Your PhD Application, Seen from the Other Side

Fri, 02 Oct 2026 00:00:00 +0000

Our faculty met today to discuss PhD applications for Spring 2027. A few hours later, two more emails from prospective students arrived in my inbox.

Reading them, I noticed how strange my position is. Not long ago I was the one writing those emails. Now I am the one reading them, and from this side it is obvious that there is an information gap. Applicants put real effort into things that tell me nothing, and underinvest in the things I look at first. I did the same when I was applying, because nobody told me how the file reads from the other side.

So here is how it reads to me.

One caveat before I start. This is the view from a program like ours at NIU. If you are applying to top-ranked programs, the bar and the weighting are different, and some of what follows may not transfer.

The question I am trying to answer

When I open an application, I am trying to answer one question: is this an excellent student, or one with real potential, and how likely are they to succeed in a PhD?

I cannot observe that directly, so I look for evidence. The evidence I trust has two properties: it was expensive to produce, and I can check it. A transcript has both. A line that says “proficient in PyTorch” has neither. Almost everything below follows from that.

What carries information

Grades. This matters more than applicants seem to think, especially for students applying without a master’s degree. I do not expect someone coming straight from a bachelor’s program to have meaningful publications, so the transcript is the main evidence I have. I read the math and programming courses most closely: calculus, linear algebra, probability, algorithms, data structures. A high GRE quantitative score helps for the same reason.

Publications at good venues. A paper at a venue I respect is as strong a signal as an excellent transcript. The venue matters, and so does the paper, because I will open it.

Recommendation letters with substance. What the letter says matters more to me than who signed it. A specific paragraph from a lecturer who supervised your project for a year is worth more than three generic sentences from a well-known name. I have read letters in which the writer admits to having rarely met the student. I have read sets of letters for one applicant that were nearly identical to each other. I have read letters that contained negative remarks, and letters that contradicted the CV or the transcript. You may never see your letters, but you choose who writes them. Choose people who know your work well enough to describe it.

A professional CV. Your CV is the first piece of your writing I see, and I treat it as a work sample. Publications should be listed with full venue and year. There should be no typos. Typesetting matters too: a CV written in LaTeX usually looks better than one written in Word, and it tells me you already use the tool you will write papers in.

What carries almost none

Hackathon placements, minor competitions, and certificates. Unless the competition is one I know to be hard, a ranking in it tells me very little. Online course certificates tell me less.

Small projects. A list of course-project-sized items (a sentiment classifier, a to-do app, a chatbot wrapper) does not separate you from other applicants, because nearly all of them have the same list.

Lists of languages and packages. Anyone can type “Python, C++, TensorFlow, PyTorch”. I cannot verify it, so I do not weigh it. Public code that I can read tells me far more.

What works against you

This is the part I think applicants miss most. Weak material is not neutral. Every item you include invites me to evaluate it, and I do.

Links that show inactivity. If you give me a GitHub link, I will click it. If the contribution graph shows a handful of commits over the past year, the link has hurt you. The same goes for a LeetCode profile with a small number of problems solved. Either would have been better left off.

Weak publications. A paper at a venue with no real review, full of errors, is not a small plus. The most common kind I see takes an off-the-shelf CNN or LSTM, runs it on a classification dataset, and reports the accuracy. These papers are worthless to me as evidence of research ability, and they show me the standard you are willing to put your name on.

Long descriptions of weak research. Some statements spend a page or more on past projects that are poorly motivated or poorly executed. The length does not help. It gives me more to find problems in. If a project was a learning exercise, two sentences are enough.

Claims the rest of the file does not support. If a CV lists reviewing for top conferences but shows no publications at those venues or comparable ones, I start to question the other lines as well. Once I doubt one claim, I read everything else less generously.

Stale interest in my work. Many emails say the applicant is interested in my research and then cite a paper of mine from three years ago, on a topic I no longer work on, without saying anything about it beyond the title. That tells me the applicant searched my name and took the first result. If you want to show interest, read something recent and tell me one thing you thought about it: a question, a limitation, an idea for what to try next. One real sentence beats a paragraph of praise.

What to do with this

Cut before you add. Remove every line and every link that would not survive a reviewer clicking on it. A one-page CV with a strong transcript, one solid project with readable code, and nothing else is a better application than three pages of filler around the same core.

Then spend the time you saved on the things that carry weight: your grades in the courses that matter, one piece of work you can defend in detail, and letter writers who know you.