Skip to main content

How to Write a Letter of Recommendation for a Student

For professors, PIs, and managers writing grad school rec letters: the structure, the comparative calibration, and what committees discount.

Nirmal Thacker, Founder, GradPilot · CS, Georgia TechPublished Jul 15, 2026 · Updated Jul 16, 20269 min read
Free SOP ReviewExpert scoring + feedback

How to Write a Letter of Recommendation for a Student (Graduate School Edition)

You said yes, and now there is a portal email in your inbox with an upload link and a deadline. This guide is for the person writing the letter — the professor, PI, TA-supervisor, or manager recommending a student to a graduate program — and it is deliberately practical: the structure that works, the sentences committees actually quote, and the patterns that get letters silently discounted.

Most guidance on this topic lives on university career-center pages written a decade ago for employment references. Graduate admissions letters are a different genre with different rules, because they are read differently: a faculty committee is trying to predict whether this person can do sustained, independent, graduate-level work, and your letter is the closest thing to eyewitness testimony they will get. (If you are the student trying to orchestrate this process, start instead with our complete guide to grad school letters of recommendation.)

What the committee is doing with your letter

Three things, in roughly this order:

  1. Checking your basis of knowledge. How long did you observe this student, in what capacity, and how directly? A letter's weight is proportional to the witness's proximity.
  2. Harvesting evidence. Specific incidents — what the student actually did — that corroborate or complicate the rest of the file.
  3. Calibrating. Where does this student sit relative to everyone else you have taught or supervised? You have seen hundreds of students; the committee is borrowing your sample.

Everything below serves one of those three reads.

The anatomy of a strong letter

1. Open with a relationship statement

One short paragraph: who you are, how long you have known the student, in what capacity, and how directly you observed their work. "I supervised Maria's undergraduate thesis for 14 months and met with her weekly" is a credibility header. Skipping it — or burying it — forces the reader to guess how much your praise is worth.

2. Build the letter on incidents, not adjectives

The Common App's official guidance to recommenders compresses this into four words: "Anecdotes outshine adjectives. Always." MIT's admissions office asks for letters that are "specific and storied." The AAMC's official letter-writer guidelines — the closest thing this genre has to a published rubric — tell writers to describe directly observed behavior: the situation, what the student did, and what came of it.

Compare:

Weak: "Priya is a brilliant, hardworking student with excellent analytical skills and a real passion for research."

Strong: "When the assay Priya had spent six weeks building returned contaminated results, she traced the problem to a reagent batch, redesigned the protocol to include a control we now use lab-wide, and still delivered her section of the paper on schedule. I have watched many students hit that wall; most wait to be told what to do next."

The first sentence could be pasted under any name — and committees read ten thousand of it per season. The second could only be about one person, and it quietly demonstrates independence, debugging instinct, and follow-through without naming a single trait.

A useful drafting discipline: for every claim you want to make ("independent," "creative," "resilient"), write the incident that proves it, then consider deleting the claim and keeping the incident.

3. Calibrate comparatively — with a named group

The single most quotable sentence you can write looks like this:

"Of the roughly 60 master's students I have supervised over 15 years, she is among the strongest three."

Comparison group, time window, position. This converts your opinion into data a committee can use. Uncalibrated superlatives ("one of the best students I've ever taught") are the weak form; they gesture at calibration without committing to it.

Be honest with the numbers, for two reasons. First, your credibility across future letters depends on it. Second, most graduate portals will also ask you to place the student on a rating gridthe University of Florida's form uses a percentage grid, and Stanford's Knight-Hennessy program asks you to rate the student "relative to the group you specified". If your prose says "extraordinary" and your grid says "top quartile," the reader believes the grid and re-prices the prose.

4. Speak to what the program is predicting

For research-oriented programs (especially PhD), the committee is predicting research potential: independence, idea generation, tolerance for dead ends, and the ability to finish. If you supervised research, that material belongs at the center of the letter — our companion piece on what PhD admissions committees look for in letters unpacks this from the reader's side. For professional master's programs, evidence of analytical horsepower, writing, and learning speed carries more of the weight.

If you are a manager or employer writing for a returning student, the job is translation — converting workplace achievements into evidence of academic capacity. That case has its own guide: employer letters of recommendation for grad school.

Health-professions programs are their own genre with their own reader expectations — we keep dedicated writer-side guides for medical school and PA school, and veterinary school's VMCAS is the rare system that publishes an explicit content checklist for letter writers.

5. Handle weaknesses like a scientist, not a salesman

Assess; don't advocate — that is the AAMC's framing, and it is right for every graduate genre. One honestly framed growth area ("her first drafts run long; her revising discipline improved markedly over the year") makes the rest of the letter more believable. What you should not do is hedge the core endorsement. "I believe Daniel would likely do well in a doctoral program" reads, to a committee, as a warning delivered politely.

6. Close with an explicit endorsement tier

Committees read closing lines as a code: "I recommend her without reservation" / "my strongest recommendation" / "I recommend him" each mean different things, and the absence of a superlative in the final paragraph is itself information. End with the strongest sentence you can defend, and make it match everything above it.

Length, format, and mechanics

What gets letters discounted

A compressed blacklist, assembled from official reader guidance and the portal forms themselves:

  • Transcript recitation. The committee has the transcript. "Earned an A in my course" is not evidence; what she did to earn it might be.
  • Course description instead of student description. Yale Law's tip sheet for recommenders specifically warns against describing the course and re-summarizing the résumé — general praise letters get discounted, detailed comparative ones get quoted.
  • Adjective stacks with no incidents. See above; priced at zero.
  • Grid-prose contradiction. The checkbox says top 15%, the letter says "best in my career." Pick the true one, use it in both places.
  • Damning brevity. A three-line letter is read as a message, whatever it says.

About AI drafting (read before you paste into ChatGPT)

Letter-writing workload is real, and surveyed teachers openly report using AI tools to draft recommendation letters — about one in three, in an Education Week survey. Meanwhile, in our own audit of 174 university AI policies, not one addresses AI use in recommendation letters — every rule binds the applicant, none binds you.

But the policy gap is not the real problem; genericity is. An AI draft built from three bullet points produces exactly the adjective-stack letter that committees discount, and it produces it in the same register for every student you paste in. If you use drafting tools at all, the discipline is the same as the rest of this guide: the incidents, the calibration sentence, and the endorsement tier must be yours, from your observation, in your voice — and the draft needs a hard human edit before it goes near a portal.

One more scenario: if you have asked the student to produce the first draft — a common workload shortcut with genuine ethical and policy hazards, including an explicit ban at Stanford — read our guide to self-drafted recommendation letters before proceeding. Some programs treat it as an application-terms violation regardless of who suggested it.

Before you hit submit

A 60-second checklist: relationship statement in the first paragraph · at least two specific incidents · one comparative sentence with a named group and time window · endorsement tier explicit and consistent with the rating grid · one to two pages · letterhead + institutional email · the student's name spelled the same way throughout (it happens).

If you want a second set of eyes that is not a colleague, GradPilot reviews the letter you are drafting the way a committee will read it — relationship evidence, incident specificity, comparative calibration, and the generic-praise patterns above, with AI detection included in the full review. Paste in the draft, get the review, keep authorship entirely yours. We review letters; we never write them.


Sources

Get SOP Feedback

See how your statement of purpose scores on proven rubrics used by admissions committees.

Related Articles

Your Statement Deserves a Second Look

Rubric-based scoring and actionable feedback before you submit

No credit card required