Nine students ask you for a recommendation letter in the same two-week window every fall. Each one deserves something specific and true, not a template with the name swapped out. And each one has its own deadline, portal, and word limit. This is the part of the job nobody schedules time for, so it eats your weekend instead.
The fix isn't writing faster. It's front-loading the raw material so the actual writing takes twenty minutes instead of two hours.
Build one input document per student, not one letter
Before you write a single sentence, collect what's true and specific about this student in one place:
1. Two or three concrete moments, a project, a comment in discussion, a piece of writing that stuck with you
2. One thing they struggled with and worked through, colleges want growth, not just achievement
3. A sentence about who they are outside your class, if you know it, a club, a job, a sibling they care for
4. The angle the student asked you to emphasize, if they told you one (many students will, if you ask)
This takes ten minutes per student and it's the only part that has to come from your memory. Everything after this can be sped up.
Let AI build the first draft, not the substance
Once you have that input document, this is a legitimate place to use AI: paste your notes and ask for a full draft in a warm, specific, professional tone, addressed to an admissions committee. What you get back won't be right yet, it'll be generic in the connective tissue even when your input was specific. But it gives you a structured shape to edit instead of a blank page.
Read it back and ask: does this sentence apply to *this* student, or would it be equally true of any B+ student who participated in class? Cut every sentence that fails that test. Replace it with something from your notes.
The paragraph that actually gets remembered
Admissions readers see hundreds of letters that say "hardworking," "kind," and "a pleasure to have in class." The paragraph that gets remembered is the one with a specific scene: what the student actually said, did, or made. If your draft doesn't have one concrete scene by the second paragraph, stop and add one before you do anything else.
Managing the volume without dropping one
- Keep a simple tracker: student name, deadline, portal link, status (notes collected / draft written / sent)
- Write letters in deadline order, not alphabetical order, a letter due in three days always jumps the queue
- Send a short confirmation to the student once it's submitted; they're anxious about this and a "sent it, good luck" message costs you nothing
Our [student feedback writer](/student-feedback) is built for a related problem, turning specific observations about a student's work into language that reads like you, not like a form, and the same instinct applies here: specific beats generic every time, and the tool exists to speed up the polishing, not to invent the substance.
What doesn't get outsourced
The moments you remember about a student are the actual letter. AI can help you turn ten minutes of notes into a structured draft fast, but it can't supply the one detail that makes an admissions reader put the letter down and think "I want to meet this kid." That part is still entirely yours, the tools just protect the time you have left to write it.
