Likely AI-written
The passage has predominantly AI-like signals. Look at the sources, drafts, and context before drawing a conclusion.
Before you publish · Free beta
Could it be AI? Take a closer look.
Check writing and images before you share. A second opinion, not proof of authorship.
- No account
- Nothing is sent until you agree
- CouldItBe does not save your content or result
- How we handle your content
Text detector Checking…Image detector Checking…
Formats, limits & sharing
Paste at least 50 words, or open TXT, DOCX or a selectable-text PDF. Review the writing, then consent to GPTZero screening.
Add your writing
A detector scan needs at least 50 words. TXT, DOCX or selectable-text PDF up to 8 MB also work.
Scan paper pages with a camera
Choose how to check
Add your writing to continue.
Your result
See a sample report (opens in a new tab)Your result will appear here
Add your writing, choose how to check, and press the button.
- ✓ A plain-language assessment
- ✓ What it means — and what it doesn’t
- ✓ Suggested next steps you can tick off
- ✓ A report to keep (your text is excluded)
AI detectors can miss AI and flag human work. Use a result to guide a closer look, never as the sole basis for an accusation. See our actual detector test results.
Part of your editorial review
Check the content.
Keep the context.
Review a draft, a supplied image, or a published page. Use the result to decide what deserves a closer look. Follow the review checklist →
- 01
Add your content
Paste writing, open a document, or select an image.
- 02
Run the detector
Agree to the scan. GPTZero checks text; Sightengine checks images. Availability depends on our shared beta allowance.
- 03
Follow up on the result
Compare drafts, sources and disclosures. Save a report for your review, and ask the contributor about anything that needs context.
Reading the result
An answer you can actually use.
Text results come from GPTZero. Confidence describes its assessment, not the share of words written by AI.
Mixed signals
AI-like and human-like signals appear together. Editing, collaboration, and writing style may all play a part.
Likely human-written
The passage has predominantly human-like signals. This doesn’t rule out AI assistance.
See a sample report before you scan →
Checking an image? You’ll see a Sightengine model score instead. Higher scores mean stronger AI-generation signals, not a verified percentage of the image made by AI.
Make yourself comfortable
A space that feels like you.
Four ways to see the same tools. Only your theme preference is saved in this browser.
More ways to explore
Supporting tools
Website text can be screened after you review it. Video offers experimental still-frame screening. Local technical reviews remain available separately.
Good questions
Before you press check.
The useful details, without the fine-print hunt.
How much text should I check?
Use at least 50 words and no more than 10,000 characters for a detector scan. A complete, representative passage is more useful than a few isolated sentences. Longer text can still receive local observations, but it won’t be silently shortened and sent.
Can I upload a document or scan a paper page?
Yes. Open a UTF-8 TXT, DOCX, or selectable-text PDF up to 8 MB. Paper pages can be transcribed with the camera workflow. Review the extracted text, then consent to a detector scan.
Is my writing or image saved?
Local review uses browser memory. A detector scan sends your consented text to GPTZero or image to Sightengine through CouldItBe. CouldItBe does not save the scan input or result. Research contributions are a separate choice. Read the privacy policy.
What if the detector is wrong?
Detectors can flag human work or miss AI. Treat the result as a second opinion. Compare the original, drafts, and sources, and ask about editing or assistance. Don’t use a result alone to accuse someone.
What can I check before publishing?
Use Text for a draft or document, Image for an illustration or supplied image, and Web for the extracted writing on one public page. Files checks a queue of TXT documents and images. Video is an experimental check of three still frames, not a full-video or deepfake verdict. Compare what each tool covers.
Does this find plagiarism or identify a specific AI model?
No. AI detection is different from matching copied text, and these checks do not reliably identify the exact model or person behind content.
Why is a detector sometimes unavailable?
Provider availability, rate limits, and our shared beta allowance can affect access. Retry the availability check. If no detector result is produced, local observations stay separate and are not presented as a replacement classification.
Research and how we improve the tools
Our local writing patterns are experimental and are kept separate from the detector’s result. We publish our pilot findings, including where the patterns failed to separate writing groups. Optional feedback and research contributions never automatically train the live detector. Contributions contain only derived measurements and an unverified source label, never original writing or media. Study pages explain adult eligibility and whether submissions are open.
Share feedback · Read the methodology · Writing study · Audio study · Video study