The four AI tools I actually use (and what each one is good for)

There are probably 80 AI tools you could be using right now. I use four. That’s it.

I want to talk about why, because I think the most common mistake busy physicians make with AI is trying to learn too many tools at once. You don’t need a stack. You need a default. Pick one, get comfortable, and you’ll already be ahead of 80% of your colleagues.

Here are the four I open most weeks, what each one is best at, and the one task I would not use it for.

1. Claude

Best for: Long writing, document work, anything where I want a thoughtful second draft.

This is where I do most of my real work. Drafting a long email to my property manager. Writing a kids’ birthday party invite that doesn’t sound like every other birthday party invite. Reviewing a personal contract before I send it to a lawyer. Outlining a blog post (yes, including this one, although the words are mine).

Two things Claude does well that I lean on constantly:

• Long documents. I can paste a 30-page PDF, or upload it directly, and ask real questions about it. The summarization stays on topic. The citations are usually right.
• Voice and judgment. The default writing voice is the most natural of the bunch. If you ask it to sound less formal, it actually sounds less formal, instead of sprinkling exclamation points around.

What I would not use it for: Real-time information lookups. Claude does not have live web access in the way some other tools do, depending on the version. For “what’s the weather in Lisbon next week,” I open something else.

Pro move: Use Projects (the persistent workspace feature). I have a “Household” project, a “Real Estate” project, and a “Life After Call writing” project. Each one keeps the context I need so I am not re-explaining myself every time.

2. ChatGPT

Best for: Quick lookups, brainstorming, image generation, and the kid-projects-at-9pm tab.

This is my switchboard. If I have a question and I’m not sure what flavor of help I need, this is the one I open. The interface is the one most people know, and the features are the most plentiful.

Where it shines:

• Brainstorming. “Give me 10 names for a family game night themed around the 90s.” It’s fast, it’s good, and the iteration loop is tight.
• Image generation. When my daughter needed a poster about the water cycle the night before it was due, this is what I used. Two minutes.
• Quick web-grounded lookups. With browsing turned on, it’s good for “what time does this place close on Sundays.”

What I would not use it for: Anything requiring a long, careful read of a document. Not because it can’t, but because the experience is built around quick turns, and you can feel the model getting impatient on long context. I send that work to Claude.

Pro move: Use a custom GPT or a custom instructions block. Tell it once that you are a physician, what you care about, and what you don’t want it to do (like, say, give clinical advice). Save yourself the same disclaimer every time.

3. Gemini

Best for: Help inside Google Docs, Sheets, and Gmail.

This is the tool I use the least on its own, and the most while doing something else.

I live in Google Workspace for my personal stuff. Gmail. Docs. Sheets. Drive. Calendar. When I am inside one of those, the “ask Gemini” sidebar is just there, and it knows what I am looking at. That changes the math.

Specific uses I rely on:

• Long Gmail threads. “Summarize this thread and tell me what the actual question is.” Saves me five minutes I do not have.
• Sheets formulas. “Make a formula that flags any row where the rent collected is below the expected rent.” It writes the formula. I check it. Done.
• Docs editing. Tightening a paragraph. Rewriting a sentence I have rewritten four times already.

What I would not use it for: Standalone writing where I’m not already in Google’s apps. The standalone Gemini app is fine. It is just not where I gravitate.

Pro move: Treat Gemini as the assistant that already knows what you are looking at, instead of a separate tool. The value is in not switching tabs.

4. NotebookLM

Best for: Asking questions of a stack of PDFs, articles, or notes.

This is the underrated one.

You drop in sources, up to dozens of them, and then you ask questions of the whole notebook. It cites which source the answer came from. It will not invent things from outside your uploads.

A few real uses:

• Onboarding a new topic. I dropped in a stack of articles on short-term rental tax treatment, and asked it questions until I felt like I could speak to my accountant in plain English.
• Reading my own writing. I dropped in two years of my own journal entries (yes, journal entries) and asked it to identify the months I sounded most burned out. Useful. Slightly haunting.
• Decision packets. Pre-school decision time? Drop in the brochures for three schools and ask which one has the longest day and which one has the most outdoor time. Done.

What I would not use it for: Anything that requires general knowledge. NotebookLM is intentionally narrow. It will not answer “what’s a good restaurant in Lisbon” unless you upload travel articles about Lisbon first.

Pro move: Use the “audio overview” feature for long source stacks when you’re driving. It generates a podcast-style conversation about the sources. Sounds like NPR. Genuinely useful on the commute.

Which one should you start with?

If you have used none of these before, the honest answer is: pick either Claude or ChatGPT and ignore the other three for two weeks.

• Pick Claude if you do a lot of writing or document work, or if you care about the voice of what comes out.
• Pick ChatGPT if you want the most flexible all-purpose tool and don’t mind that the voice can feel a little eager.

The other two come later, once you know what kinds of tasks you’re actually using AI for. Gemini becomes a natural add-on once you notice you’re copying and pasting things into Google Docs. NotebookLM becomes a natural add-on once you find yourself reading three articles to answer one question.

What I am not using (and why)

For completeness, here are tools I have tried and decided against keeping in regular rotation, with the one-line reason.

• Perplexity. Good at sourced search, but I already have search-grounded answers in both ChatGPT and Gemini. Not different enough for me to switch.
• Local models on my laptop. A fun experiment. Not yet good enough at the things I care about to be worth the friction.
• AI scribes and clinical tools. Hard rule, see the disclaimer at the bottom of every post on this site. Not what I write about.
• The latest viral agent framework. I will try it, and if it survives a month in my actual life I will write about it. Most don’t.

The point

The point is not to know about every tool. The point is to have one default for the kind of work you do most, and to actually finish the task in front of you. The tool is not the work. The tool is just the thing that gets the work out of your head.

If you take one move from this post, it is this: open the tool you already have, pick one annoying non-clinical task in your week, and finish it faster than you would have without help.

That is the whole game.

Strictly non-clinical. Nothing on this site is medical advice. I do not post about patient care.