One of our customers — a course creator with several years of content available — told me something that stuck: "I'm selling a lot of courses, but a significant number of people never study them. I have students who've had courses for two or three years and everything I try isn't working." That's not a rare problem. Across 32,000+ courses on Ruzuku, we see it constantly: students who enroll with real enthusiasm and then go silent. No course platform proactively flags these students for you. But you can use ChatGPT and a spreadsheet to build your own early warning system.
What you’ll walk away with:
- A system identifying students showing disengagement signals
- Personalized outreach templates for common at-risk patterns
- A tracking workflow monitoring which interventions work
The silent student problem
Here's what the data tells us: first-week engagement is the strongest predictor of course completion. Students who don't engage in week one rarely catch up. Yet most course creators don't realize someone has gone quiet until the cohort is half over — if they notice at all.
One of our customers running a plant medicine program wanted to reach out to participants who'd been inactive for three or four weeks. The challenge? She had to manually dig through a CSV export to figure out who was behind. Another customer found that students were deleting their enrollment emails and never even opening the course. These aren't edge cases. They're the default experience for most online course creators.
Enterprise learning management systems have automated risk detection. Thinkific launched an AI teaching assistant in early 2026 that answers student questions, but even that doesn't proactively identify at-risk students. No creator platform — Teachable, Kajabi, Thinkific, Skool — flags disengaged students for you automatically. What I'm describing here is a manual workaround. It's not elegant. But it's the best tool available to independent course creators right now.
How to build your early warning system
Export your progress data
Export a CSV from your course platform that includes, at minimum: student name, email, enrollment date, lessons completed, and last active date. On Ruzuku, you can export this from your course's student progress page. If your platform doesn't include a "last active" column, use the last lesson completion date as a proxy.
Open the CSV in Google Sheets and do a quick scan. Remove any test enrollments or revoked students — I've seen creators get confused by students who were removed from a course but still appear in progress reports, polluting the data.
Ask ChatGPT to flag patterns
Copy your spreadsheet data (headers included) and paste it into ChatGPT. Use one of the prompts below to ask for a risk assessment. ChatGPT will categorize students by engagement level and suggest who needs attention first.
Draft personal check-ins
For each at-risk student, have ChatGPT draft a short, personal check-in message. Not a mass email — a message that acknowledges their specific situation. "I noticed you completed the first two modules but haven't been back in a few weeks" is specific enough to feel personal without being invasive.
Research by Engagement Mining (Jayaprakash et al., 2014) found that the sixth week is the optimal intervention point in semester-length courses — late enough that the pattern is real, early enough to make a difference. For shorter cohort courses, scale that proportionally. A 6-week course? Check in at week 2.
Prompts to try
Risk assessment from CSV data
"Here is student progress data from my online course [paste CSV data]. The course has [X] total lessons and started on [date]. Categorize each student into three groups: (1) On track — completed at least [X]% of lessons released so far, (2) Falling behind — completed some lessons but more than 2 weeks behind, (3) At risk — haven't completed anything in 3+ weeks or never started. Show the results in a table with columns: Name, Lessons Completed, Last Active, Status, Days Since Last Activity."
Personal check-in messages
"Based on the at-risk students you identified, draft a short personal check-in email for each one. Each email should: (1) mention them by name, (2) reference their specific progress (e.g., 'I see you completed the first two modules'), (3) express real care without guilt-tripping, (4) offer one specific, low-effort next step they could take. Tone: warm and supportive, like a teacher who noticed a student has been absent. Keep each email under 100 words."
Engagement trend analysis
"Look at this student progress data [paste CSV] and identify patterns: (1) What percentage of students are inactive? (2) Is there a specific lesson where students tend to drop off? (3) Are there students who were active initially but stopped — and when did they stop? Summarize the findings in plain language. I'm a course creator, not a data analyst."
The human layer
Be clear about what this approach can and can't do. It identifies who might need help. It doesn't tell you why they stopped, and the "why" matters enormously.
Sometimes students go quiet because life happened — a family emergency, a work deadline, a health issue. Sometimes they got stuck on a specific lesson and felt too embarrassed to ask for help. Sometimes they realized the course wasn't what they expected. Each of those requires a different response, and AI can't distinguish between them.
What you can do: send the check-in message, then listen. One of our customers told me she sent a simple "I noticed you've been away — everything okay?" email to a student who'd been silent for a month. The student replied within hours, explained she'd been overwhelmed, and ended up completing the entire course. The email took two minutes. The impact lasted.
The other caveat: this is a workaround. You're manually exporting data, pasting it into a chatbot, and acting on the results. It works, but it doesn't scale gracefully beyond 50-100 students per cohort. If you're running larger programs, you'll want to do this check weekly and focus on the highest-risk students first.
What it gets wrong
- False positives from incomplete data. A student at 94% completion who hasn't clicked the final checkbox looks "incomplete" in the data but may have finished everything meaningful. Always sanity-check before reaching out — you don't want to email someone who's actually done.
- Doesn't account for intentional browsers. Some students buy courses as reference material with no intention of completing every lesson sequentially. A "0% complete" student who's accessing specific lessons on demand isn't at risk — they're using the course differently than you expected.
- Check-in tone can backfire. ChatGPT's default check-in messages can sound slightly clinical or guilt-inducing. Edit heavily. "I missed seeing you in the discussions" is better than "I noticed your progress has stalled."
- No real-time alerting. This is a snapshot, not a monitoring system. By the time you export, analyze, and reach out, a few more days have passed. For cohort courses where timing matters, do this check at least weekly.
Related guides
- Tracking Student Progress in Google Sheets — set up the spreadsheet that feeds this workflow
- Recording Async Video Messages with Loom — sometimes a personal video beats an email
- Generating Discussion Prompts with AI — increase engagement before students go quiet
- Writing Student Emails with AI — draft other types of student communications
Don't let students disappear
The silent student problem isn't a platform problem — it's a visibility problem. Until course platforms build real-time risk detection, this manual approach is the best tool independent creators have. Start free on Ruzuku and build courses where you can see student progress and step in before anyone falls through the cracks.