DRD: Which is the best AI tool for me? The Best AI Tool for Doctors Doesn't Exist.

Every few weeks, a doctor asks me the same question.

"Which AI tool should I be using?"

I understand the instinct. Medicine trains you to find the right answer — correct diagnosis, right drug, right dose. So when a new instrument lands in your hands, you want to know which one is best.

But that question is built on a broken premise.

There is no best AI tool. There is only the right tool for the right job.

Why the Question Itself Is Wrong

Think about how you learned to use a stethoscope. Nobody called it the best medical instrument. You understood what it was for. You reached for something else when you needed something else.

Asking "which AI is best?" is like asking whether a stethoscope is better than a CT scan. The answer depends entirely on what you're trying to find out.

The four tools most doctors start with — ChatGPT, Claude, Perplexity, and Gemini — are all genuinely good. They're also genuinely different.

The Big Four AI: What Each One Actually Does Well

ChatGPT is the most versatile. Strong at drafting — discharge summaries, referral letters, patient handouts, research proposals. Reliable for writing tasks that don't involve patient data. What it isn't: a trustworthy source of citations. Ask it for references and it will confidently give you papers that don't exist. Never use a ChatGPT citation without independently verifying it.

Claude produces writing that sounds more measured and less mechanical. It handles long documents well — paste in a 40-page guideline and ask it to pull key clinical points, and it does that with fewer errors than the others. Good for drafting protocols, teaching case discussions, or reviewing lengthy documents.

Perplexity functions closest to a search engine with a brain. It answers questions with real, clickable citations. For quick clinical queries where you need verifiable sources, this is the most transparent of the four. It won't fabricate a reference and present it with confidence. In medicine, that matters.

Gemini is strongest with uploaded files. Upload a lab report, dense PDF, or clinical document and ask questions about it — Gemini handles this well. Google's Med-PaLM research also showed strong performance on clinical exam-style questions.

Where the Big Four Fall Short

All four general tools share the same honest problem: they were not built for medical research. They were built for everything.

For a clinician running a literature search, writing a systematic review, or building evidence for a clinical protocol, that's a liability.

Elicit searches across 138 million indexed scientific papers and grounds its answers in actual studies — not reconstructed training data. A study in Hepatology Communications compared ChatGPT and Elicit on clinical literature searches directly. ChatGPT hallucinated references. Elicit found real papers with accuracy matching human researchers. For literature reviews, Elicit is a different category of tool entirely.

Consensus is designed for evidence-grounded Q&A. Ask it a clinical question and it shows you where the literature agrees, where it conflicts, and where evidence is simply insufficient.

Scite checks whether a paper is still supported by newer research or has been contradicted. Essential for anyone updating clinical guidelines.

The emerging workflow among researchers: Elicit to find papers → Consensus for evidence questions → Scite to verify → Claude or ChatGPT for writing. General models last. Not first.

How to Compare Models Before Committing: OpenRouter

Most doctors haven't heard of OpenRouter (openrouter.ai). They should know it exists.

OpenRouter publishes live rankings of which AI models are actually being used for which tasks — writing, research, coding, summarisation — based on real usage data across millions of users, not marketing copy. If you want to trial a model before paying for a subscription, several are available free through the platform.

Their data from over 100 trillion tokens also shows something worth knowing: healthcare is the most fragmented AI category on the platform. No single tool dominates. Usage is spread across medical research, clinical guidance, and education. That's not a gap — it confirms what we're saying here.

A Simple Decision Map

Task	Tool
Writing reports, letters, handouts, SOPs	ChatGPT or Claude
Quick clinical queries with verifiable sources	Perplexity
Querying uploaded documents, PDFs, lab reports	Gemini
Literature review, evidence synthesis	Elicit
Evidence Q&A from published papers	Consensus
Verifying citations	Scite
Comparing models before committing	OpenRouter

The Trap to Avoid

Most doctors try a tool for three days, feel uncertain, and switch. Two weeks later, same story.

The tool is not the skill. Prompting is. What you get out of any AI tool depends far more on how you frame the question than on which platform you're using. Pick one tool for your single most time-wasting task. Use it for 30 days before judging it.

Start with your problem, not with the tool. The rankings of which model is "winning" change every month. That principle doesn't.

What's your single most time-wasting clinical task? Drop it in the comments — let's find the right tool for it.

Which AI Tool Is Right for You? (Answer 5 Questions)

Go through these quickly. Be honest.

1. What do you mostly need help with? Writing and drafting → ChatGPT or Claude Finding and verifying information → Perplexity Searching medical literature → Elicit Working with uploaded documents → Gemini

2. Do you need sources you can verify? Yes, always → Perplexity, Elicit, or Consensus Not critical → any of the four general tools

3. Are you doing research or writing a paper? Yes → start with Elicit, finish with Claude or ChatGPT No → skip Elicit entirely for now

4. Do you have 30 minutes to learn a new tool? Yes → try the specialised tool that matches your task No → stick with ChatGPT or Perplexity, both have near-zero learning curve

5. Are you unsure which tool to trust for your use case? Yes → go to openrouter.ai/rankings, filter by task, see what others are actually using