The AI that argues with your students.
Before a student drafts a word, PKL challenges their thesis, tests their evidence, and makes them answer the strongest counterargument. You see the reasoning, stage by stage. Not detection. Not surveillance. Visible thinking.
The Problem
You can’t see how your students are thinking. AI is making the problem worse.
Black box reasoning
A student hands you a five-paragraph essay. You can grade the result, but you can’t tell whether they understood the book, or just relied on AI to understand it for them.
Detection fails
Half your class is using AI. Detection tools don’t catch it — the surface patterns they flag also describe a developing writer, and even OpenAI shut down its own classifier.
Sources: Liang et al. (Stanford, July 2023); Vanderbilt disabled Turnitin’s AI detector, August 2023; OpenAI classifier shutdown, same month.
The coaching you’d give, if you could
Every English teacher knows what good writing instruction looks like: a one-on-one conversation with each student about their thesis, evidence, and interpretive move. You’ve never had the time.
What works instead
So we don’t detect — we make the writing process visible. PKL records each step a student takes through the essay: thesis, evidence selection, outline, draft, revisions. You can see whether they thought through the problem or arrived at a finished essay with no work shown. See how ↓
What makes PKL different
PKL doesn’t help students write the essay. It makes them defend it first.
Inside the Evidence stage, PKL opens a five-round Socratic exchange. It pushes back on the student’s locked thesis from four angles — power, perspective, plain reading, counterclaim — and the student has to answer with reasoning, not a rewrite. The exchange produces a transcript teachers can toggle on or off per assignment.
Frankenstein · student thesis under pressure
"Victor is the real monster because his abandonment turns a being capable of compassion into a vengeful destroyer."
PKL
01
Plain readingRead the timeline plainly. Victor flees within hours of animating the creature. On the text alone, how do you know the creature was capable of compassion before the abandonment happened?
Student
02
The creature later learns compassion from the De Lacey family. That capacity had to come from somewhere.
PKL
03
CounterclaimA skeptical reader could argue Shelley shows the opposite: the creature learns compassion by imitating humans, which means capacity was acquired, not innate. Under that reading, the abandonment did not corrupt an existing capacity; it just delayed a learned one. What in the text rules that reading out?
Student
04
Fair. I need a version of the claim that does not depend on innate compassion. Maybe the argument is about what Victor owed the creature as its maker, not about what the creature already was.
What teachers get
37.5 out of 40. No comments. When he asked what to fix, the teacher couldn’t say.
A high school junior we know spent weeks working on the two biggest essays of the year, one in English, one in History. He got an A– on each. Both came back with a numerical score and nothing else. When he asked his teachers what he could have done better, neither could say.
PKL’s answer
PKL doesn’t write your feedback. You do. What PKL takes off your plate is the part of grading that doesn’t need a human reader:
- ·Rubric scoring against your rubric. AI proposes a score per criterion; you click to confirm or override.
- ·A short summary of how the student got there: time on task, what they revised, where they got stuck. You don’t have to compare five drafts to see who actually engaged.
- ·Per-paragraph diagnostics the student already worked through during drafting. You see what they saw.
- ·Issue-spotting on the argument layer: where a claim doesn’t hold, where a quote got paraphrased instead of analyzed, where a topic sentence drifted. PKL flags the spans so your eye lands on the right places first. You still write the comment.
That used to take your weekend. PKL does it in the time the student finishes writing. The hours you get back go where they belong — to the specific, sentence-level feedback only you can give.
What a per-paragraph diagnostic looks like
Student paragraph: Frankenstein, body 2
“Victor abandons the Creature right after he creates it. The Creature says ‘I ought to be thy Adam, but I am rather the fallen angel.’ This shows that Victor is a bad father because he doesn’t take care of his creation. The Creature is sad and wants to be loved but Victor won’t love him.”
PKL Diagnostic
“Victor is a bad father” is a moral judgment, not a literary claim. What is Shelley showing us about creation, responsibility, or abandonment?
The “fallen angel” quote is the right pull and is woven into the sentence. Strong evidence selection.
The student paraphrases the quote but doesn’t analyze why Shelley chose “fallen angel” specifically, a religious image with weight the student is missing.
“The Creature is sad and wants to be loved” restates the obvious. What does Shelley want the reader to feel about Victor here? About themselves?
How PKL is different
Why your existing tools aren’t solving this.
The category difference is not another feature column. It is the moment at which the tool enters the student’s work, and the kind of intellectual labor it asks the student to perform.
PKL tests the argument itself.
Before the student writes a word of a draft, it presses the student to narrow a claim, confront the strongest objection, distinguish evidence from assertion, and leave a visible record of the reasoning that came before the prose.
How it works
Three steps to get a class running. Free for beta teachers.
Create a class. Share the code.
Pick a book, write a prompt, set a due date. Students join with a 6-character class code. Canvas LTI and Google Classroom integration are on the roadmap.
Students write in PKL through a structured workflow.
Thesis → Evidence → Outline → Draft. PKL coaches each step. When a student pastes a quote, PKL retrieves the passage from the indexed book and verifies it. No fabrication, no hallucination. Inside Evidence, students opt into the five-round Socratic exchange described above.
You see the reasoning, not just the result.
Open your class dashboard to see each student’s thesis, revisions, and diagnostic history. Grade in PKL or export the data to your gradebook. The hours you used to spend writing margin comments go back to your weekend.
