A PhD application is a hiring decision. A professor is choosing someone to work beside for years, and the application has to answer the questions any hire raises: can this person do research, are they fun to work with, and do they want to do what we do.
I applied to PhD programs last year, and these are my notes on what I think matters. I hope this demystifies parts of the process and encourages you to apply. :))
1. A research paper is the easiest way in
A professor needs to know you can do research before they take you on. A published work is the most direct evidence and gets your application read.
If that feels out of reach right now, it does not need to be a full conference paper. Two lighter options that you can aim for:
- A workshop paper. The major conferences run workshops alongside the main event. Workshop papers are shorter, reviewed less strictly, and a far more realistic target for someone who has not yet done research.
- The ICLR blog post track. ICLR is one of the main machine learning conferences, and its blog post track publishes reviewed posts that analyze and build on existing research. No new result is required. What is required is that you understand the work and have something to say about it.
They both show engagement with the research community. It shows that you can read papers, analyze them, and build on them. That is the skill a professor is looking for.
2. Letters prove you are sane
A paper shows you can do research. Letters show you are a person someone would want to do research with.
Consider the professor's side of it. Taking on a PhD student is a bet of $300k to $500k, spread over the years you are in the program. Nobody wants to make that bet on a stranger. The letters are the reference check.
So the letters need to say that you are delightful to work with, and you can do research. (A letter that says you earned an A in a class says neither.) The right letter writer is someone who has watched you do research. (If you are unable to directly work with a professor, try working with their PhD students who can vouch for you to their PI, who can then write you a letter.)
3. Research alignment decides the interview
I cannot emphasize this enough. Once you have an interview, alignment is what makes or breaks it.
I had four interviews. I got into the best of the four, because that was the only place that did exactly what I wanted to do. At the other three I was trying to connect their research to mine and find some intersection we could both live with. That does not work. :((
So do not look for an intersection. Build on top of them. Read what the group has published, work out where it goes next, and arrive with that. A professor who hears their own next question coming from an applicant needs much less convincing.