HomeAI Detector › For Teachers

AI Detector for teachers, built around the conversation, not the verdict.

Bulk-scan a whole class in minutes, get sentence-level evidence for honor-code talks in about thirty seconds, and a defensible PDF report you can hand to a department reviewer. Calibrated for ESL writing so structured English does not get over-flagged. FERPA-aware, GDPR-aware, and student text is never used to train the classifier. Free to try. No card.

Start free, no card See pricing
Education discount on Pro FERPA-aware Sentence-level highlights
Who it is for

Built for the classroom, not the lab.

For K-12, middle and high school, and undergraduate teachers grading mixed-ability classes under time pressure with a department policy that now asks for evidence rather than instinct.

You are not trying to catch every AI essay. You are trying to spend less time wondering and more time teaching. TextSight is built around that priority: short scan, clear evidence, easy export, and a tone that respects student context.

Middle and high school English teachers

Five-paragraph essays, in-class writing, take-home assignments. The free tier covers casual sanity-checks on a single suspect paper. Pro at $19.99 a month (or $14.99 a month on yearly) covers the steady weekly load and adds the 90-day history that matters when a grade gets contested two weeks later.

Undergraduate instructors and TAs

Larger classes, longer essays, more pressure for documentation. Bulk upload of a class folder is the unlock here. Pro handles individual teachers. Business at $39.99 a month with 5 seats is the right fit for a department coordinating across instructors and TAs.

Department coordinators and writing-program directors

You need consistent evidence formatting across multiple teachers, shared history for moderation panels, and an audit log when a case escalates. The Business tier shares scan history across a workspace and produces the one-page PDF format reviewers want to see.

Fits beside Turnitin

A pre-Turnitin workflow that finishes before grading does.

TextSight is not a Turnitin replacement. It is the same-class scan that runs while you grade, so you have sentence-level context before the official institutional report lands a day or two later.

Before grading begins

Bulk-upload the class folder of PDF or DOCX submissions. TextSight returns a class dashboard in a few minutes with Authenticity Scores per essay. Anything below 50 is surfaced for a longer look. Pro and above.

During grading

You see the Authenticity Score next to the student name. Above 75, grade normally. Between 50 and 75, glance at the sentence highlights and factor them into your contextual read. Below 50, the essay gets a slower review with the full report open.

After grading

One-click PDF export per scan when you need a record. The PDF carries the student text, the score, the sentence flags, the timestamp, and the classifier version. This is the format department reviewers want, not a screenshot of a percentage.

Compared to reading every essay cold, suspecting AI in some, and escalating without evidence, the scan-then-grade pattern usually costs 30 to 45 minutes for a class of 25 essays and saves the painful escalations.

Plans & pricing

Pricing for teachers and departments.

Verified institutional emails get an education discount on Pro automatically at signup. Department licensing through Business. Full details on the pricing page.

Free
$0/forever

 

Sanity-check a flagged paper. No card, no email.
  • 3 scans / day
  • 5,000 chars per scan
  • Sentence-level highlights
  • Authenticity Score
Start free
Starter
$7.49/month

Billed $89.88/year — Save $30

For a teacher checking one class a week.
  • 20 scans / day
  • 20,000 AI rewriter words/mo
  • Chrome extension
  • Email support
Get Starter
Business
$29.99/month

Billed $359.88/year — Save $120

For departments and writing programs.
  • 100,000 AI rewriter words/mo
  • 5 team seats, shared history
  • Audit log, REST API
  • White-label PDFs
Get Business

Education discount applies automatically when you sign up with a verified institutional email. View full pricing →

What you see in a scan

Sentence-level evidence, not a single suspicious number.

A percentage by itself is not a basis for a conversation. The TextSight result panel shows where the classifier reacted and why, so you can read the evidence the same way the student can.

Sentence highlights

Every sentence is colour-coded by its individual AI-likeness score. Red sentences clustered in one paragraph are a stronger AI signal than scattered yellows. Scattered yellows in otherwise structured prose often mean the student writes formally and is not using AI. You read the pattern, not just the headline.

Paragraph cards

Each paragraph is rolled up into a card showing its own score and the dominant signal driving it. Useful when you want to ask the student about a specific section without scrolling through highlights.

Perplexity and burstiness signals

The underlying signals the classifier weighs are surfaced read-only on Pro. Low perplexity plus low burstiness across the whole essay is the classic AI fingerprint. Mixed perplexity with normal burstiness is the human pattern. These are diagnostic context, not verdicts.

Authenticity Score and Plagiarism Risk

Two scores side by side. Authenticity Score is the inverse-AI reading. Plagiarism Risk is a separate signal that catches copied passages. A clean essay scores high on authenticity and low on plagiarism risk. The two together give a fuller picture than either one alone.

Fits with the tools you already use

Integrations with your existing classroom stack.

Native LMS plugins are not shipped yet. Here is the honest 2026 picture of what works today and what is coming.

Today: export and upload

Export student submissions as PDF or DOCX from Canvas, Blackboard, Brightspace, Google Classroom, Schoology, or Microsoft Teams. Drag the folder into TextSight bulk upload. Get the class dashboard back in a few minutes. Copy scores back into your gradebook, or save the PDF report for your records.

Today: paste-in workflow

For single essays, paste the text into app.textsight.ai. The free tier covers casual one-off scans up to 5,000 characters. Pro extends per-scan length and unlocks bulk upload.

Today: Chrome extension

One-click scan from any web page including hosted submission viewers. Useful when you are reading inline rather than downloading. Available on Starter and above.

