AI for Higher Ed (Part 3): What parts 1 and 2 didn't cover
Recommendation letters have an AI blind spot nobody's named yet, plus a full prompt library for your practice and your clients.
Part 2 of this series gave you the data hygiene protocol, the four highest-value workflows, and the client disclosure conversation. Read it first if you haven’t. This issue does not repeat that ground.
This newsletter grows from your shares. Send it to a fellow colleague today!
In today’s issue (free analysis):
What actually changed in the AI-detection landscape since our essay-voice issue, and the one tool now doing something none of its predecessors could
The recommendation letter blind spot: what AI can ethically touch in the brag-sheet-to-letter pipeline, and the exact point past which it becomes a different kind of problem
A quick, practical note on AI notetakers, and why the honest answer here is simpler than the internet’s current AI-panic content would have you believe
For paid subscribers:
The list-building workflow that produces a genuinely differentiated list instead of the same twelve schools every AI tool recommends
The exact brag-sheet prompt to use with a student, and the line it must never cross into letter-drafting territory
How to talk to a family about their own AI use, since most of them are already experimenting with it whether you bring it up or not
A full prompt library: ready-to-use prompts for your own workflow, and a separate set built to hand directly to students and parents
Our take: why marketing “we use AI too” may be undermining the actual case for hiring you
Paid subscribers get the full analysis every issue, including the practitioner frameworks in the section below the paywall. If this newsletter helps you stay ahead in your work, a paid subscription is one of the better professional investments you can make. Many readers expense it as professional development.
Don't take our word for it. Here's what the higher ed community thinks.
What actually changed since we last covered AI detection
When we wrote about essay voice and AI detection, the honest conclusion was that detection tools were unreliable enough that no family should build a strategy around one specific tool catching or missing anything. That conclusion still holds. What’s changed is that the market has stopped treating all detectors as roughly equivalent.
A University of Chicago Booth School of Business working paper, by researchers Jabarian and Imas, benchmarked the major AI detectors, including Turnitin, GPTZero, Originality.ai, and Pangram, against both raw AI-generated text and text run through “humanizer” tools designed specifically to evade detection. Pangram was the only detector that held a false positive rate near zero even after humanizer processing degraded every other tool’s accuracy. That result is doing real work in the market: Pangram has moved from a niche entrant to the tool that admissions-adjacent detection services and several institutions now treat as the reference standard, while Turnitin’s AI-detection product, disabled outright by Vanderbilt in 2023 over accuracy concerns, continues to lose institutional trust into 2026.
Worth being precise about, since this is exactly the kind of nuance that gets flattened in most AI coverage: GPTZero disputes this paper’s ranking, arguing the researchers queried the wrong field of its API and that on a corrected re-run it matches or beats Pangram. Read that dispute closely and it’s narrower than it sounds. GPTZero is contesting its own recall score, its ability to catch AI text at all, not Pangram’s false positive rate, the number that actually matters for a family worried about a human-written essay getting wrongly flagged. The two detectors may be closer on catching AI text than the headline numbers suggest. On wrongly accusing an honest student, the independent finding still stands unchallenged.
The detail worth carrying into a client conversation is not “Pangram is better.” It’s that Pangram’s output is more granular than a simple flag. It can distinguish AI-written from AI-edited text at something closer to the sentence level, which matters directly for the coaching guidance in our essay-voice issue: a student who used AI to reorganize a paragraph they wrote themselves is now more distinguishable, in principle, from a student who generated the paragraph outright, than detection technology allowed for even a year ago. That is a reason for tighter coaching discipline, not looser. The safest version of AI-assisted editing keeps the tool at the outer edge of a draft, structure, pacing, word choice at the margins, and never inside a sentence the student didn’t write first.
The recommendation letter blind spot
Every issue in this series has covered AI’s effect on work a student produces directly: essays, activity lists, brag sheets. None of it has addressed the piece of the file an IEC is closest to without ever touching: the recommendation letter.
You don’t write recommendation letters. That’s the point, and it’s exactly why this deserves scrutiny rather than a pass. IECs routinely help students prepare the brag sheets and resumes that get handed to teachers and counselors, the raw material a recommender draws from to write the letter. AI has made that preparation dramatically faster. It has also created a specific failure mode most IECs advising on this process haven’t been warned about.
