Drawing Out a Student’s Own Words: An AI College Counselor for Ontos Admissions Insight
Ontos Admissions Insight brings a twenty-five-year college counseling method to schools as software. The AI’s job is to ask the next question, surface each student’s story and leave the words to them.
The work
The Challenge: The Essay Has to Belong to the Student
College admissions asks a seventeen-year-old to explain who they are, where they come from and what they might bring to a campus, in a few hundred words.
The students with the most support have someone to help them find that story. A good counselor draws it out in conversation: another question, a noticed theme, a push past a generic answer. That attention takes time, and a school counselor may be responsible for hundreds of students. Teachers face the same math when recommendation season arrives, with dozens of individualized letters to write in hours they do not have.
AI is the clear way to expand that capacity, and it comes with an obvious risk. Ask a generative AI tool for help with a college essay and it can simply write one. The result may be polished, but the most important thing about the essay is gone: it has to belong to the student.
Ontos has spent twenty-five years building a college counseling methodology on that belief. Admissions Insight is that method as software, for the schools where individual attention is hardest to provide.
The Strategy: An AI Counselor, Not an Author
We built Admissions Insight around a different role for AI: counselor, not author.
The platform uses AI to draw material out of students. It recognizes when an answer is vague, spots something worth exploring, asks a better follow-up question and helps a student see connections across their own experiences. The AI guides the conversation. The ideas and the words remain the student’s.
- The constraint is designed in and tested. Enforcing it took more than instructions to the model. The refusal to write is built into the product and tested as part of the system itself.
- Prompts are production infrastructure. The counseling methodology is encoded as versioned, observable prompts in PromptLayer, with review, regression testing and traceability, so the behavior can be reviewed and improved over time.
- OpenAI does the reasoning. The hard work is weighing what a student has shared, what is missing and what question would help them go deeper. That is what a frontier model is good at.
- Four audiences, one journey. Students, counselors, school leaders and families each get an appropriate view into the same underlying record. Much of what students share is deeply personal, and many of them are minors, so trust was a design requirement from the start.
The Outcomes: Individualized Admissions Support at Scale
Over time, the interactions add up to something more useful than a chatbot transcript: a portrait of the experiences, values and themes a student keeps returning to.
Students get help discovering what they actually want to say. Counselors see where each student is in the process, what has emerged through reflection and where their expertise will matter most. Teachers and counselors open recommendation season with richer, better-organized material about every student they advocate for. And schools can offer consistent, individualized admissions support across a far larger student population.
AI doesn’t become the student’s voice. It helps more students find their own.
The detail
Why it mattered
- A strong essay starts before the writing. Students need to find the experiences, values and moments that say something meaningful about who they are. A good counselor draws those out in conversation, and that attention takes time a counselor responsible for hundreds of students rarely has.
- The obvious use of AI is the wrong one. Ask a generative tool for help with a college essay and it can simply write one. The result may be polished, and the most important thing about the essay is lost: it needs to belong to the student.
- The users include minors. Students share family experiences, setbacks, aspirations and reflections they may not have put into words anywhere else. That information set requirements across the product, from the model’s role to who sees what.
- Four audiences, one record. Students, counselors, school leaders and families each need an appropriate view into the same underlying journey.
How it was built
- The AI reasons like a counselor. It weighs what a student has already shared, what might be missing and what question would help them go deeper. It can ask for a specific memory, press on an overly broad answer or return to a theme that keeps appearing.
- Prompts are production infrastructure. The counseling methodology is encoded as versioned, observable prompts, with review, regression testing and traceability, so behavior can be inspected and improved over time.
- The interactions add up to a portrait. The platform surfaces the experiences, values and themes a student keeps returning to. The student gets better material to write from, and the counselor gets a richer understanding of who they are advising.
- The method leads. The Ontos methodology remains the foundation. The software extends its reach.
Client leadership
Matthew O’Bryant
Matthew O’Bryant has led platform work for universities and education organizations, where a tool that writes for a student instead of with them fails on its own terms.