How Emeritus is Revolutionizing Assignment Grading Through AI
- The Opportunity: Why AI-Assisted Grading Matters
- Leveraging ChatGPT in Grading: The Early Experiments
- Understanding Grader Experience: Survey Insights
- Challenges Encountered in Early Phases
- Transition to Custom GPT Models for Complex Assignments
- Impact of Custom GPT Models: Higher Accuracy and Consistency
- AI for Feedback Quality Control (QC)
- Looking Ahead: The Future of AI-Enabled Grading at Emeritus
As a working professional, your time is limited—and when you invest in learning, you need feedback that is clear, timely, and genuinely useful. Whether you’re juggling deadlines, managing teams, or preparing for a career transition, the quality and speed of feedback can make the difference between truly mastering a concept and simply completing an assignment.
At Emeritus, learners submit more than 700,000 assignments each year across programs in AI, management, cybersecurity, and more. Every submission deserves feedback that helps you immediately understand what you did well, what needs improvement, and how to apply the learning on the job. To maintain this standard, our teams review 6,000+ pieces of personalized feedback monthly, ensuring clarity, accuracy, and alignment with learning goals.
As cohorts grow and professionals’ expectations evolve, we’ve integrated AI into our grading process to deliver the kind of feedback today’s adult learners need: faster, more consistent, more personalized, and more actionable. These improvements help you progress with confidence, stay engaged, and translate new skills into real-world impact—without slowing down your life or career.
AI and ML Courses
The Opportunity: Why AI-Assisted Grading Matters
Scaling grading while maintaining quality is a challenge faced by institutions worldwide. For Emeritus, adopting AI-supported systems allows us to align our assessment strategy with our broader mission. AI-enabled grading provides:
- Enhanced scalability: Supporting higher volumes of submissions without sacrificing quality.
- Improved consistency: AI assistance ensures standardized evaluation across complex cohorts.
- Faster feedback loops: Faster turnaround increases learner engagement and supports better learning outcomes.
- Empowered graders: By automating repetitive tasks, graders can focus on higher-impact evaluation.
These advantages position Emeritus as a leader in shaping the next wave of technology-supported education.
Leveraging ChatGPT in Grading: The Early Experiments
In 2023, Emeritus initiated a structured experiment using ChatGPT-3.5 to support grading activities across selected programs in Digital Marketing, People Analytics, Talent Management, Business Management, and Leadership.
Prompt Engineering and Structured Inputs
To enable consistent and high-quality outputs from the model, the team designed customized prompts for each assignment. These prompts included:
- Detailed assignment instructions
- Rubrics aligned to learning outcomes
- Solution sets or answer keys
- Sample submissions and feedback curated by instructional designers and subject matter experts
Because early versions of ChatGPT had limitations related to file uploads, word count, and accuracy, graders manually copy-pasted relevant content into ChatGPT and followed strict training on how to interpret and refine outputs.
Understanding Grader Experience: Survey Insights
To assess how ChatGPT performed across actual grading tasks, Emeritus conducted a survey across graders who collectively used prompts for more than 1,000 submissions. The survey captured perspectives on ease of use, speed, accuracy, consistency, and overall satisfaction.
Key results from the survey included:
- 70% found ChatGPT faster than manual grading
- 72% found the tool easy to use
- 65% rated it highly effective for grading tasks
- 68% observed strong consistency across regenerated responses
- 60% found the outputs to be highly accurate
- 75% reported high overall satisfaction
These findings showed that AI-assisted grading could meaningfully improve speed and consistency for simpler, text-based assignments.
Leadership Courses
Challenges Encountered in Early Phases
While the initial results were encouraging, several challenges emerged:
- Accuracy concerns: ChatGPT-3.5 performed well for simple text-based tasks but struggled with multi-criteria rubrics and assignments requiring deeper analytical judgment.
- Reliability: Some responses required multiple regenerations, and occasional platform issues disrupted workflow.
- Time investment: Preparing detailed prompts and manually entering assignment content reduced some of the time savings.
These limitations indicated that while early AI tools could supplement grading, more specialized models were required for complex tasks.
Transition to Custom GPT Models for Complex Assignments
To address complexity and improve accuracy, Emeritus developed custom GPT models using a licensed version of ChatGPT-4.0.
These models were designed after in-depth analysis of assignment structures and requirements across domains such as cybersecurity, finance, and technology. Unlike earlier prompt-based workflows, custom GPT models could handle:
- Longer and more detailed learner submissions
- File attachments
- Multi-dimensional, domain-specific rubrics
- Nuanced evaluation criteria
Following comprehensive pilot testing, a domain-expert grader was trained to use these models, resulting in measurable improvements.
Impact of Custom GPT Models: Higher Accuracy and Consistency
A survey conducted after the pilot revealed significant benefits:
- 86% of assignments rated the model as highly effective in supporting grading
- 71% found outputs highly accurate and aligned with human grader scores
- Speed improved, with 29% of graders reporting faster completion
- 71% observed highly consistent results across regenerations
- Overall satisfaction reached 86%
These results demonstrated that custom-built models could support grader judgment in a reliable, scalable way, particularly for complex assessments.
AI for Feedback Quality Control (QC)
As grading volume increased, ensuring consistent, high-quality feedback became a growing challenge. Initially, QC reviewers spent 90–100 minutes per grader each month reviewing feedback for three assignments. Due to workload, sample size had to be reduced to two assignments per grader.
To support QC at scale, Emeritus developed a custom GPT model in July 2024 to evaluate grader feedback. This model was trained on detailed QC rubrics that assessed clarity, tone, personalization, and practicality.
Improvements After AI Integration
- QC time reduced to approximately 75 minutes, even after returning to a three-assignment sample size
- The model analyzed feedback more efficiently, enabling increased coverage without increased effort
- Continuous refinement based on reviewer feedback helped maintain accuracy and fidelity
The QC process is now faster, more scalable, and more consistent—helping graders continuously improve the quality of feedback provided to learners.
Looking Ahead: The Future of AI-Enabled Grading at Emeritus
Building on these early successes, Emeritus is pilot-testing several new AI-based innovations:
- An AI Tutor that gives learners instant, preliminary feedback before final submission
- Dynamic, chatbot-based assignments that test learner knowledge through interactive conversation
- AI-generated assessment for late submissions, supported by a disclaimer and optional human review
These tools aim to make learning more adaptive, personalized, and accessible.
AI-enabled grading at Emeritus is designed to give you what matters most as a working professional: faster, clearer, and more consistent feedback. By combining human expertise with AI support, we help you quickly understand how to improve and apply what you learn—without slowing down your work or personal commitments.
The takeaway: you get timely, actionable feedback that keeps you progressing with confidence and helps you see real impact in your day-to-day role sooner.
This article was collaboratively researched and written by Akshay Vyavahare, Assumpta Fernandes, Pallavi Mhatre, Lara Mirgh, Vishnuvardhan TP, Suchitra Thingalaya, and Glen Mohr.
