Thoughts on writing PhD reference letters.

2025 was the first year where I wrote reference letters for research students who I supervised. In this post I will explain my strategy towards writing letters and how students can best support this. Feedback is welcome (especially from more senior people who've written many letters).

Working backwards: letter as a tool for admission

The point of a reference letter is to help a professor (or admissions committee) make an appropriate decision to accept or reject a student's application. I view being a referee a bit like being a defence lawyer: you make the best possible case for the applicant to persuade the "judge" to admit them, while still remaining truthful. When assessing PhD applications, I imagine the key questions are:

1. Will the student succeed in the PhD program if admitted?
2. Will the student find a suitable advisor in the department?
3. Is the student among the best students to admit (compared to other applicants)?

Therefore, my goal with the reference letter is to answer these questions as positively as possible (while still being honest).

Standard outlines do answer these questions

Claude 4.5 suggested the following reference letter outline:

1. Brief context (how long you supervised them, on what)
2. Overall assessment with comparative ranking
3. Detailed research contributions with technical specifics
4. Intellectual qualities and growth trajectory
5. Fit with target program
6. Unambiguous recommendation strength

A letter with this structure actually answers all the questions above. Points 1-6 all contribute towards success in the PhD program, (3,4,5) all relate to finding a suitable advisor, and (2,4,5) can be used to compare between different applicants. So far past letters I've written have mostly followed this structure.

I'll go point by point through the outline and give my comments.

1. Brief context: easy, neutral / objective, no further comments.
2. Overall assessment: this is most challenging for me personally since I am not a professor and have not worked with a lot of students. So far I have qualified my assessment with the fact that I am fairly junior, and instead commented on whether they seem roughly on par with the PhD students in my PhD lab. I try to minimize this section because I don't think my claims here carry a lot of weight. Instead, I will try to elaborate more on points 3-5.
3. Research contributions: here I try to explain what parts of the project were challenging and explain the most difficult work that the student did. However, I realize I can't always recall precisely what parts of the project were the student's initiative vs my suggestions. So far I have erred on the side of describing things which unambiguously were the student's work.
   • Note to self: take better notes when meeting with students!
   • Note to students (also repeated below): it would be helpful to get a reminder of what some of your best research contributions were so I don't miss them!
4. Intellectual qualities: this is basically a placeholder to write nice things about the student. As a default I will write what skills the student learned in the project and how they seemed to improve. Usually this is a mix of programming skills and fundamental knowledge about the topic of the research project (eg Bayesian optimization, Gaussian processes). Beyond this I'll insert anything (positive) that comes to mind. Eg: curiosity, critical thinking, communication skills.
5. Fit with target program: although this feels like a potentially important section, so far I have not included it in any letters so far. Ideally what I would write here is:
   • How the student's research interests overlap with those of one or more professors. However, I would only write this if I actually know the professors' work well: the student will probably mention the same professors anyway, so I feel I mainly add value if I think I can write a better case than the student, which will probably only be possible if I understand the professor's body of work better than the student (possible if it's within my core research expertise) or if I know the professor personally (and therefore have a more nuanced understanding of who they would like to work with).
   • How the student's research interests are "especially" suited to a particular department. Usually this is not the case because most departments are sufficiently general, but I would mention if I feel the department is particularly strong in the student's proposed research (something I might be better-positioned to judge than the student themself).
6. Unambiguous recommendation: basically a summary where I try to re-state my "overall assessment" (point 2) again.

What information should students provide?

Strictly speaking I don't need any information to write a reference letter, but as mentioned in the previous section there are some points where information would be helpful:

• A reminder of the student's research contributions and the most difficult tasks they accomplished (helps to write section 3).
• Who their intended supervisor(s) are in the department (helps with section 5)
• Their proposed PhD research (helps with section 5).
• (optional) Any intellectual qualities the student wants to highlight (used in section 4, although I am less likely to include this in the letter compared to section 3).

