AI in Action for Educators

One of the most common questions educators ask right now is:

“Can AI tell if a student used AI?”

It seems like a simple question — but the answer is complicated.

Many schools have begun using AI detection tools to identify writing generated by systems like ChatGPT. However, current research shows these tools are far from perfect. Studies have found that AI detectors can both miss AI-generated writing and incorrectly flag human writing as AI.

This means educators must approach AI detection thoughtfully. Detection tools may provide clues, but they should never be the only evidence when evaluating student work.

Understanding the strengths and limits of these tools helps teachers:

• avoid false accusations

• design better assessments

• guide students toward responsible AI use

This lesson explores what AI detectors can do — and what they cannot.

In this lesson you’ll learn:

• How AI detection tools work

• Why they often produce unreliable results

• Real examples of AI detection used well — and used poorly

• What AI “humanizers” are — tools designed to disguise AI writing so it looks more human

• Strategies for designing assignments that discourage AI misuse

• Ways to talk with students about academic integrity in an AI world

The short answer is:

Sometimes — but not reliably enough to prove misconduct.

Research comparing multiple AI detection systems found average accuracy around 39.5%, and when simple editing or paraphrasing was applied, accuracy dropped to about 17%.

Other studies similarly conclude that AI detectors are not sufficiently reliable for high-stakes academic decisions.

Even when detection works well for raw AI text, accuracy drops significantly if a student edits or rewrites the content.

This is why many universities now advise faculty to treat AI detection scores as signals for further investigation — not proof of cheating.

Most detectors look for patterns such as:

Perplexity
AI writing often uses predictable word patterns. Perplexity measures the randomness or unpredictability of text; high perplexity indicates human writing (varied, unexpected), while low perplexity indicates AI writing (predictable, common word choices). Low perplexity means an AI model is not "surprised" by the word choices.

Burstiness
Human writing tends to vary sentence length more than AI writing. Burstiness refers to the variation in sentence structure and length, often characterized by the natural, rhythmic fluctuation between long, complex sentences and short, simple ones in human writing.

Statistical language patterns
Detectors compare writing to patterns seen in AI-generated text. Detectors look for overly smooth phrases, monotone text, and less varied or unexpected word choices.

However, these patterns are not unique to AI writing.

For example:

• strong writers can produce “predictable” sentences

• English language learners may write in patterns detectors flag

• editing or paraphrasing can hide AI fingerprints

This is why false positives occur.

Instead, they analyze patterns in the writing.

• Good writers often use predictable sentence structures.
• English learners may write in ways detectors misinterpret.
• AI text can be edited or paraphrased.
• Different detectors use different algorithms.

Research shows that when AI-generated text is edited or paraphrased, detection accuracy can drop dramatically, sometimes to around 17% accuracy in experimental conditions.

AI detectors can be helpful when they are used:

✔ as one piece of evidence
✔ as a conversation starter with students
✔ to identify patterns across multiple assignments
✔ alongside teacher knowledge of a student’s work

Problems arise when detectors are used:

✖ as proof of cheating
✖ without teacher review
✖ without giving students a chance to respond
✖ without understanding false positives

Research shows detectors sometimes misidentify human writing as AI-generated and fail to detect AI writing at all, particularly with students who write well.

Different AI detection tools use different algorithms and can produce very different results.

Common tools used in education include (but there are many, many options for these types of tools):

Turnitin AI Detection

• Integrated into many school plagiarism systems

• Often used by universities and institutions

• Can detect AI-generated sections within essays

Limitations:

• May miss edited AI writing

• Can produce false positives

GPTZero

• One of the earliest tools designed for educators

• Analyzes sentence patterns and predictability

Pros:

• Easy to use

• Free options available

Limitations:

• Can struggle with mixed human/AI writing

Originality.ai

• Often used by publishers and professional writers

• Combines plagiarism and AI detection

Pros:

• High accuracy in controlled tests

Limitations:

• Paid subscription

• Still vulnerable to edited AI text

One independent test comparing five detectors found overall accuracy ranging from about 82% to 90% depending on the tool.

Even in these tests:

• False positives occurred (human writing flagged as AI)
• False negatives occurred (AI writing missed)

This is why many experts recommend using detection tools only as indicators, not proof.

Just as schools began using tools to detect AI-generated writing, a new category of tools has emerged that attempts to hide AI writing rather than detect it.

