Your US Admissions Counselor May Not Beat My AI Planning Workflow

Show, don't tell. So instead of arguing the point, here is what the workflow actually produces — described concisely, with the document itself doing the talking.

The core idea: feed a student's background into GPT-4o, write the planning template and constraints into the prompt, generate LaTeX output directly, and compile to PDF. The result is not a vague list of suggestions. It is an executable plan.

Six design principles

1. AI and typesetting in separate lanes. GPT-4o for content generation, LaTeX for layout. The Gantt chart currently resolves to monthly precision; weekly precision is possible with additional work.
2. End-to-end generation. Student background goes in; LaTeX source comes out; PDF compiles. Minimal manual reformatting in between.
3. A human judgement layer on top. Beyond what the AI generates, I add SWOT analysis for each competition and specific recommendations for what to do at each phase. That part does not get automated.
4. Four-colour annotation system. Four colour-coded comment blocks distinguish content by type (see below).
5. Built around the Common App structure. Core sections are competitions and activities. Educational history, standardised test scores, and other elements are included but subordinate.
6. An explicit executability standard. Tasks are clear and actionable. Timelines are realistic. The student's profile is coherent. Learning progression is logical. Competition and activity choices are consistent with each other. There are specific expected outcomes for each phase, with fallback options built in.

The four-colour annotation system

Why counselors may not beat this

This is not a claim that AI understands admissions better than experienced humans. It is a claim that what most counselors actually deliver — vague timelines, generic advice, activity lists with no competitive logic — is something a well-prompted AI can surpass without difficulty.

What remains genuinely hard to automate is strategic judgement: which competitions suit this particular student, what the SWOT looks like for each one, how the long-arc narrative holds together. That is still my contribution. The AI handles structure and typesetting; I handle strategy and evaluation.

One important scope note: I only run this workflow for physics, computer science, and engineering profiles, because those are the domains I know well enough to evaluate what the AI produces. I do not use it outside those areas.

The standard I hold a plan to: tasks are clear and actionable, timelines are realistic, the student's identity is coherent, learning progression is logical, competition and activity content is consistent, and there is a chain of specific expected outcomes with fallback options at each stage. If the plan you currently have does not meet these criteria, it may be worth revisiting.