Free Standard plan during website beta.
Built for educators navigating questions of academic integrity in student writing — and the researchers studying style itself. Unlike AI detectors that flag writing against a one-size-fits-all norm, comparewriting.com measures each author's own style to find matches. Compare two texts, match an unknown document to a known writer, or explore what makes a piece of writing distinctive, all with peer-reviewed methods, explained in plain language.
See it in action.
Probably different authors. The small, habitual words — articles, prepositions, and connectors that writers use without thinking — appear in noticeably different patterns across the two samples. The submitted essay also favours longer, more subordinated sentences, while the student's sample moves in shorter, simpler units. Punctuation habits are the one feature that stays close. No single measurement is definitive, but the evidence here points away from a shared writer.
Probably different authors. Function-word distributions and character n-gram profiles both show above-average cosine distance between the samples. Mean sentence length differs by roughly six words, and dependency distance patterns are inconsistent. Punctuation distribution is the closest-matching feature family and does not offset the other signals. At this sample size results should be treated as provisional; stylometric reliability improves substantially above 2,000 words per text.
Probably Author A. The clearest overlaps are in the small, habitual words — conjunctions, prepositions, and qualifiers people use without thinking — and in the rhythm of sentence length. Author B's profile shows consistent differences in those same features, and Author C diverges further still across nearly every measurement. Remember, this is stylistic evidence, not proof: context about each person and the document still matters.
Probably Author A. Function-word distributions and character n-gram patterns show the closest pairwise match against Author A's reference profile. Author B's profile differs consistently on the same feature families; Author C's differs more substantially still. Sentence length variance and subordination rate align with Author A's established range on the reference samples. Attribution at this scale of candidate pool benefits from corroborating evidence: document provenance, known writing occasions, and contextual knowledge the data cannot capture.
This is confident, formal prose with a strong scholarly voice. Sentences are long and carefully constructed, vocabulary is precise, and passive constructions appear regularly, all hallmarks of academic writing by someone comfortable in the genre. Sentence length varies in individual, idiosyncratic ways, suggesting an established voice rather than a templated approach.
This text exhibits a high formal register (Flesch Reading Ease: 28.4; Flesch–Kincaid Grade: 16.8) consistent with advanced academic writing. Mean sentence length (34.2 words, SD = 14.7) and a subordination ratio of 0.62 indicate complex syntactic structuring. Type-token ratio (0.54) and hapax legomena rate (0.38) both sit above the corpus median, suggesting broad lexical range. Passive voice frequency (21.4%) is elevated relative to general prose but within the expected range for the sciences.
The plain / academic toggle is available on every results page. Switch at any time once your results are ready.
Three steps to a rigorous result.
Paste or upload texts in any common format, including photos of handwriting. Use writing you know someone wrote to compare it to a text you're unsure of, or just upload texts to learn more about their style.
We measure 80+ stylometric data points using established statistical methods to determine authorship verdicts. An LLM steps in at the end to translate the results into plain language. No hallucinated data, ever.
Every analysis produces a plain-language verdict backed by a full breakdown of every measurement. Switch to academic mode for a technical version with literature references. Share a read-only link with a colleague, or export the full PDF report. Findings are framed as a starting point for judgment, not a final answer.
What makes comparewriting.com different.
Every authorship decision involving student writing is framed as the beginning of a conversation, not an accusation. Findings are presented as evidence to examine together with your students, with space for innocent explanations, open-ended questions, and the judgment only an educator can bring. The tool produces data; you bring the relationship and the context.
Send a read-only link directly from the results page. Anyone with the link sees exactly what you see, no account required. Or export a full PDF report with the full reasoning, all measurements, and the AI's interpretation in one document. Shareable links and PDF export are available on all paid plans.
Upload a photo of handwritten work — an in-class essay, a signed letter, a notebook page — and comparewriting.com transcribes it before running the analysis. No scanning software required. Handwriting support is included on Standard and above, which means in-class writing samples are just as easy to analyze as anything submitted digitally.
Two audiences, one set of methods.
Built for teachers and academic integrity officers working with college-level student writing. When a submission reads differently from a student's other work, comparewriting.com gives you the evidence, and interprets it with care, framing findings as a starting point for conversation rather than a verdict. Because the tool measures each student against their own baseline, it avoids the bias that trips up norm-based AI detectors.
For literary scholars, linguists, and forensic stylometrists. Full metric breakdowns in academic mode, with PDF reports that meet the standards of academic research. Burrows' Delta and related methods computed transparently, so every finding is traceable to real data.
This product is entirely free while the website is in beta. A free plan will always be available, and subscription options will be announced later.