Drop in your draft
Upload a Word file or PDF, or paste the text. Genie pulls out the words. Layout, fonts and footers stay out of the way.
Drop in a draft. Writing Genie finds copied passages and machine-sounding prose, explains what gave it away, then rewrites the weak spots without losing your point.
Drop your draft right here
or browse your files
Your draft is not saved. Large uploads are deleted after extraction.
Upload a Word file or PDF, or paste the text. Genie pulls out the words. Layout, fonts and footers stay out of the way.
Two scores, sentence-by-sentence heatmaps, and the exact phrases that made a detector raise an eyebrow.
One click rewrites the flagged bits: stock phrases gone, rhythm loosened, copied passages paraphrased. Then it checks again.
AI detectors look at style. Plagiarism checks look at source overlap. Writing Genie keeps them separate, so you know which problem you actually have.
Measures phrasing, sentence rhythm, transitions, contractions, punctuation habits and formulaic patterns, then explains the signal in plain English.
Heuristic · explainableCompares your sentences against Wikipedia, OpenAlex academic abstracts, and a library of well-known quotations. Matching spans are linked to their source.
Public reference scanNo, and nobody honestly can. Detectors are inconsistent and produce false positives on human writing all the time. Writing Genie reduces the stylistic patterns those tools tend to flag and shows you the reasoning, so you can make the final call.
Not the whole web. It checks Wikipedia, OpenAlex academic abstracts, and a curated quotation library, plus repeated passages inside your own document. A complete web index requires a paid service like Copyleaks or Turnitin.
Locally, the file stays on your machine. On the hosted version it is sent to Vercel Blob, processed by the serverless function, and deleted immediately after extraction. Only short quoted phrases are sent to the public Wikipedia and OpenAlex APIs for source matching.
It swaps words and reshapes sentences, so small changes in tone are possible. Read the result. If a passage is a direct quote, cite it. Don't paraphrase it into something the original author didn't say.
Word documents (.docx), PDFs with selectable text, plain text and Markdown. Scanned PDFs need OCR first. Legacy .doc files should be re-saved as .docx. Long documents are split into roughly 60,000-character parts automatically, processed one after another with a progress indicator, and merged into one final report.
Yes. The cleaned draft can be downloaded as DOCX or PDF with a styled title, subtitle, Heading 1 to 3 levels, bullet lists and quotations. Word documents use real heading styles, so the navigation pane works. Structure is taken from uploaded Word headings where possible and inferred from the text otherwise.
Up to 3,000,000 characters and 25 MB per file on the hosted demo (12 MB locally). Large files upload directly to Vercel Blob, so the usual 4 MB serverless request limit does not apply. Documents beyond 60,000 characters are processed in parts automatically; above 1.2 million characters the sentence heatmap is skipped to stay inside serverless response limits.