The section where extra information from the student can make the biggest difference is section 5 (fit with the target program). By default this is something I won't write because I think it's better to say nothing than to say something that contradicts the student's application. For example, if I write that the student could work with Professor A but the student only mentions Professors B,C,D in their application, all parties will probably view this as a bad sign.¹ A second example: I wouldn't want to write a letter which assumes the student's PhD research will be similar to the project they did with me, because if the student's proposed research is different it will seem like I am not recommending them for their new research direction (and switching directions between Master's and PhD is super common).

However, a "fit statement" is probably valuable to the department making admissions decisions because it helps answers Q2 (advisor) and Q3 (is this student the best student), so I'd like to include it if possible. Providing me with the information required for a "fit statement" necessarily means that some parts of the application (the ones the letter might reference) need to be done (or almost done) before asking for the reference letter.² This moves the timeline of the entire process forward, leading me to my next point.

Timeline: please give me 2 weeks to write the letter

I ask students to give me at least 2 weeks notice to submit the letter (preferably more). Why is this?

Going through the previous sections, the actual tasks involved in writing a good letter are:

• Reviewing their research project to create a narrative of what they did and what their accomplishments were (~20m).
• Reflecting on what their best intellectual qualities were (~15m, usually requires a bit of thinking to recall details from previous meetings)
• Forming and justifying an "overall assessment" (~10m)
• (if info provided) Reviewing their research proposal and proposed supervisors to describe the fit with the target program (~20m)
• Actually writing the letter (<30m)

All together this is maybe ~1.5h of work. The reason I ask for 2 weeks to complete this work is to account for the fact that at any given time I may have other tasks competing for my time (eg other research projects, project deadlines, attending a conference, holiday). It is very unlikely that I will be unable to find a 1.5h block in a 2 week timeframe, regardless of what other tasks I have. However, there are times when I might not be able to schedule a 1.5h block in a 5 day window (example in footnote³).

Asking for multiple letters

The 1.5h quote above applies to the first reference letter. For additional letters, the reflection/review work can be re-used, saving some time. However, the "fit with the target program" section cannot be re-used because it will necessarily change for each letter. Furthermore, from what I have seen each application usually has a slightly different prompt for its reference letter, requiring some customization. Therefore, I would conservatively budget 1h to write a strong additional reference letter. This is shorter than ~1.5h for the first letter, but not that much shorter! Therefore I would still like 2 weeks notice.

A possible exception to the 2 week notice time is if a student wants a bunch of similar reference letters around the same time (eg for US PhD applications which are all due in December). In this case I would prefer a big request up front with clear timelines instead of a bunch of small requests. This allows me to plan the work more efficiently. You should send the request around the time you decide to apply, not just close to the deadline. This will probably be ~6 weeks in advance.

Examples of good and bad requests for reference letters

Good (single request)

Nov 28 Hello Austin, I am applying for a PhD in X at University of Y and need 3 referees. Are you willing to write a strong reference for me based on my [master's research / undergrad research / etc]? The reference is due December 15.

For my PhD research I would like to focus on topic T. I am applying to University of Y because Professor A is an expert in topic T. I might also want to collaborate with Professor B.

To help write the letter I've attached my draft statement of purpose for University of Y's application, an up-to-date CV, and my master's thesis. In case it is helpful to you when describing my master's research, in my opinion these were my biggest contributions to the project:

• Writing a differentiable solver for sparse matrix in Jax (idea to write a solver was from you, the implementation was entirely me).
• When method K didn't work I suggested studying parameters L,M,N and eventually discovered that we had misspecified N.
• I suggested the initial modification to method K which we collaboratively refined over several iterations to become method Q.

Thanks, Student

Analysis: this email is sent > 2 weeks before the deadline, includes a summary of contributions, information on the intended supervisor / research topic, and some bonus information (CV + master's thesis) which could be useful (better to include it than to not include it).

