The Research

New ways to assess
in an AI world.

I teach high school. I built these tools because the assignments I grew up giving stopped telling me what my students knew. This page is the why — with the citations, so you can check me.

Which side of the line?

There's a line. On one end: “AI taught me.” On the other: “AI did it for me.” Most kids who used AI last night are somewhere between, and the only job that matters is figuring out where. A detector can't tell you. The kid can — if you put them on the spot and ask them to show you.

“AI taught me” or “AI did it for me”? Ask the kid to show you. That's the whole method.

That's what every tool on this site does: it creates a moment where the student has to show what they actually know, with the AI out of the room. Everything below is the research that convinced me that's the only honest way left.

The short version

AI detectors aren't reliable enough to be evidence against a student.12 Generative AI writes passing answers to a lot of the assignments we hand out.57 And the people who study this for a living have landed on the same conclusion: rules and warnings don't fix an assessment — changing the task does.8

So I stopped trying to catch AI and changed what I ask for.

In class, in a locked window. In a quiz written from the student's own paper. In a test no two periods share. The proof of learning moves to a place AI can't reach, and nobody gets accused of anything.

AI detection isn't evidence

Most schools learned this the hard way in 2023, and nothing since has changed the picture.

  • Researchers tested 14 AI-text detectors, Turnitin's included. None reached 80% accuracy. All of them leaned toward calling AI text human, and a light paraphrase made them worse.1
  • Detectors flagged more than half of essays by non-native English speakers as AI-written — while scoring near-perfectly on essays by American eighth-graders.2
  • OpenAI shut down its own detector in July 2023, citing low accuracy.3 A month later Vanderbilt turned Turnitin's off over false-positive risk.4

A detector gives you a percentage. It doesn't tell you what a kid knows, and it's not something you can defend at a parent conference. We don't sell one.

What AI does to a normal assignment

  • Across ten subjects at seven universities, ChatGPT turned in passable work on many standard assessments — best on online quizzes — and struggled with oral, visual, and hands-on lab tasks.5 A 2024 rerun with newer tools found the gap had only widened.6
  • In a blind, real-world trial, researchers slipped fully AI-written answers into a university's actual unsupervised take-home exams. 94% went undetected. They outscored the real students.7
  • The assessment-research community's verdict: telling students “no AI on this one” is a discursive change and doesn't restore validity. Structural changes to the task do.8
Honesty note

These are university studies. Nobody has run the equivalent controlled trial on ninth-grade essays. I read them as a warning about how the assignments most of us grew up giving now behave — not as a verdict on your classroom.

Two frameworks worth knowing

You don't need to read the papers. These two ideas are enough to redesign a unit.

University of Sydney · Liu & Bridgeman, 20239

The Two-Lane Approach

Lane 1 is secured: in-class work, live demonstrations, exams — the moments where you have to know it's the student. Lane 2 is open: AI use is expected and taught. The authors' blunt line is that there's no viable middle ground. Every RCNR assessment tool is a Lane 1 tool.

Perkins, Furze, Roe & MacVaugh, 2024 · revised 202510

The AI Assessment Scale (AIAS)

Five levels of permitted AI use on any task, from No AI to AI Exploration, now used in hundreds of institutions. It's a communication tool — it makes the expectation on each assignment explicit. Its own authors are clear that relabelling an unchanged task doesn't make it valid. The task has to change too.

Teaching centres that have worked through this land on the same short list of designs that hold up:1112

  • Secured, in-class work
  • Oral or live demonstration
  • Staged drafts with a visible process
  • Personal, local, or reflective prompts

How that turned into four tools

I built for the Lane 1 problem, because that's where teachers were stuck. Each one is a design from the list above, made cheap and fast enough for a single teacher to run on a Tuesday.

