How the Alerts System Works

One benefit of student reflection is that students often use the space to share important things that are impacting their learning and development that educators would not otherwise know. Sometimes, students write reflections that may need an intervention outside of Sown To Grow. There are several lines of defense to catch these reflections, STG’s alerts system is designed to serve as an additional layer of protection for student well-being. Like any other tool, it has its limitations (more on that below), but it is a powerful tool that adds meaningful safeguards to help educators identify students who may need support.

How Alerts Are Triggered

The system sends alerts in three ways:

• Automatic alerts: Sent immediately when the algorithm’s confidence score is high or there is explicit language indicating self-harm or other obvious concerns.
• Reviewed alerts: When the system flags content as potentially concerning but with lower confidence, an expert reviews it before any alert is sent.
• Teacher flagged alerts: Teachers can manually flag reflections directly in the platform if they find them concerning. This ensures important context (such as a student’s history, tone, or external circumstances) can be taken into account, even if the system does not automatically detect the reflection as concerning

Most alerts are sent within an hour of a reflection being written. If a reflection is edited, it goes through the system again and may trigger a new alert. For reflections submitted outside school hours, the automatic alerts are sent right away but the reviewed alerts are sent the following business morning.

Alerts are sent via email to the designated admins and teachers. The email includes a link that takes educators directly to the app, where they can view the reflection and take steps to support the student. Once someone from the team has reviewed the reflection and taken appropriate action, admins are recommended to mark the reflection as “Addressed” so other admins know the student has received support. Primary administrators are authorized to enable or disable who receives urgent alerts.

Admins can also log in to the app to access the full alerts portal, while teachers can log in to see all student reflections and use flagged filters to find the ones they need to review.

What the System Considers

A number of factors determine whether a reflection warrants an alert, including:

• Language that suggests self-harm, depression, or loss
• Language that suggests elevated stress or anxiety
• Other contextual signals within a reflection such as a combination of seemingly mild words, repetitive patterns (e.g., ‘alone, alone, alone’) , or strong emphasis on a seemingly minor experience (e.g., ‘they ALWAYS ignore me’) can make a reflection more concerning, even if individual words alone wouldn’t trigger an alert.
• The emoji a student selects alongside their reflection

This means that certain reflections, such as those mentioning serious topics but paired with “neutral” or “happy” emojis, may not trigger alerts. While this helps reduce false positives, it also means the system can sometimes miss important signals.

How Often Reflections Are Flagged

• Each week, the system processes hundreds of thousands of student reflections.
• Of these, about 2–3% are flagged as potentially concerning by the algorithm.
• After expert review, fewer than 1% of total reflections lead to alerts being sent to administrators.

This layered process helps avoid overwhelming schools with false positives while still surfacing the most serious concerns.

Strengths of the System

• Timely alerts: Serious concerns are surfaced quickly, most commonly within an hour.
• Balanced design: We work hard to strike the right balance - catching the most serious concerns without overwhelming you with notifications. Out of hundreds of thousands of reflections each week, fewer than 1% trigger alerts, so the ones you receive are likely the most concerning ones. 
• Admin dashboards: Even when alerts are not triggered, the Emotional Well-Being dashboard highlights patterns such as students who repeatedly select “not so good,” “awful,” or other lowest-tier emojis. This allows educators to proactively spot trends and reach out before concerns escalate.
• Multiple safeguards: Teachers can flag reflections directly if they notice something troubling, adding another line of defense.
• Continuous learning: The system learns over time when administrators mark unconcerning alerts as “non-critical” (or provide feedback) and when and as teachers flag reflections that were missed by the algorithm but are concerning - helping make the alerts system more accurate. 

Limitations to Keep in Mind

• Language complexity: Words or phrases that are harmless in isolation (“I don’t want to live in California”) can be deeply concerning in other contexts (“I don’t want to live anymore”). The system sometimes struggles to distinguish between these.
• Spelling and phrasing: Typos, slang, or unconventional phrasing can make it harder for algorithms to detect intent (“my uncul died”, “iam depresed”). Often, educators would reach out to us informing us about certain slangs (such as “kms”) that we have over time trained our system to catch.  
• Subjectivity: Emotional expression is nuanced, and some situations may be more concerning for certain students depending on their individual circumstances. Because educators and administrators are better equipped to understand each student’s context, we often err on the side of caution and rely on their judgment to evaluate (For eg “I hate school”, “family problems”).
• Not foolproof: While super powerful, like many tools, it can make mistakes - sometimes reflections can be missed (for eg “I want to disappear into my book”) or flagged incorrectly (for eg “My favorite chapter from the book was when her dad died”). That’s why multiple layers of support and human judgment remain essential.

 

A Layered Approach to Safety

This system is meant to supplement - not replace - the vigilance of educators. Teachers and administrators remain the first and most critical line of defense, by:

• Teachers reading and responding to student reflections every week
• The ability for teachers to flag alerts to administrators and counselors
• Admin dashboards that surface student reflections indicating they may be feeling down - visible in the Student Activity tab
• Following mandatory reporting requirements whenever a student is believed to be at risk of harm; the system does not replace educators’ legal responsibilities to act.

In Summary

The alerts system is an added line of defense, providing timely notifications to help identify at-risk students. While powerful, it has limitations, so human judgment and context remain essential. Each alert must be manually addressed, and you will be notified once per alert; if it’s not addressed, the notification count will continue to increase. Together with teachers reviewing reflections weekly, the ability to flag concerns directly, and dashboards highlighting students who may be struggling, these layers form a stronger safety net to support student well-being.

If you have additional questions regarding concerning reflection alerts, please reach out to our support team at [e-mail].