Document AI detector
Inspect a complete document.
Upload PDF, DOCX, HTML, TXT or Markdown. Text sections and embedded images are analysed separately and combined into a warning-only report.
- Free to use, no account and no sign-up. Heavy automated use is throttled so the checker stays available
- Documents are split into text sections and embedded images
- Sampled video submits up to 5 representative frames
The file stays in your browser. Extracted text sections and images are submitted separately for analysis.
PDF, DOCX, HTML, TXT or Markdown · max. 20 MB · up to 40 sections and 24 imagesChecking API availability…
Document analysis
Document analysis
Section and image results will appear here.
Your recent checks
Kept in this browser only and never uploaded. It records the verdict, not the file: no image, no share link and nothing that identifies you.
What this result can tell you.
The detector looks for patterns associated with synthetic or generated content. It cannot prove who made a file, which tool was used or whether the content is truthful.
- Use the signal to decide what to inspect next
- Review source and Content Credentials when available
- Do not accuse a person or block content from this result alone
- Unknown is a valid outcome
Uploads are temporary and are not used for model training by default. Avoid sensitive, unlawful or children’s media.
Read the privacy draftThe same detector as an Android app or a Chrome extension — both direct downloads.
Get the appsAI document detection guide
Check complete documents section by section.
A document score is only useful when the extracted text and images still represent the source. This checker makes selection, processing, section results and limitations visible.
Supported document formats
- PDF files with selectable text, positioned text or page images
- DOCX files containing paragraphs, headings and supported embedded images
- HTML and HTM after scripts, styles, navigation and hidden elements are removed
- Plain TXT and Markdown documents
The visible upload state confirms the exact filename, file size and MIME type. Analysis starts only after the user presses the analysis button.
What happens in the browser
The file structure is read locally. Text is divided into bounded sections and supported embedded images are prepared separately. Those extracted units are then submitted for warning-only analysis. The original complete document is not itself treated as a single meaningful percentage.
Limits on section and image counts keep the workflow usable. If a document is truncated, scanned, malformed or password-protected, the status explains that instead of silently pretending extraction succeeded.
PDF extraction and layout limitations
A PDF is a page-description format, not necessarily a clean sequence of paragraphs. Words may be stored as individual characters, columns can share coordinates and text boxes may appear out of reading order. The parser repairs common spacing and column patterns conservatively, but you should still compare the displayed sections with the original document.
Scanned pages may contain no selectable text. Page images can provide visual signals, but this is not the same as accurate OCR. Tables, formulas, footnotes and decorative headings can also produce short sections that are unsuitable for writing-style analysis.
DOCX and HTML handling
DOCX
Headings and paragraphs are extracted from the Office document structure. Supported embedded images are checked independently. Damaged archives, unsupported objects and protected files produce a visible error rather than an endless loading state.
HTML
The parser removes scripts, styles, templates, navigation, footers and hidden elements before extracting readable text. This avoids analysing tracking code and interface labels as if they were article content.
Understanding the downloadable PDF report
After a successful analysis, the checker creates a PDF detection report containing the overall warning label, analysis time, filename, modality, detector details, processing state and available section context. The wording is deliberately limited to No strong AI signal found or Possible AI-generated indicators found.
The report is not a certificate of authorship, truth or authenticity. A client-generated file can be copied or altered. If a server-signed verification URL is included, use that URL as the authoritative record and still apply the stated limitations.
A careful document review workflow
- Confirm that the filename and file type shown after upload match your source.
- Compare extracted headings and text samples with the original pages.
- Review section-level results instead of relying only on an average.
- Check embedded images separately when they affect the document's meaning.
- Collect drafts, references, revision history and declarations for consequential decisions.
- Download the report as a record of the check, not as proof.
Read the PDF and DOCX review guide, how to interpret scores and the trust principles.
AI document detector FAQ
Which files can I upload?
PDF, DOCX, HTML, HTM, TXT and Markdown files up to the displayed size limit.
How do I know selection worked?
The upload area turns into a visible selected state and shows filename, size and MIME type. The button changes to Analyse document.
Can it read scanned PDFs?
Scans may contain no selectable text. Visual page checks are not equivalent to reliable OCR.
Is the PDF report proof that a document is legitimate?
No. It records a detector warning and limitations; it does not prove origin, authorship or authenticity.
Why should I compare extracted text?
Columns, text boxes, custom spacing and damaged fonts can change PDF reading order or word boundaries.