Roadmap: native plugins

Canvas, Google Classroom, Blackboard, Brightspace, Schoology, and Microsoft Teams plugins are on the roadmap. We are not promising dates while the integration partners change their plugin requirements; we will not ship a thin wrapper that breaks every term.

Student data protection

Privacy-first by default, FERPA-aware and GDPR-aware.

Student submissions are protected by FERPA in the US, by GDPR in the EU and the UK, and by local equivalents elsewhere. TextSight is designed to honour those rules out of the box.

No training on student work

Student text submitted for scanning is never used to train the classifier or any other model. This is a contract clause, not a setting you have to find.

Deletion on request

Any scan can be deleted from history. On Pro you can delete individual records. On Business, deletion can be applied across a workspace by an admin in a single action.

Standard DPA for institutions

Business and Enterprise tiers ship with a standard Data Processing Agreement. Larger institutions with custom DPA needs are handled via the contact form, usually inside a week.

Honest scope on residency

Hosting is on Hetzner in Germany for ML inference and on DigitalOcean for the API. EU institutions get EU residency by default. For US institutions that require US-only residency under FERPA, the Enterprise tier is where that is contractually scoped.

Calibration tool, not verdict

A detector is a conversation starter.

Treating any single number as proof a student used AI is unfair and unreliable. The 2026 expectation is that teachers use detection as one input among several, and the design here is built around that expectation.

Use the score as a signal, not a sentence

A low Authenticity Score means the essay reads more AI-like to the classifier. It does not mean the student used AI. False positives are real, particularly for ESL writers, formally-taught Oxford-style English, and standard topics where phrasing overlaps with AI defaults. Your contextual judgment is the final layer, not the classifier.

Do not auto-fail on a score

Department policies that auto-fail on a single detector percentage produce bad outcomes and have already led to lawsuits in the US and UK. The defensible path is conversation first, evidence second, decision third, with the scan as supporting context throughout.

Conversation starters, not accusations

The sentence-level highlights are the lever. Ask the student to walk you through how they wrote a specific flagged paragraph. Ask for earlier drafts or research notes. The flags give the student something specific to respond to, which is fairer than a vague accusation based on a percentage.

Design-side defences matter more than detector arms races

Deeply-rewritten AI text is hard for any detector to catch reliably. The durable defences are assignment design: in-class drafting, multi-stage submissions with conferences, prompts that require personal experience or course-specific knowledge. The detector is one signal in that ecosystem, not the whole defence.

FAQ

Teachers frequently ask.

Can a teacher use TextSight if the school already runs Turnitin?
Yes, and many do. Turnitin's institutional report often lands a day or two after submission, and the report does not always show sentence-level highlights. TextSight gives teachers a same-class scan with sentence-level evidence in about thirty seconds, useful for deciding whether a paper needs a longer look before grades go in.
How do I bring up AI suspicion with a student without false-accusing them?
Use the sentence-level highlights as conversation starters rather than verdicts. Ask the student to walk you through how they wrote a specific flagged paragraph. The report shows which lines triggered flags, which gives the student a fair chance to explain, share earlier drafts, or acknowledge. A percentage on its own is not a basis for a conversation.
What about false positives for ESL or multilingual students?
False positives on non-native English writing are real and well-documented. Detectors over-flag structured English from non-native writers at several times the rate of native US writers. Before treating any flag as evidence, look at whether the student is an ESL writer, whether the topic uses standard phrasing that overlaps with AI defaults, and whether the flagged sentences cluster or scatter. Scattered flags in structured prose often reflect false positives, not actual AI.
Is there an education discount?
Verified teachers signing up with an institutional email such as .edu, .ac.in, .ac.uk, or .edu.au get Pro at a reduced rate. For department-wide or school-wide licensing the Business plan at $39.99 a month includes 5 seats with bulk upload, team workspaces, and an audit log. Larger institutional contracts are quoted via the contact form.
Does TextSight integrate with Canvas, Google Classroom, or Schoology?
Native LMS plugins are not shipped yet. The honest workflow today is to bulk-upload PDF or DOCX exports from your LMS into TextSight, get the class dashboard back in a few minutes, and copy results back into your gradebook or save the PDF report. Canvas, Blackboard, Brightspace, Google Classroom, Schoology, and Microsoft Teams integrations are on the roadmap.
Is a TextSight scan defensible in an academic integrity hearing?
Reports are designed to support hearings, not replace teacher judgment. Each scan stores the input text, the Authenticity Score, the sentence-level flags, the date and time, and the classifier version used. Pro adds 90-day history with one-click PDF export. Treat the report as supporting evidence alongside the student conversation and earlier-draft review, not as a verdict on its own.
Will TextSight train on student work?
No. Student text submitted for scanning is never used to train the classifier or any other model. Data retention is bound to the user's history settings, deletion on request is supported, and a standard DPA is available on Business and Enterprise tiers.
What should I do with a borderline score in the 40-60 range?
Look at the highlighted sentences, not the percentage. A cluster of red sentences in one section often signals a student leaned on AI for that part. Scattered yellow flags through otherwise structured prose usually means the student writes formally and is not using AI. Borderline scores are exactly where teacher judgment matters most, and the tool is built to support that judgment rather than replace it.
Related

More for teachers and schools.

Scan a class. Talk to a student. Move on.

Free to try. No card. Education discount on Pro for verified institutional emails.

Start free, no card See pricing
FERPA-aware · GDPR-aware · Sentence-level evidence · 90-day audit history on Pro