Introduction: the "post-GPT" era
For the last two years, the medical world has been in the "chat" phase of artificial intelligence. Clinicians experimented with ChatGPT, marvelling at its ability to pass exams but worrying about its tendency to hallucinate.
In 2025, the novelty has worn off, and we have entered the era of integrated clinical workflows. We no longer want a chatbot to talk to; we want a co-pilot to work with. The messy, crowded market of "AI for doctors" has finally segmented into three clear, distinct categories that solve specific problems:
- Scribes (They listen so you don't have to type).
- Retrievers (They find facts so you don't have to search).
- Reasoners (They brainstorm so you don't miss a diagnosis).
This guide cuts through the noise to identify the best-in-class tools for each category.
Category 1: the "knowledge retrievers" (decision support)
These tools are designed to solve the "search problem." When you are with a patient, you don't need a creative essay; you need a specific fact, dose, or guideline, and you need it instantly.
iatroX
Best for: UK Guidelines (NICE/CKS) & "Grounded" answers. The difference: iatroX is built specifically for the UK clinician. Unlike general LLMs, it uses a "grounded" architecture (Retrieval-Augmented Generation). It does not guess; it retrieves. When you ask about hypertension management, it doesn't give you a generic medical answer; it gives you the specific NICE NG136 protocol. Key benefit: Hallucination-free citation. If iatroX cannot find the answer in a trusted UK guideline (NICE, BNF, CKS), it will tell you, rather than inventing a plausible falsehood.
OpenEvidence
Best for: US-centric deep research. The difference: If iatroX is the specialist for UK guidelines, OpenEvidence is the academic librarian for the global literature. It excels at synthesizing vast amounts of peer-reviewed data from journals like NEJM and The Lancet. It is an excellent alternative for deep-dive research queries where US guidelines or global evidence bases are required.
Category 2: the "scribes" (documentation)
This is the most crowded sector. These tools use ambient listening to transcribe consultations and format them into structured notes (SOAP, etc.).
The contenders: Heidi Health / Corti / Nabla
- Heidi Health: A favourite for its "freemium" model and flexibility, allowing clinicians to turn a raw transcript into letters, referrals, and notes with custom templates.
- Nabla: Known for its speed and "zero-retention" privacy promise, aiming to generate the note before the patient has even left the room.
- Corti: Often found at the enterprise level (e.g., emergency dispatch or large primary care networks), Corti listens in to provide real-time nudges and safety checks ("Did you ask about chest pain?").
The differentiation: Scribes save you typing time; tools like iatroX save you thinking time. You likely need both, but they serve fundamentally different functions in your day.
Category 3: the "diagnostic partners"
These tools are designed to augment your clinical reasoning, acting as a "second opinion" for complex cases.
Glass Health
Best for: Complex differential diagnosis (DDx) brainstorming. The difference: Glass Health is not for checking a dose; it is for when you are stuck. You input a one-liner ("30F, recurring fever, joint pain, negative ANA") and it generates a structured schema of differentials to consider. It helps combat cognitive closure by suggesting "don't miss" diagnoses you might not have considered.
Comparison table
| Tool Name | Primary Function | Cost Model | Best For |
|---|---|---|---|
| iatroX | Knowledge Retrieval | Free / Freemium | UK Clinicians needing NICE/BNF/CKS accuracy. |
| OpenEvidence | Knowledge Retrieval | Free (Verified HCPs) | Deep Research & US-centric guidelines. |
| Heidi Health | Ambient Scribe | Freemium / Pro | GP/Outpatient notes & admin reduction. |
| Corti | Co-Pilot / Scribe | Enterprise | Real-time safety & decision support during consults. |
| Nabla | Ambient Scribe | Free Tier / Paid | Speed & privacy-conscious documentation. |
| Glass Health | Diagnostic Reasoning | Freemium / Pro | Complex DDx & learning clinical schemas. |
Verdict: the "stack" concept
The era of the "one app to rule them all" is over. In 2025, the most efficient clinicians will utilise a Clinical AI Stack:
- One for notes: (e.g., Heidi or Nabla) running in the background to handle the admin.
- One for decisions: (e.g., iatroX) open in a tab to verify guidelines, doses, and local protocols instantly.
By unbundling these tasks, you ensure you are using the sharpest tool for each job, maximising both safety and efficiency.