Create something great with GramProof today

The checker allows three successful scans per UTC day per anonymous session, with 100 to 2,000 words per scan. A real AI detector must be configured for analysis. Paid plans are not connected yet.

Predictable phrasing, repetitive structure, short passages, and heavily edited writing can be difficult to interpret. A flag does not establish who wrote a passage. Review the surrounding context and the writer’s process before drawing conclusions.

Keep your notes, outlines, sources, and draft history. These can help explain how your writing developed. The planned Writing Replay feature is designed to make that process easier to share; the replay above is an interactive illustration.

Scan History is planned for signed-in accounts so you can revisit results and reports. This demo does not save scans, and refreshing the page clears the editor.

English is the supported language in this preview. Additional language support and language-specific performance information will be published when available.

The displayed plans use a monthly credit allowance that expires after one month and renews monthly, even when billed annually. These are preview plan details; final billing terms will be available before purchase.

Three successful detector responses are allowed per UTC day. Failed requests do not use the allowance. An anonymous session cookie keeps the server-side counter connected across refreshes; persistent account-level limits are not connected yet.

No account is needed for the current checker. The server enforces an anonymous session allowance.

Scans send your submitted text through our server to the configured AI detector. The app does not save your text or results; the detector service has its own retention policies. Uploaded documents are converted to plain text in your browser. An HttpOnly cookie identifies your server-side scan allowance.

An AI score is intended as an estimate of patterns associated with AI-assisted writing. It is not proof of authorship or a definitive count of AI-written words. The checker displays the score returned by the real detector. Confidence is shown only when available, and sentence highlights require provider sentence-level results. The separate product showcase still uses illustrative results.

No. Consider it alongside drafts, sources, writing history, assignment requirements, and a conversation with the writer. Avoid making consequential decisions from a score alone.

Add or upload text, choose an allowance, and scan. The server submits the document to the configured AI detection provider and returns its results. The allowance compares against the raw score without changing it. No score is generated when the service is unavailable.

Detecting AI-assisted paraphrasing is a planned capability. Rewritten text can be challenging to assess, and no result should be treated as certain. Capabilities depend on the configured detector. No result should be treated as certain, especially for rewritten or heavily edited text.

A false positive occurs when human-written content is incorrectly flagged as AI-generated. It is one reason to treat detection results as a prompt for further review rather than a verdict on a writer’s originality.