Teacher Lite
Getting started
Welcome to Teacher Lite. Here is how to get the most out of it.
Set expectations first
AI detection is probabilistic, not deterministic. The tool produces a signal and an explanation of what triggered it. It does not produce a verdict, and it should not be used as one.
This guide covers how scans work, how to read a result, what AI detection cannot tell you, and how to use Teacher Lite responsibly in a classroom workflow.
The basics
Run your first scan
Go to the homepage, paste the text you want to screen into the input box, and submit. You can also submit a URL and the tool will pull the readable text from the page. The result returns in a few seconds with a verdict, a confidence score, and per-passage explanations.
How scans count
Each text scan uses one of your 50 monthly credits. URL scans cost the same. There is no difference in price between a short paragraph and a full essay; one submission is one credit.
Your monthly limit
Your 50 scans reset on your billing date each month. Unused scans do not roll over. If you run out, you can upgrade to Teacher Pro for 200 scans per month or wait for the reset.
Reading a result
Every scan returns one of three verdicts. The verdict is a summary of the underlying score; the score itself is a probability, not a percentage of how much AI was used.
The text shows patterns consistent with human writing. Strong vocabulary variation, irregular sentence rhythm, personal voice. This is the result you would expect from a piece that has been written, revised, and re-revised by a person over time.
Patterns are mixed. Could be human writing that happens to be formal, AI text that has been edited, or a heavily-collaborated piece. Treat this as “needs another data point”, not as evidence of AI use. Compare against earlier work from the same student, or ask the student to walk you through their process.
The text shows patterns strongly consistent with AI generation. But remember: a high confidence score is not proof. It is a signal that warrants a conversation. The explanation pane will show you which passages triggered the score and why, which is more useful than the score itself when you sit down with a student.
The five things AI detection cannot tell you
This is the part teachers most need to understand before acting on a result. None of the following is something any detector can do reliably, including this one.
1. Which AI tool was used
A detector cannot tell you whether the text came from ChatGPT, Claude, Gemini, LLaMA, or any other model. The features that flag AI writing overlap heavily across the major systems. A “Likely AI” result tells you the patterns look machine-generated; it does not identify the source.
2. How much was AI-written, versus lightly edited
A student who pasted a prompt into ChatGPT and submitted the output unchanged will look very similar to a student who used AI for a rough draft and then rewrote half the sentences. The score cannot meaningfully separate those two cases. The per-passage view helps, but it cannot quantify how much human revision happened.
3. Whether every model is covered
New models ship monthly. Detection accuracy on each new model needs to be re-established, and there is always a gap between a model’s release and a detector’s ability to flag its output reliably. Smaller open-source models, in particular, often produce text that current detectors struggle with.
4. With 100% accuracy
False positives are real. Published research has shown several detectors flag legitimate writing by non-native English speakers at materially higher rates than native-speaker writing. Clinical academic prose, formulaic genres, and certain creative styles can also trigger detection. Any result should be read with this limitation in mind.
5. Whether academic misconduct occurred
A detector score is not proof of misconduct. It is one input. Universities and schools that have treated AI detection scores as proof have already had to walk that back. Use the result as a starting point for inquiry, not as the conclusion of one.
How to use Teacher Lite responsibly
Use scans as a starting point, not a verdict
A single “Likely AI” result is the moment to look more carefully, not the moment to make a decision. The score opens the question; it does not answer it.
Compare samples from the same student over time
Run multiple pieces from the same student across several weeks. A drastic style change between two assignments is a stronger signal than any single scan. Keep your own informal record of how a student normally writes; that baseline is more useful than any percentage the tool will give you.
Have a conversation before any formal process
If you suspect AI use, talk to the student before escalating. Ask them to walk you through their thinking, their sources, and how they arrived at specific passages. A student who wrote the piece will be able to do this; a student who pasted output will often struggle. That conversation is more reliable evidence than a score.
Include full context in any escalation
Downloadable evidence reports are coming soon. When you share a result with a department lead or head of year, include your own context: prior work from the student, what the scan flagged, what came out of your conversation. The score on its own is not a case.
Limitations published
Full detail on how detection works, what is reasonably reliable, and where it is unreliable or impossible is on the methodology page. It is worth reading before you act on a result for the first time.
Calibration against a public dataset of roughly 1,000 known passages is in progress. Accuracy figures will be published when that work is complete. Until then, results should be treated as probabilistic guidance, not measured precision.
Help and feedback
Found a false positive? Reply to your account email. We read every one and use it to improve the tool. A real example of a piece that was scored incorrectly is one of the most useful things you can send.
Questions about the tool, your subscription, or anything else? Email Paul directly at paul@searchintel.tech.
Ready to run a scan
The fastest way to learn the tool is to use it on a piece you already know the answer to. Paste something you wrote, then paste something an AI wrote, and see what the result says.
Run your first scan