Jeffrey Neill, director of college counseling at Graded: The American School of São Paulo, described the honest math of the job to EdWeek: a recommendation letter used to take him about three hours, and at least ninety minutes of that was spent collecting information from scattered sources rather than writing. That is the legitimate use case: AI aggregating a student’s transcript, brag sheet, and a counselor’s own notes into an organized summary the counselor then writes from. The AI never touches the letter. It only organizes material a human was always going to use, which is exactly the aggregation the American School Counselor Association’s own guidance already expects counselors to do before writing, gathering ample information about a student first, regardless of what tool does the gathering.
The line gets crossed when the aggregation quietly becomes drafting, when a brag sheet arrives so fully written, in such polished paragraph form, that a busy recommender is functionally being handed a draft letter to lightly edit and sign. A 2024 study on AI use in the college process found roughly one in three students and teachers had already self-reported using generative AI to help with essays or recommendation letters, and a co-author of the study has since said publicly she expects that number has only grown. Admissions readers who review hundreds of letters a cycle have documented pattern recognition for exactly what that produces: language that sounds plausible about a student without being specific to them, the same handful of adjectives, “resilient”, “curious”, and “hardworking” recurring because those are the words AI reaches for when it’s filling gaps rather than recalling a real memory.
This was already a known problem with brag sheets before AI entered the picture: college advisers have long warned that a generic, checklist-style brag sheet produces recommendation letters that read like every other recommendation letter, because every recommender working from the same template reaches for the same language. AI does not introduce that failure mode. It makes it much faster to produce at scale, and much harder for a recommender to notice they’ve fallen into it.
Here is the part of this that should actually change how you talk to a student about their brag sheet, not just how a recommender uses it. A student has no control over whether their counselor or teacher ran the finished letter through an AI tool, no visibility into it, and no way to defend against a flag if the letter reads like AI output. A student can write a flawless, entirely human essay and still end up with a file that raises questions, because of a choice a recommender made in a different part of the application the student never saw. It’s a real reason to make sure the brag sheet you help a student build gives a recommender enough specific, human detail that they don’t feel pressure to lean on AI to fill in what’s missing.
The standard worth holding, and worth explaining to a student before they hand a brag sheet to a teacher: AI can help a student remember and organize what actually happened, a specific class discussion, a specific project, a specific moment a teacher would recognize, but it should never generate the descriptive language a recommender is expected to supply from their own memory and judgment. This isn’t a new rule invented for AI. It’s the same confidentiality and honesty standard IECA’s own Principles of Good Practice already hold this profession to, applied to a tool that didn’t exist when those principles were written.
A quick, practical note on AI notetakers
Otter, Fireflies, Fathom, and the built-in transcription tools in Zoom and Teams have become close to standard in professional meetings generally, and plenty of IECs now use one to avoid typing notes by hand during a family call. To be direct about it: this is a sensible, low-risk thing to do, and the case for treating it as a serious legal exposure is thinner than a lot of current AI commentary suggests. A family who hired you is not adversarial to you, and the realistic path from “my consultant used a notetaker” to any actual dispute is close to nonexistent for the overwhelming majority of practices.
The one habit worth building anyway, because it costs nothing and removes any ambiguity: say so upfront. When a new family starts working with you, mention during intake or in your engagement letter that you may use an AI note-taking tool on calls so you can stay focused on the conversation instead of your notepad, and that they’re welcome to ask you to turn it off at any time. One sentence, said once at the start of the relationship, does essentially all the protective work a much longer conversation would otherwise be trying to do.
Everything above is diagnosis: what changed, where the blind spot is, what the actual risk looks like. Below the paywall is where this issue turns into something you can use immediately: the list-building workflow that keeps AI from handing every family the same twelve schools, the exact brag-sheet prompt and the line it can’t cross, how to actually raise AI with a family, and a full prompt library built for your practice and ready to hand straight to students and parents.
Higher Ed Insights is reader-supported. If you are reading this as a free subscriber and you work in college counseling, financial aid, enrollment management, or high school education, the paid section is written directly for you.
Keep reading with a 7-day free trial
Subscribe to Higher Ed Insights to keep reading this post and get 7 days of free access to the full post archives.