AI text detection
Two language models trained on millions of human and AI passages score your text, paragraph by paragraph.
AI Text Detector
Find out if writing came from ChatGPT, Claude, Gemini or another AI. Paste up to 300 words free, no sign-up needed.
- Free AI text checks
- No sign-up needed
- Your content is never stored
- Every check for $9.99/month, free for 7 days
Trained to recognize writing from ChatGPT, Claude, Gemini, Llama and Mistral, and images from Midjourney, DALL·E, GPT Image, FLUX, Stable Diffusion, Imagen and Firefly.
Four checks in one place
Start free with AI text detection. Go Pro for images, plagiarism, fact checking and documents.
AI text detection is free right now. Images, plagiarism, fact checking and documents are coming soon.
AI image detection
Forensic vision models find the pixel traces generators leave, and read Content Credentials embedded in files.
Plagiarism check
Finds passages that appear in published sources, with a similarity score and a link to each source.
Hallucination check
Tests every claim in an AI answer or summary against your source or Wikipedia, and shows the evidence.
Documents of any length
Upload PDFs, Word files, slides and e-books, several at a time, with passage highlights and no word limit.
Private by design
Content is analyzed in memory and discarded. Never stored, never sold, never used for training.
Tested on thousands of real examples
We measure accuracy on public benchmarks the models were not tuned on, and we publish the misses too. See the full results.
94%of AI-generated images caught
2%of real photos wrongly flagged
99%of text from 2026 frontier AI models caught in an independent benchmark
<1%of human writing wrongly flagged on 7 of 8 public text benchmarks
Image results: 4,011 images from 20+ generators and 4 real-photo collections. Text results: 5,000+ texts from the RAID, MAGE, COLING 2025, Beemo, Ghostbuster, TextSight 2026, Epoch AI 2026 and Russell et al. 2025 benchmarks, at the 60% threshold; the 99% is Epoch AI's plain-request texts from Claude Opus 4.8, GPT-5.5 and Gemini 3.1 Pro. Some test sets come from the same families as our training data. Real-world accuracy varies, especially for short, edited, formal or translated content. Results are estimates, not proof.
How it works
- 01
Paste or upload
Add text, a document, or an image. Nothing to install.
- 02
We analyze it
Detection models read writing patterns and image pixels, search for sources, and check claims against evidence.
- 03
Get a clear answer
A score, a plain-English verdict, and the evidence behind it. Read the details.
Every check, one low price
Free now, Pro coming soon
AI text checks are free for everyone, no account needed. Pro, with AI images, plagiarism and fact checking, is on its way.
AI text, AI images, plagiarism and fact checking for $9.99/month, with no word credits to count. Try everything free for 7 days and cancel anytime in two clicks.
Free
$0
No account needed.
- AI text checks up to 300 words
- Overall AI likelihood score
- AI image detection
- Plagiarism check
- Hallucination check
- Document uploads and highlights
7 days freeComing soon
Pro
$9.99/month
7 days free, then billed monthly. Cancel anytime.
Soon
Not available yet. Everything below is on its way.
- AI text checks of any length
- AI image detection
- Plagiarism check with source links
- Hallucination and fact check
- PDF, Word and other document uploads
- Passage-by-passage highlights
- No monthly word credits
How Pro compares
| Clarity Pro | Typical AI detector plan | |
|---|---|---|
| Monthly price | $9.99/month | $14.95 to $18 |
| AI text detection | Included | Included |
| AI image detection | Included | On some plans |
| Plagiarism check | Included | Included |
| Hallucination and fact check | Included | Enterprise plans only, or not offered |
| Monthly word or credit limits | None | Yes |
| Try before you pay | 7 days, everything unlocked | A free plan or a limited trial |
"Typical" reflects the entry-level monthly plans of Originality.ai ($14.95) and Winston AI ($18) as listed on their websites in October 2026. Competitors' plans and prices change; check their sites for current details. Pro includes fair-use daily limits on plagiarism and fact checks.
Frequently asked questions
How accurate is the AI detector?
In our benchmark tests on 4,011 images, it caught 94% of AI-generated images while wrongly flagging about 2% of real photos. For text, on Epoch AI's independent 2026 benchmark it caught 99% of writing from Claude Opus 4.8, GPT-5.5 and Gemini 3.1 Pro, and 92% when those models were told to imitate a real person's style, without wrongly flagging any of its human texts. Across 7 of 8 public benchmarks, under 1% of human writing is wrongly flagged. No detector is perfect, so results are shown as a likelihood with an 'Uncertain' band rather than a forced yes/no.
Which AI models can it detect?
Text from ChatGPT, Claude, Gemini, Llama, Mistral and other large language models. Images from Midjourney, DALL·E, GPT Image, FLUX, Imagen, Adobe Firefly, Stable Diffusion, Ideogram and many more.
Is it free?
Yes. You can check text up to 300 words for free without an account. Pro, with image detection, plagiarism and fact checks, document uploads, unlimited length and passage-by-passage highlights, is coming soon.
How does the plagiarism checker work?
It searches for distinctive sentences from your text, fetches the matching pages, and compares them word sequence by word sequence. You get the share of your text found in each source, the matched passages highlighted, and links to the sources. It finds copied and lightly edited text; heavily paraphrased text is better caught by the AI detector.
What is the hallucination checker?
AI chatbots sometimes state things that are false or not in their source. The checker splits a text into claims and tests each one against evidence, either a source document you paste (best for summaries and AI answers about your own material) or Wikipedia. Each claim is marked supported, contradicted, or not verifiable, with the evidence shown.
Do you store my text or images?
No. Content you check is analyzed in memory and discarded immediately. It is never saved, sold, or used to train AI models. Plagiarism and fact checks send short excerpts (a sentence or phrase at a time) as search queries to Wikipedia and our search provider to find sources; nothing else leaves our servers.
What file types are supported?
Documents: PDF, Word (DOCX and DOC), Google Docs exports, ODT, RTF, TXT, Markdown, HTML, EPUB and PowerPoint. Images: JPG, PNG, WebP, HEIC (iPhone photos), AVIF, GIF, TIFF and BMP.
Can I use the result as proof that someone used AI?
No. Detection results are probabilistic estimates and can be wrong. Use them as one piece of evidence alongside others, never as the sole basis for an accusation or a decision about a person.