Fits your teaching
Adapts to your rubric, your writers, your books.
Your rubric, your framework
Your framework
Pick a preset (Iowa CEEA, with AP Rhetorical support coming) or build your own. Rename any axis to match your vocabulary. If your students learn “topic sentence” instead of “claim,” that’s what PKL calls it too.
Your rubric
Upload the grading rubric you already use. PKL maps its diagnostics to your categories and weights, so what students see matches what you’ll grade them on.
Your exemplars
Upload two or three essays you consider strong student work. PKL calibrates its expectations to your standards, not a generic “good essay.”
PKL doesn’t replace your judgment. It extends it across 30 students at once.
Real writers, model essays
Feedback grounded in real writers, never generated advice. Once a student has internalized the framework, PKL points them to the writers who handle the same craft move with more skill.
Joan Didion, James Baldwin, Toni Morrison, George Orwell, Virginia Woolf, David Foster Wallace, Norman Maclean, Anton Chekhov, Benjamin Franklin.
PKL also pulls from model student essays — annotated paragraphs from past student work that show the same craft move at a teenager’s level. Students see what’s realistic at their level, not only what mastery looks like. Nothing is AI-generated; every passage is attributed and verified.
38 books indexed
Fully indexed novels, plays, and short stories. Most other AI essay tools have zero.
When a student pastes a quote, PKL retrieves the passage from the source text and verifies it. No fabricated quotes, no generic literary advice. The AI is grounded in the book the student is actually reading.
The Great Gatsby
Hamlet
Macbeth
Romeo and Juliet
Othello
Frankenstein
Pride and Prejudice
Sense and Sensibility
Emma
Jane Eyre
Wuthering Heights
Great Expectations
A Tale of Two Cities
The Scarlet Letter
The Adventures of Huckleberry Finn
Heart of Darkness
The Awakening
The Picture of Dorian Gray
Dr. Jekyll and Mr. Hyde
The Yellow Wallpaper
Crime and Punishment
Candide
The Odyssey
Beowulf
Ethan Frome
A Doll's House
The Importance of Being Earnest
The Time Machine
Frederick Douglass: Narrative
Quicksand
The Necklace
The Tell-Tale Heart
The Cask of Amontillado
The Gift of the Magi
The Story of an Hour
Bartleby, the Scrivener
Sweat
The Lady with the Dog
Don’t see your book? Request it from inside PKL — we prioritize new titles by what beta teachers ask for.
For department chairs and academic leaders
The answer to AI cheating isn’t detection. It’s dialectic.
Detection tools misfire, generate false accusations, and tell you nothing about what a student can actually do. Monitoring tools confirm a student followed a process. Neither one answers the question your teachers are actually asking: can this student think?
PKL answers it. When a student has to defend a thesis against structured pushback, respond to counterexamples, and sharpen evidence before drafting, the record of that exchange is something no shortcut can fake and no detector needs to police. The thinking becomes the deliverable. Your teachers grade reasoning they can see.
Email Adam about a pilot →A full data processing agreement is available before the pilot begins. Privacy and compliance details, including our DPA, are below.
What a pilot looks like
- 01
A bounded academic setting
One or two sections, one semester, one defined group of texts and assignments.
- 02
Teacher judgment stays central
Faculty set the prompts, review the reasoning record, and decide what counts as a stronger argument.
- 03
A semester-scale review
At the end of the term, an outcomes memo written for your academic leadership and your board covers student engagement, diagnostic progression, and teacher time saved.
- 04
No cost during the pilot period
No IT integration required. Students join with a class code.
Pricing
Free during the teacher beta.
School and department pricing available after the beta. Per-student licensing starts around $5/year.
Privacy, at a glance
- FERPA-aligned; COPPA-aligned for students under 13
- SOPIPA-compliant (California AB 1584 available)
- No student writing ever trains any AI model
- SOC 2 Type II in progress (Supabase, Vercel, OpenAI audited)
Full privacy & compliance details(expand)
What we collect
- ·Student name and the essays they write
- ·Diagnostic scores from PKL
- ·Teacher comments and grades
- ·Class membership
What we don’t
- ·Sell or share student data, ever
- ·Train AI models on student writing
- ·Use student data for advertising
- ·Track students outside of PKL
FERPA-aligned. COPPA-aligned for students under 13. You can export or delete student data on request, anytime. Full data processing agreement available for school districts.
SOPIPA compliant. PKL meets California’s Student Online Personal Information Protection Act and will execute AB 1584 compliant agreements with California districts. Ask about the California Student Data Privacy Agreement.
Built on a writing-teacher rubric. Hand-tuned over the past year — 51 prose-craft issues across 7 essay genres, each with named-writer exemplars and detection logic. The depth is why student feedback reads like a writing teacher’s, not a chatbot’s.
AI privacy. Student writing is sent to OpenAI for diagnostic processing under OpenAI’s Data Processing Addendum with Zero Data Retention. No student writing is used to train any model: ours, OpenAI’s, or any third party’s.
SOC 2 Type II in progress. Currently in audit. Existing infrastructure is hosted on SOC 2 Type II audited providers (Supabase, Vercel, OpenAI).
Join the beta
Use PKL with one class this term. Free.
We’re working with a small group of English teachers this semester. If that sounds like you, send me a one-paragraph email about your class and I’ll get you set up within a week.
Email Adam to request beta access →Or write me directly at adam@thinkpkl.com. I read every message.