A Guide to Cold Emailing

Everyone should be sending more cold emails (that is, emails without a prior contact) as a way of becoming more connected with their research community. As I tweeted once (ze tweet), almost every good opportunity in my research life is either the direct or secondary outcome of a cold email. To put some numbers on this the outcomes of my cold emailing have been

The problem is that people stress out about cold emailing too much. Yes, people will ignore many or most of your emails (a 10% response rate wouldn't be unusual). But importantly, no one will hold those emails against you unless you either sound like a huge jerk in your email or get mad at them when they don't respond. Let me stress this point: either you'll get a response or they'll forget you ever emailed them at all. You matter much less than you think. As a consequence of people forgetting, it's basically something with only upside. You might have an interesting discussion, open up a new collaboration, learn something new. It's a great way to build a network with researchers you respect and appreciate.

What makes sense as a cold email in a research context / what sort of emails am I talking about? Generally, they have something to do with trying to get a new project off the ground or to understand a paper more deeply. Most of the time it's one of these options:

Sometimes these emails lead to interesting conversations, sometimes they lead to collaborations, and occasionally they have led to internship offers. There's no real plan, I'm just looking to get answers to my questions and sometimes the subsequent discussion leads somewhere interesting.

It's important to keep in mind that the format of these cold emails has varied significantly throughout my career and are a strong function of how established I am in the sub-field of the person I've emailed. The formula is something like:
$$\text{level of specificity} = \frac{1}{\text{how established you are}}$$
. It definitely does get easier as you get further into your career; I'd say I currently have a 25% response rate on my emails but it used to be more like 10%. What content should these emails contain?

Alright, but that's too abstract. It's useful to work through some examples and why I think they're good or bad. Some examples, paraphrased for anonymity, that I have received, ranked from quite bad to good Quite bad:

These are generic, do not even attempt to provide a reason why I might want to help them or a connection, and have approximately zero chance of getting a response. With some low probability p I will respond to an email like this with useful advice although that advice is likely instructions on how to send more productive cold emails. In the future it's likely to just be a pointer to this guide.


These are specific to the reader and make them think that they are likely part of a small group of people that are capable of answering your questions.


The latter email is probably the best variant because it demonstrates:

  1. The author is familiar with the work and is actually interested enough to read a paper carefully
  2. The author understood the work well enough to think of interesting extensions of it
  3. The ask is specific.

It goes without saying that you should only send such an email if you're genuinely planning to do this work. Otherwise, it's a pretty manipulative thing to do. Of course, such an email is risky! Maybe you haven't understood the work at all and it's now obvious to the recipient that you don't understand. I'd say not to stress out too much about this. You understand what you understand, you can't change that except by time and work, and how someone responds to your lack of understanding is often a good measure of how kind they are. I find such kindness filters useful as effective mechanisms for filtering out people I wouldn't want to interact with in the first place.

Alright then, I hope you're convinced, have already started drafting emails and are taking your first steps towards fostering a more connected, kinder research community.

Quick addendum: some element of this is totally privilege. It's pretty well established that people with names like mine get better responses than people with names not like mine. I'm not sure what the solution is (outside of solving racism) and am quite interested if anyone has some good ideas here.