AI-Assisted Feedback Using Gemini Gems
- Elvin Flores
- 3 hours ago
- 2 min read
Over the past year, I’ve facilitated more workshops and faculty consultations on AI than I can count. In almost every session, I hear the same two questions:
Will this actually save me time?
How can I realistically incorporate this into my teaching?
Beyond the impressive demos of Generative AI (GenAI) tools, faculty are looking for practical, workable solutions that can meaningfully support some aspect of their teaching. What I’ve come to see through my conversations with faculty is an opportunity to intentionally design feedback workflows using GenAI tools grounded in user-supplied documents, a technique often referred to as Retrieval-Augmented Generation (RAG).
Specifically, if we use course materials such as assignments, rubrics, and readings to anchor an AI tool for structured pre-submission feedback, while reserving all final evaluation for yourself, the feedback it generates can be tightly aligned to the course learning outcomes. The goal is not to outsource feedback, but to help students identify gaps before you ever see the draft. This can be done by uploading course resources to a custom “Pre-Submission Writing Coach” Gem [NYU log-in required].
Instead of offering generic comments, the tool can generate feedback using clearly defined criteria, maintain a consistent tone, and, if student-facing, avoid rewriting the student’s work. The structure matters because it keeps the tool operating within educational guardrails rather than replacing instructor judgment.
To create a “Pre-Submission Writing Coach,” begin by grounding an AI tool in your course learning outcomes, session/unit learning objectives, an assignment rubric, and an assignment prompt (i.e., uploading documents or linking a Google Drive folder to a Google Gem). When a student submits a draft, the tool can respond in targeted ways.
Instead of offering general advice, it might say:
“It looks like your thesis statement does not clearly address Criterion 3 (Historical Context) of the rubric. Review the background materials provided and consider how your argument situates the event within its broader context.”
The purpose is not to rewrite the paper. It is to point students back to the criteria and prompt revision aligned with assignment requirements.
Sequencing Within the Assignment
Sequencing also matters. Rather than using AI at every stage, treat it as one phase in the overall workflow. “Train” the Gem with sample essays, test tone and alignment, and apply it consistently within a specific assignment window. Establishing clear boundaries can help protect workload and expectations.
Below is an example of how this Gem might be implemented in an already existing assignment.
As more students experiment with AI in their own workflows, our role as educators is not to keep up with every feature. It is to model best practices. When we build constraints, define boundaries, and align tools with learning outcomes, we teach students how to move beyond surface-level use of AI and engage it in ways that deepen their thinking.
Want to learn more about using Gemini Gems?
See our AI resources below (NYU log-in required) or contact us at nexus@nyu.edu.
Teaching and Learning with AI (Guides)
Creating a Google Gem (Guide)
Introduction to Google Gemini (Workshop)