How to Cold Email PhD Students

Get research experience by asking for something else.

Note: I am writing this as someone who does research in ML/AI. Advice may not hold for different fields.

I frequently (about once a week) receive cold emails from UC Berkeley undergraduates asking if I have a project they could help with. I suspect this is because I am listed somewhere on the BAIR website. While there is nothing “wrong” with the way the emails are written, I believe the senders may benefit from a different approach. To better explain this approach, I must first describe what most of the cold emails I receive look like. While it is important to me that I include my motivation behind this post, I do so at the end so as to not slow down the pace for most readers.

What 95% of the Cold Emails I Receive Look Like

Almost every cold email regarding research has the same structure:

1. The student introduces themself with their name, year, and major(s).

2. They say they are interested in my work on XYZ topic¹. The well-prepared students may mention they read a paper I wrote and think it is somewhat interesting.

3. They will list their experience and qualifications: coursework, internships, etc. Then they will state their resume is attached.

4. The email will end with them asking if there is a project that would be a good fit for them, and potentially give their availability in terms of hours per week.

To be clear, there is nothing wrong with this type of email—it is respectful, to the point, and gives me all of the information I may want if the stars aligned and I was indeed actively looking for an undergrad.

However, I am not typically looking for undergrad assistants. But this does not mean that I am completely useless to them. Rather than trying to go from total strangers to research collaborators, students should adopt an in-between step to gain a bit of interaction first.

My Suggested Cold-Emailing Method

My proposed email template takes advantage of two facts:

1. Academia is a small world.

2. PhD students rarely get questions about their papers.

The first point is important, because it means that PhD students talk to each other. Multiple times, I have received emails from students proclaiming interest in incredibly niche topics which I have worked on briefly—only for me to ask the other students in my office to learn that they also were asked about different niche topics by the same student! While the small-world nature of academia hurts students taking a spray-and-pray approach, it can be leveraged to help those taking a more targeted approach when combined with the second point.

Since PhD students typically do not receive questions about their work, they are typically more than happy to help others understand decisions made in their methodology, explain experimental results, or fix any code issues happening when reproducing their results. Instead of sending 10 different PhD students emails about your interest in topic XYZ, take the time to read some papers and email 2 PhD students questions about a paper they wrote on that topic! It will take time to read the papers and come up with decent questions, so you will have to take a more targeted approach and contact a smaller number of students. The questions don’t have to be uniquely insightful. You can ask why they chose a certain definition for their method, about a certain step in the proof of a theorem, etc. If coming up with “good” questions about the paper is hard, you can try to take their code and run it yourself—and see if any questions arise about their implementation. By asking these questions, you can open the door for a quick 30-minute chat².

In short, if I were to cold-email a PhD student, I would do the following:

1. Introduce myself with my name, year, and major(s).

2. Subtly try to show interest in a particular topic through action—not by explicitly stating it.

   1. E.g. “I have been trying to learn more about mechanistic interpretability, and in doing so came across your paper [insert paper name].”

3. Ask/mention your question, but do so in a way that could lead to a 30-minute meeting (i.e., not in a way they could answer in a brief email).

   1. E.g. “I was reproducing your results to play around with your method and noticed that [insert some type of design choice]—I am curious about why you chose to set up this experiment this way.”

4. Follow up your question with an invitation to meet.

   1. E.g. “Is there any chance you would be able to chat about the paper for 30 minutes to help me better understand? I would be happy to meet over Zoom or go to your office, whichever works best for you.”

5. Don’t try to pitch them on your experience or attach your resume, just end the email politely.

I suspect that this type of email will have a much higher success rate than the ones jumping straight to research positions. But here, success means something different: establishing a relationship. Come to the meeting prepared, and ask good questions. Not all questions have to be about the paper; you can ask about best ways to learn more about topic XYZ, which classes they recommend you taking, etc. Then, at the end of the meeting, ask about potential follow-up work to the paper in question. You could mention things you think would be interesting to follow up on, or you could just ask “Are you planning on doing any follow-up work to the paper? If not, do you have any related projects in the pipeline?”. The follow-up work is your chance to ask if they could potentially take you on.

If they do have follow-up or related work planned, you should then ask if they could use an undergraduate research assistant. If they do not, you can ask if they know anybody who does similar work who may need one. Even if they cannot take you on, they are still a valuable connection. Remember fact #1: academia is a small world. If they have some sort of relationship with you, they can likely ask around and connect you to someone who is looking for an undergraduate assistant. Plus, then you have gained a mini-mentor in them that you can ask questions regarding broader academic interests and goals. It might take multiple meetings to reach this point, but the in-between step of having a human connection with someone in your field of interest can be a huge help.

Final Note: This Assumes Not Everybody Takes this Approach

The reason this should work is because almost nobody is doing this. If a PhD student gets a couple of emails per week asking to meet for 30 minutes, they are not likely to meet. However, if you are the only one asking serious questions about their work, the probability is high that they will be happy to chat. Part of this should be solved by the increased time barrier in taking this approach: it will be much harder for students to send many PhD students legitimate questions about their work than it is to email describing some broad interest, so hopefully this results in less emails. Otherwise, I guess I should prepare to be inundated with meeting requests from undergrads, regretting my own advice. Would that count as suffering from success?

My Motivation for Writing This

When I arrived at UC Berkeley, I was surprised at how few PhD students in machine learning-related fields came from large public universities. For example, in my Statistics PhD program, a strong majority of the students who attended undergrad in the US went to smaller private colleges (myself included). This is something that struck me as odd, as I know many incredibly bright people who went to large public universities.

I suspect that one (of many) contributing factors to this is the sheer size of the student bodies at public universities. I attended Rice, which has an undergraduate population of about 4000. UC Berkeley has about 33,000 undergrads. As a result, students here have a much harder time forming relationships with professors and other researchers in their subject area, making “research” more of a nebulous concept due to the difficulty in gaining research experience. Given that students who are able to obtain research experience are more likely to choose to do a PhD (and will probably have better chances in admissions), I expect this has a lot to do with the imbalance of public/private college attendees in PhD programs like mine. My hope is that by writing this, students at larger schools are able to create interpersonal connections with people doing interesting work, which in turn helps them get involved in research.

1. Sometimes, the topic they mention isn’t something I work on.

2. 30 minutes is the perfect amount of time for such a meeting. Long enough to have a substantive discussion, but short enough that the PhD student can find the time to take it.