Check your content. Understand the result.

Choose the right tool

• Text: paste writing, open TXT, DOCX or a PDF with selectable text, or photograph a paper page. Correct extracted text before scanning. Local review measures writing patterns; the optional GPTZero scan provides a separate AI classification.
• Image: open the original JPEG, PNG, WebP or GIF. Review visual measurements and embedded generation information. A screenshot or re-saved image may lose useful metadata. When available, a separate consented Sightengine scan adds an AI-generation score.
• Website: enter one public page URL. We temporarily retrieve its available text, metadata and structure. Login pages and pages loaded mainly by scripts may not provide enough content. This does not scan an entire site or establish who wrote its code.
• Audio: open a supported recording and review its technical measurements. Noise, silence or clipping are recording characteristics, not proof of an AI voice.
• Video: open a supported video or a direct public MP4/WebM link. Review sampled frames and transitions. Sampling can miss edits; the current tool does not classify a video as a deepfake.

Each tool shows its supported formats and size limits before you start.

Text: AI screening with supporting evidence

1. Paste or open your writing. For camera input, use good light, keep the page flat, and check every extracted sentence.
2. For an AI classification, check the consent box and select Check for AI writing. GPTZero screening needs at least 50 words and no more than 10,000 characters. The same report also includes local observations.
3. To keep writing on your device, select Review local patterns only. This measures structure and wording but does not classify AI authorship. Document type affects only this local comparison.

GPTZero receives the editor text through our server only for that requested scan. Its finding appears first in the main evidence report. Editing the text clears the server result and consent; another scan needs consent again. Our local report and GPTZero result are not combined into a percentage.

Image: local evidence and optional AI screening

1. Select an original image. Local inspection runs on your device.
2. If the Sightengine option is shown, use a JPEG, PNG or WebP up to 10 MB. Check the upload consent box and select Check image for AI. GIF remains available for local inspection only.
3. Read the provider score separately from the local metadata and pixel findings. A higher score indicates stronger AI-generation signals according to Sightengine, not independently measured accuracy.
4. Download the screening result separately from the local evidence report. Neither download contains the image bytes. Selecting a different image clears consent and the old screening result.

The optional scan sends the selected file, including embedded metadata, through our server to Sightengine. It does not add that file to our research or training. If the option is absent, server image screening is not enabled.

What the result means

AI-like, human-like or mixed describes GPTZero’s assessment. “Human-like” does not prove that AI was absent, and “AI-like” does not prove misconduct. Provider confidence is not a measured probability that the author used AI.

Local patterns describe the content we measured. Missing metadata is neutral. Embedded generation information is a declaration unless independently verified. If two analyses disagree, keep both findings and check original files, drafts or editing history.

A limited result means the tool cannot support a stronger conclusion from this input. Use a longer passage or a clearer original where appropriate. Never change a document’s type just to obtain a different result.

Text report download and print controls include the completed GPTZero finding, provider confidence, timestamp and local evidence. Source text is excluded from downloads. Browser Print / Save as PDF is available where the tool provides a print control.

You control sharing

Local analysis, a server scan, and a research contribution are separate choices. Agreeing to one does not agree to the others. We do not currently retain GPTZero input or results to train our engine.

Optional measurement contributions preview what is shared and exclude your original text and media. They do not automatically retrain the live tool. Save the deletion receipt if you contribute. Separate volunteer studies have their own consent requirements.

Only send content you are authorized to share. The public server scan is not intended for confidential or sensitive student, medical or government records. See the privacy policy and terms.

If something does not work

• No extracted text: use a selectable-text document or photograph the page and correct the transcription. Handwriting may need manual correction.
• Server scan unavailable: it may be temporarily unavailable or its shared beta allowance may be exhausted. Local review remains available.
• Scan button disabled: check the word/character limits and the consent box.
• Scan limit reached: wait before retrying. A shared allowance may require us to restore capacity.
• Hard to read: choose Light, Dark or Graphite from the theme selector at the top of the page.

Report a problem or visit About & contact. Please do not include sensitive source material in support messages.