"Cold" Emails (and the in-person equivalent)

Faculty get "cold" emails about research opportunities—emails asking to work with the prof without prior contact or not in response to a job ad—all the time. A "cold" professional approach is a completely fine and normal thing to do, but it's also the case that most such emails are ignored.

This page is about how to write a more effective "cold" email (or approach in person), but more so about how to get into a position where you can write a more effective "cold" email—at the end of the day, the best way to get a position is to actually be a good fit for it.

Opinions here are my own—different things work for different people, and different faculty have different opinions and preferences. YMMV.

First, why would a professor want students? There are a few overlapping reasons:

• The prof has a research program and needs help with getting things done
• The prof wants outstanding collaborators and mentees: no one wants to miss out on working with a future MIT PhD grad and Stanford professor just because they didn't respond to an email.
• The prof wants to work with students in general: it can be intrinsically rewarding to work with people and to help someone start their career, and it's also just part of the job for many faculty to mentor students.

A "bad" cold email basically says "you probably need students, I'm a student, can I have an interview." This sometimes works!

But a better cold email signals to the prof that they'll get the good outcomes they want from mentoring the student with the least amount of hassle.

• "I basically understand what you do and I can get the job done in terms of helping you" is in most cases the best case you can make.
• If you can make the case (humbly) that you're a future star, that can help (but obviously, don't overdo it).
• If you can make the case that you'll be easy to work with, that can help enormously, but that is challenging to do in a cold email—everyone says they're a good colleague. That is one of the reasons that prior connections and references help a lot.

The prof (and often, more so their students and postdocs than the prof themselves) know what it takes to help with their research. You often don't—you're looking for research experience!

Still, you want to somehow show that you will be helpful.

Obviously, your transcript is a piece of evidence: good grades mean precisely that you were asked to do a job (prep for an exam, do a course project), and you did it well. But that's not the only or the most important thing.

A portfolio of previous projects that are (somewhat) related to the prof's research can be very helpful. Students often call this "previous research experience." In my opinion: that's a misleading frame.

More than lines on the resume, I want to see: what were you working on, what was the challenge, what did you actually do that was difficult, what ideas did you bring, how much of an independent understanding of the project do you have (if it wasn't a solo project)? In a cold email, your job is to convey that concisely (ideally, with a link to a website where the prof can read more if they want). In person, I personally would carry a literal portfolio (folder) with printouts from my projects, with diagrams etc., ready to talk about it.

Ideally, you want to demonstrate understanding of and enthusiasm for the field. This is easier in some fields than others (e.g., good luck demonstrating understanding of string theory if you haven't studied it for a decade or two). Still, most cold emails do a pretty terrible job of it. Here is what you can do.

• Read recent papers. Try to form an understanding of them and convey that understanding in your email. That is hard work, but if you want a research position, presumably the reason you want it is that you want to do research! (Of course, we all have career goals and so on; but part of becoming a good cold emailer is developing the genuine disposition to do research.)
• The easiest way to demonstrate understanding of a paper is to mention how to possibly extend it. There is a "cheat" there: many papers actually have a "future works" section or talk about future work in the conclusion. (Some of the time, the prof won't think much of those: they are often boilerplate where the plan is obvious, or infeasible, or already being worked on, or all of the above.) I think echoing a "future works" section is better than nothing. The hard work of a good cold email (and research) is to come up with something actually good.
• You don't have to say you want to work on the prof's research. Reading widely, having an interest in something else that's related, and having concrete ideas about that can often be better.

A non-terrible idea is to try an LLM with this webpage and the prof's research being part of the prompt. As a start, this could work, but be careful and judicious.

Here are some Gemini-generated emails to me using this strategy, along with my comments:

Example 1

My comments: This is pretty good! The person (or their LLM) read my work and has a reasonable idea. But: there are several red flags. First, fine-tuning RoBERTa is a bit old hat as of 2026; that's fine, but I'd be looking for at least an acknowledgement of the fact that it's not state-of-the-art to know the person knows that. Second, the plan makes sense, but for someone with the experience being claimed, it just lacks ambition: they propose a straightforward extension that, if you think about it, has some significant issues with it. That is quite likely to end up as the actual project (and if they need a small project, then, fine, why not), but someone with the technical background being claimed would usually aspire to more than that if they are being genuine, or else they'd mention the challenges one would face with actually making the idea work.

Example 2

My comments: This is also pretty good! I would interview this person. Some issues: the "negative result" with HoG is actually not so impressive to me (or probably anyone), it's just the best that we could have come up with; praising it is a bit of a red flag, although I wouldn't actually hold it against an undergrad. The extension idea is so reasonable that we've already been trying it for a year now, it's just stubbornly not working.

So, am I just prompt-engineering you to send me dozens of emails? Kind of. I think that iterating with an LLM (while also thinking for yourself and making sure you don't say things that don't make sense and that you're not ready to elaborate on during the interview) can help.

I strongly suggest not generating the whole thing with an LLM—very smooth prose with no rough edges + red flags like the above strongly suggest "LLM" and reduce your chances. I already have a Gemini Ultra account, and I don't need a research assistant to prompt it for me.