What I Teach AI Before I Ever Ask It to Help Me Write
What I Teach AI Before I Ever Ask It to Help Me Write
Why I stopped using generic prompts and built a custom evaluator profile instead
One of the biggest shifts in my workflow happened when I stopped expecting AI to understand my job without any context.
Early on, I used it the way most people do. I would type something in, hope for something useful, and then spend half my time rewriting the output or re-explaining what I actually needed. For simple tasks, it was fine. For evaluation work, it was not cutting it.
What changed everything was building a much stronger set of custom instructions. At this point, I think of it less as a settings box and more as a custom evaluator profile.
The tool now knows I work with emergent bilingual students. It knows I lean on the WISC-V in Spanish, the Batería IV, and the WIAT-4 depending on the referral question. It knows I use CSEP and cross-battery frameworks to interpret patterns across domains. It knows what an FIE is supposed to do and how I tend to structure one. It knows my reports need to be readable by a parent who is reading them for the first time at an ARD table, not by another diagnostician.
That is built in before I ever type a prompt.
What I actually mean by a custom evaluator profile
A lot of people hear "custom instructions" and think a few lines will do it. Something like:
I'm an educational diagnostician. Write clearly. Keep it parent-friendly.
That is a starting point. It is not enough for the kind of support I want.
My setup includes the types of evaluations I write, how I structure an FIE, the tone I want across sections, the reading level I'm aiming for, the bilingual and culturally responsive lens I bring to every case, the batteries and measures I use most often, and the tasks I am comfortable offloading versus the ones that stay in my hands.
That last distinction matters. I am not using AI to interpret scores or make eligibility decisions. Those are mine. I am using it to reduce the cognitive load on the parts of the process that do not require my clinical judgment so I can spend more of my energy where it actually matters.
What that looks like in practice
AI helps me the most with:
Organizing de-identified notes into logical report sections
Cleaning up repetitive background or demographic language
Rewriting a dense paragraph into something a parent can actually understand
Tightening up a section that says what I mean but says it badly
Explaining a pattern of strengths and weaknesses in clearer language
Keeping tone and structure consistent across a long document
What I am not doing is asking AI to decide what a student's profile means or whether that student qualifies for services. That judgment belongs to the evaluator.
Why the setup matters more than the prompts
A lot of people try to get better output by collecting better prompts. I understand the appeal. Prompts feel tangible. They feel like the shortcut.
But if the tool does not understand your scope, your audience, the frameworks you use, your writing style, and what your reports are actually for, even a well-crafted prompt will produce something generic.
The setup is what makes the tool useful across your workflow, not just for a single task.
Where the payoff actually shows up
The biggest benefit is not speed. It is that I am not re-teaching the tool from scratch every time I open a conversation.
I do not have to explain what BICS and CALP mean, why bilingual students need to be evaluated in both languages, what an ARD meeting is, or why it matters that a parent walks away from that table actually understanding their child's evaluation. That context is already there.
So I can use AI to take some of the weight off the repetitive, language-heavy parts of the process without giving up the parts that require an actual evaluator.
That is the point of the whole thing.