Good (batch request)

Oct 15 Hello Austin, I've decided to apply for PhD programs this year in the rough topic of T. Would you be happy to write a strong reference letter for some (or all) of the programs? Their deadlines are mostly in early-mid December.

My shortlist of programs and advisors is as follows:

• University of X:
  • Intention: top choice, will definitely apply
  • Deadline: Dec 5
  • Potential Supervisors: Professors A, B
• ...
• University of Y:
  • Intention: "backup" school, not top choice but hopefully less competitive
  • Deadline: Dec 1
  • Note: they only require 2 referees (not 3) and therefore I might not need a letter from you here (I haven't decided who it would be best to ask)
• ...
• University of Z:
  • Intention: on the fence, might not apply, just 1 professor.
  • Deadline: Dec 10
  • Potential supervisor: Professor Z

In total this is between 3-8 letters, depending on exactly where I choose to apply. If you are happy to write the letters, I will send more information closer to the deadline. For now, here is my up to date CV.

Best, Student

Nov 10 Hello Austin, I've made a final decision about where to apply. I will apply to universities X and Y, but not university Z. I would like you to be one of my 2 referees for university Y. I am still working on my statement of purpose for each application, but have attached a rough draft for each one to give you a sense of what I am thinking. I don't anticipate the key details of topic/supervisor changing.

When describing my master's research, it might be helpful to note the following contributions:

• ... (same as before)

Best, Student

Nov 28 Hello Austin, I've finalized my applications to universities X and Y, my final application materials are attached to help you customize the letters. Thank you very much!

Dec 05 Hello Austin, apologies for the last minute request, but I realized professor P at university Q would actually be a very good fit. I'd like to apply for a PhD position at university Q, although the deadline is really soon (Dec 11). Are you able to write a last-minute reference for me? My application package for university Q will essentially be the same as for university X previously, so I imagine you can use a similar letter.

Analysis: the initial email was sent well in advance, which would allow me to anticipate the requests coming later and plan around them. The email with the final requests came ~3 weeks before the first deadline (lots of time). Including an initial (unpolished) draft is more helpful than waiting until the last minute to include the final draft, but it is good to send the final draft anyway (which the student did). Finally, I realize that not everything goes according to plan so I included one extra email with an unexpected application. Although there are only 6 days left, the student gives a clear reason for applying to this school in particular and makes it clear that the letter for university X will be most similar. This information would help me save time drafting the letter and hopefully compensate for the relatively short notice.

Bad

Dec 02 Hello Austin, I'm applying for at PhD at X university and would like you to be a referee. The deadline is Dec 10. Are you able to do it?

Analysis: this ask has no context about the supervisor / program so makes it impossible to write a statement about "fit". The notice period is 8 days, which normally wouldn't be too bad, but this year in particular I was travelling on those exact dates so wouldn't have had time to write a good letter. If instead the student contacted me earlier I would have ensured to submit the letter on Dec 1 (before I went to NeurIPS).

Summary

My approach to writing a reference letter is to make the best possible case for admitting the student to the program. This can be accomplished with a fairly standard letter outline. A student can help me by reminding me of their top research contributions and describing their proposed research/supervisor. Writing a letter that includes all this information will probably take ~1.5h and I'd appreciate 2 weeks notice to complete the letter (mostly to account for other tasks I might have on my plate). If I get less than 2 weeks notice I may need to do a worse job. I'd appreciate the same 2 week notice even if I previously wrote a reference letter, and for large batches would appreciate an up front notice. I ended with some example requests.

As I said at the beginning of the post, feedback is definitely welcome!⁴ You can get in touch with me via email.

Extra bits

A few things I haven't fully thought through or formed an opinion on:

• What to do if I don't want to write a good letter? For example if the student wasn't good. I haven't encountered this so far.
• Mentioning weaknesses in a reference letter: ideally I think a reference letter should be a holistic statement and include some weaknesses, apparently reference letters skew very positive so mentioning any weakness might kill an application. So for now I probably won't include any weaknesses.