  • Digital Blue Book — secured, in-class writing. A locked page: no tabs, no paste, no other apps. Lane 1 for essays. bluebook.rcnr.net
  • ProveIt — authorship, verified from the work itself. Paste the student's paper; it writes a short quiz about that paper — the argument, the sources, the sentence on page two — taken in a locked window. The oral defense, made scalable. proveit.rcnr.net
  • TestHub — paper tests in randomized versions so no two periods share a page, lockdown entry, and AI-assisted essay grading you review. testhub.rcnr.net
  • Draft Coach — drafts that show their work. Students write with an AI practice partner and you see the revisions, not just the final file. Lane 2, on purpose. draftcoach.rcnr.net
None of these tells you whether a student used AI last night. They tell you what the student can do in front of you — which was always the question.
Because these tools touch students
  • No student accounts, no student emails. Students type a display name and their work.
  • Student work is encrypted and auto-deleted on a schedule (30 days by default).
  • Zero-data-retention agreement with our AI provider. Student names never go to the API. Nothing trains a model.
  • Built to align with FERPA and COPPA: no directory info collected, DPA available for districts.

The long version: Privacy Policy · Security & Privacy

Every tool, one flat price, no IT ticket.
Try it on one assignment.

Start 30-Day Free Trial
References
  1. Weber-Wulff, D., et al. (2023). Testing of detection tools for AI-generated text. International Journal for Educational Integrity, 19:26. https://doi.org/10.1007/s40979-023-00146-z
  2. Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(7). https://doi.org/10.1016/j.patter.2023.100779
  3. OpenAI (2023). New AI classifier for indicating AI-written text — 20 July 2023 update announcing the classifier's withdrawal due to low accuracy. https://openai.com/index/new-ai-classifier-for-indicating-ai-written-text/
  4. Vanderbilt University (2023). Guidance on AI detection and why we're disabling Turnitin's AI detector. https://www.vanderbilt.edu/brightspace/2023/08/16/guidance-on-ai-detection-and-why-were-disabling-turnitins-ai-detector/
  5. Nikolic, S., et al. (2023). ChatGPT versus engineering education assessment: a multidisciplinary and multi-institutional benchmarking and analysis. European Journal of Engineering Education, 48(4). https://doi.org/10.1080/03043797.2023.2213169
  6. Nikolic, S., et al. (2024). ChatGPT, Copilot, Gemini, SciSpace and Wolfram versus higher education assessments. Australasian Journal of Engineering Education, 29(2). https://doi.org/10.1080/22054952.2024.2372154
  7. Scarfe, P., Watcham, K., Clarke, A., & Roesch, E. (2024). A real-world test of artificial intelligence infiltration of a university examinations system: A "Turing Test" case study. PLoS ONE, 19(6). https://doi.org/10.1371/journal.pone.0305354
  8. Corbin, T., Dawson, P., & Liu, D. (2025). Talk is cheap: why structural assessment changes are needed for a time of GenAI. Assessment & Evaluation in Higher Education, 50(7). https://doi.org/10.1080/02602938.2025.2503964
  9. Liu, D., & Bridgeman, A. (2023). What to do about assessments if we can't out-design or out-run AI? Teaching@Sydney, University of Sydney. https://educational-innovation.sydney.edu.au/teaching@sydney/what-to-do-about-assessments-if-we-cant-out-design-or-out-run-ai/
  10. Perkins, M., Furze, L., Roe, J., & MacVaugh, J. (2024). The Artificial Intelligence Assessment Scale (AIAS). Journal of University Teaching and Learning Practice, 21(6). Revised: Perkins, M., Roe, J., & Furze, L. (2025). Reimagining the AIAS. JUTLP, 22(7). https://doi.org/10.53761/rrm4y757
  11. Lodge, J. M., Howard, S., Bearman, M., & Dawson, P. (2023). Assessment reform for the age of artificial intelligence. TEQSA, Australia. https://www.teqsa.gov.au/sites/default/files/2023-09/assessment-reform-age-artificial-intelligence-discussion-paper.pdf
  12. Yale Poorvu Center for Teaching and Learning, AI-Resilient Assessment; Duke Learning Innovation, AI and Assignment Design. https://poorvucenter.yale.edu/teaching/teaching-resource-library/ai-guidance-for-teachers/ai-course-assignment-design/resilient

Last reviewed August 2026. If a source has moved or been superseded, email [email protected] and I'll fix it.