StyleMatch gives you eight metrics. Verify gives you what the AI can see that the metrics can't — internal consistency patterns across the whole essay, or a ten-dimension qualitative analysis that reads the writing itself, not just its statistics.
Verify uses AI to read the writing itself — not just count words and measure sentences. Both report types draw on the same underlying stylometric signals as StyleMatch, but layer qualitative analysis on top: what the writing actually says about how it was produced, who produced it, and whether the two samples sound like the same person.
The Consistency Report works on a single submitted document — no controlled sample required. The AI chunks the essay into ~150-word segments and scores each one on six dimensions: Register, Vocabulary, Sentence Complexity, Argument Depth, Error Density, and Cohesion.
The Baseline Deviation Report goes beyond StyleMatch's statistical comparison. It uses AI to analyse ten qualitative dimensions — including patterns that numbers can't capture, like error fingerprints, characteristic phrase recurrence, and syntactic construction habits.
This is the report that truly sets Verify apart. You don't need anything except the submitted essay itself. The AI segments the document into chunks, establishes a baseline from the document as a whole, then measures every chunk against that baseline.
What it's looking for is not bad writing — it's inconsistent writing. A sudden jump in vocabulary sophistication. A chunk where argument depth drops to descriptive. A change in register halfway through. These are the kinds of shifts a teacher notices but can't quite name.
Each chunk is scored 1–10 on Register, Vocabulary, Sentence Complexity, Argument Depth, Error Density, and Cohesion. The document's own average becomes the baseline — deviations are measured against the writer's own fingerprint, not a generic standard.
Chunks are colour-coded: teal for within baseline, amber for notable, red for flagged. A clean grid gives you a visual read of the essay's consistency before you look at a single number.
If another tool has flagged specific passages as potentially AI-generated, the Consistency Report tells you whether those passages are also stylistically inconsistent with the rest of the document. It's a second, independent signal from a different angle.
The report closes with a characterisation of this writer's style — their typical register, vocabulary level, sentence complexity, and argument depth. A useful reference if you later run a Baseline Deviation Report on a second piece from the same student.
| Chunk | Register | Vocab | Complexity | Arg. Depth | Cohesion | Status |
|---|---|---|---|---|---|---|
| 1 | 3 ↓ | 4 ↓ | 3 ↓ | 2 ↓ | 4 ↓ | Within |
| 3 | 7 ↑ | 7 ↑ | 6 ↑ | 6 | 7 ↑ | Within |
| 11 | 7 ↑ | 7 ↑ | 6 ↑ | 7 ↑ | 7 ↑ | Within |
| 18 | 6 | 6 | 7 ↑ | 7 ↑ | 6 | Within |
This report is an extension of StyleMatch — it takes the statistical anchors StyleMatch produces and layers ten qualitative dimensions on top. The AI reads both writing samples and assesses each dimension, pulling actual phrases from the text as evidence rather than just reporting numbers.
These ten dimensions were chosen because they capture things that statistical metrics miss: how a writer constructs sentences, not just how long they are. Which error patterns recur, not just how many. Whether characteristic phrases reappear across both samples, or whether the vocabulary and idiom belong to two different people.
Recurring grammar patterns, article errors, tense inconsistencies, comma splices. Error patterns are among the most individual stylometric signals — and the absence of expected errors is as significant as their presence. If a student's in-class writing has consistent spacing errors around numbers that vanish in the submission, that's a finding.
Structural construction preferences beyond complexity: how the writer opens sentences, how they link clauses, what phrase-structure habits recur. Two writers can produce sentences of identical average length using completely different structural habits. This dimension captures that.
Whether the same analytical hedges, evidence-introduction patterns, and idiosyncratic expressions recur across both samples. Recent LLM-era authorship research has found that characteristic content words carry strong authorial signal — particularly when a student has a distinctive way of framing claims.
Whether the writer favours noun-heavy constructions (the implementation of, a consideration of) or verb-driven ones (implementing, considering). This cognitive-stylistic habit is remarkably stable across topics and genres — and hard to fake consistently.
The report closes with suggested questions tailored to the specific deviations found — not generic prompts, but questions that reference what was actually observed. What to ask a student about their function word patterns. How to probe the absence of characteristic errors. Ready to use before the conversation.
Every dimension in the Baseline Deviation Report is anchored in peer-reviewed research. The AI isn't pattern-matching against a generic rubric — it's applying stylometric frameworks that have been validated across thousands of texts in authorship attribution, forensic linguistics, and computational humanities research.
The research citations appear in every report — not as a disclaimer, but because a teacher handing a report to a department head or parent deserves to be able to say exactly what it's based on and where it comes from.
Verify runs on credits. Buy a pack when you need one — they never expire. Most teachers use Verify when something in Inspect or StyleMatch warrants a deeper look, not on every submission.
Inspect is always free. StyleMatch requires an active subscription.
10 reports — enough for a full class set when something comes up. Mix Consistency and Baseline Deviation as needed.
35 reports — a semester's worth of deeper investigation for a busy teacher.
100 reports — for a department or shared license. Coordinate through one account.
Student text is transmitted to generate the report, then immediately discarded. Nothing is stored after analysis completes.
Students never interact with PaperTrail. The teacher runs the report from their own account. No student data is collected.
Every report carries a disclaimer — it is a data point, not a verdict. No report should be used as the sole basis for any disciplinary decision.
Designed with North American school privacy requirements in mind. No advertising, no data resale, no third-party tracking.
Most of the time, Inspect tells you what you need. When something warrants a deeper look — before the conversation with the student — that's when Verify earns its credit.
⬇ Add PaperTrail to Chrome