These tools are often called AI bypassers or “humanizers.”

Their purpose is to take AI-generated text and rewrite it to sound more like human writing, making it harder for AI detection systems to identify.

According to a recent announcement from Turnitin, these tools are becoming increasingly common online and are sometimes marketed directly to students.

Humanizer tools typically attempt to:

• vary sentence length
• change vocabulary
• add minor grammatical irregularities
• restructure paragraphs

The goal is to make AI writing appear more like natural human writing patterns.

Because these tools are becoming more widely available, companies that provide academic integrity software are updating their systems.

Turnitin recently announced that its AI detection system now includes AI bypasser detection, designed to identify writing that may have been modified by humanizer tools after being generated by AI.

The system attempts to flag situations where:

• text was likely generated by AI,
• then altered by a rewriting tool designed to disguise it.

Like other AI detection technologies, this feature is intended to provide indicators rather than proof of misuse.

Instead of relying on detection, many educators are redesigning assignments.

Examples include:

Require:

• brainstorming notes

• outlines

• drafts

• reflections

This makes the thinking visible.

Ask students to connect learning to:

• personal experiences

• local issues

• class discussions

These are harder for AI to generate convincingly.

Short writing done during class helps establish a student’s authentic voice.

Teachers can compare this to later assignments.

Students explain their reasoning or thinking in short conferences.

If you choose to allow students to use AI tools for brainstorming, outlining, or editing, consider asking them to include their AI conversation as part of the assignment.

This creates transparency and helps teachers see how the tool supported the student’s thinking rather than replacing it.

Students can easily do this by:

• clicking the Share option in tools like ChatGPT and submitting the link
• copying and pasting the AI conversation into their document
• including screenshots of key prompts and responses

This allows teachers to see:

• what questions the student asked
• how the AI responded
• how the student used or modified the suggestions

Instead of asking “Did a student use AI?”, the conversation shifts to:

“How did the student use AI to support their thinking?”

This approach encourages responsible use, transparency, and academic honesty, while still allowing students to benefit from AI as a brainstorming or revision tool.

Ask students to complete an assignment and then answer:

• Did you use AI?

• If yes, how did you use it?

• What parts are entirely your own thinking?

This shifts the focus from policing AI to using it responsibly.

• Should AI detectors ever be used as proof of cheating?

• How can teachers balance trust and accountability?

• What types of assignments naturally discourage AI misuse?

• What expectations should schools communicate to students?

AI detection is still an evolving technology. I tested several AI detection tools using my own writing, and also by copying in text from other authors. I was surprised to find that these tools labeled 80% of that writing as written by AI, causing me to question even more how we can address this issue as educators.

Right now, the best safeguard for academic integrity is not an algorithm — it’s thoughtful teaching, strong relationships with students, and assignments that value real thinking.

As AI continues to become part of the classroom, it’s easy to get caught up in the tools themselves — the fancy images, perfect essays, or instant drafts. But the real focus remains on student thinking, intentional creativity, and meaningful learning.

Our job as educators isn’t to eliminate AI — it’s to guide students in using it thoughtfully, helping them understand when it supports their ideas, when it replaces thinking, and how to make their own choices with integrity.

The questions we ask, the conversations we have, and the reflections we encourage now will shape how students learn to create, innovate, and take responsibility in an AI-enhanced world.

— Laurie

Then an important question follows:

How are students actually using AI right now?

New research from Pew Research Center reveals something many educators may find surprising:

Most teenagers have already used AI tools — and many are using them for research, homework, and learning support.

In the next lesson, we’ll explore:

• what the latest research says about how teens are using AI
• the surprising ways middle school and high school students use AI differently
• emerging concerns about AI “humanizers” and bypass tools
• what students say they wish teachers understood about AI

Understanding how students are actually interacting with AI may be one of the most important steps schools can take as we think about the future of learning.

To help me make these lessons most useful (and to use my time wisely), I’d love your input:

How do you prefer to receive these AI lessons?

☐ Short written newsletters like this

☐ Short video walkthroughs

☐ A combination of both

You don’t need to reply unless you’d like to—this is just helpful feedback.

And finally, if you would like a short video or Google Meet walkthrough of anything we cover, please let me know. I’m happy to create videos or meet virtually with you when you'd like help.

You can view all of the AI in Action for Educators lessons on our district website through this link: https://www.parkrapids.k12.mn.us/departments/technology-homepage/ai