually be more detailed because the AI captures so much of what I say.

There's also another obvious difference.

My live scribe takes vacations and days off.

The AI doesn't.

That doesn't mean AI automatically replaces every live scribe or works perfectly for every physician.

It does mean I'm becoming increasingly convinced that this can work for me.

The important thing was surviving those first couple of weeks.

The Bigger Lesson

There's a practice-management lesson here beyond AI scribes.

When you introduce technology that's supposed to save time, don't assume it will save time immediately.

There's an implementation cost.

There's frustration.

There's learning.

Sometimes you temporarily become less efficient before you become more efficient.

If I had evaluated my AI scribe after day three, I probably would have stopped.

Two weeks later, I was starting to see what was possible.

So if you're considering an AI scribe, my advice is simple:

Give it enough time to get through the learning curve before deciding whether it works.