The Rapid Commoditization of AI Scribes
In late July, Doximity made headlines by announcing that its AI scribe—built on OpenAI’s GPT technology—would be offered free to all physicians on its platform. This is on top of its Doximity GPT, which was announced in May 2025, also for free to physicians as a writing assistant. For those of us working at the intersection of clinical practice and health tech, the announcement wasn’t a total surprise, but the implications are profound.
With this move, the AI scribe is no longer just a novel tool. It’s a commodity. It’s a must-have and table stakes for clinical applications and tools.
It signals a major shift in the market. Documentation assistants powered by large language models (LLMs) have gone from cutting-edge to commonplace in under two years. That shift comes with a new reality: Core scribe functionality alone is no longer enough to stand out.
So what now?
From Novelty to Necessity
The promise of AI scribes was always compelling: reduce administrative burden, combat burnout, improve documentation accuracy, and give clinicians back valuable time. Many of those benefits have been realized, especially in settings where scribes are well-integrated and optimized for clinical workflows.
But with commoditization comes risk.
As more players enter the space, many are relying on off-the-shelf LLMs—like GPT, Claude, or Gemini and wrapping them in slick interfaces. However, is there sufficient medical oversight, clinical validation, or product rigor for clinicians to take these offerings at face value and start using them in clinical practice with real patients? To the untrained eye, many of these tools may appear functionally identical, but for practicing clinicians, the differences matter. A lot.
The Risk of Shallow Solutions
The ease with which companies can build a minimum viable scribe using general-purpose LLMs creates a dangerous illusion: that all scribes are equally capable.
They aren’t.
There’s a vast difference between a general-purpose summarizer and a scribe trained on real-world clinical workflows, tuned for specialty-specific language, and tested for safety and efficacy in live environments.
Consider just a few critical questions:
Does the scribe actually understand clinical nuance, or is it just parroting probable text?
How does it handle complex, multi-condition visits in specialties like oncology or rheumatology?
Can it maintain structured data integrity for EHR ingestion?
How does it address variations in accents, code- and attention-switching, and interruptions?
These aren’t edge cases. They’re everyday realities in clinical care, and many LLM wrappers aren’t built to handle them. At least we haven’t seen the evidence to suggest that they can.
Clinicians Can’t Be the Guinea Pigs or the QA Department
One of the greatest dangers of this commoditization wave is the potential erosion of clinician trust. When users can’t tell the difference between a clinically robust product and a lightly skinned chatbot, they will inevitably become skeptical of all options.
That’s a high price to pay for the illusion of democratized access.
If clinicians don’t know which scribes are safe, accurate, and reliable, they will disengage, or worse, they’ll inadvertantly introduce risk into the patient record by relying on flawed output.
We cannot afford to treat clinicians as the quality control layer for unproven tech.
What Differentiates the Winners?
So what separates the AI scribes that will stick from those that will fade?
It's no longer about who can generate a SOAP note from an audio file. That capability is core, baseline functionality. Instead, winning solutions will differentiate based on:
✅ Seamless Workflow Integration
Not just “compatible with EHRs” but designed to fit into actual clinical workflows, with minimal clicks and cognitive burden. Integration with scheduling, templates, structured fields, and patient context will become critical.
✅ Specialty Intelligence
Generic LLMs aren’t enough. Clinical relevance requires tuning for each specialty’s vocabulary, documentation standards, and regulatory nuances.
✅ Real-World Adaptability
Clinical environments are messy. Patients interrupt, clinicians mumble, conversations are nonlinear. The best scribes must adapt to that messiness, not break under it.
✅ Transparency and Trust
Explainability features that allow clinicians to see why a note looks the way it does, validate what was (or wasn’t) captured, and correct it quickly will be essential for adoption.
✅ Measurable ROI
The ultimate test: Does the scribe meaningfully reduce time, improve documentation quality, and contribute to better outcomes or revenue capture? If not, it’s just another expense that clinicians and patients will ultimately have to pay. Clinicians will have to see more patients so that the healthcare system can afford the new AI tool, and additional expenses get transferred to patients in the form of higher copays and membership fees. If the AI scribe does not deliver on its promise, then will all that cost be worth it?
What Should Health Systems Do?
This moment calls for discernment. For health systems evaluating AI scribe solutions—or deciding whether to build, buy, or partner—now is the time to define what value actually means for your clinicians and your bottom line.
Key questions to ask:
What level of clinical validation has been performed?
Is there peer-reviewed evidence or real-world implementation data?
How easy is it for clinicians to edit and control outputs?
What’s the total cost of ownership, including hidden IT lift and training needs?
How well does it integrate with our current EHR and documentation workflows?
Does it support the specialties we need most?
Vendor due diligence must now include not just engineering claims but clinical and operational proof points. Don’t be swayed by flashy demos and magic beans.
A Call for Clinical Involvement
Finally, we must call on clinical leaders—those with real experience in the trenches—to be at the table for these decisions. Whether as internal champions, external advisors, or product collaborators, clinicians must shape how these tools are developed, deployed, and evaluated.
Because if we get this right, AI scribes can truly transform care delivery, not just by saving time, but by returning meaning to the clinical encounter.
But if we get it wrong, we risk another cycle of broken promises, disengaged users, and wasted investment.
Final Thought
AI scribes are here to stay, but only some will stay clinical impactful.
In a world where the tech is free, the true value lies in the design, deployment, and discipline behind it.
