Human probability score
Each scan returns a 0 to 100 human score for the whole passage. Higher means more likely written by a person.
Paste a passage and get a human probability score plus sentence-level flags from Winston AI. 300 to 150,000 characters, 3 credits per 1,000.
Paste a passage and get a human probability score plus sentence-level flags from Winston AI. 300 to 150,000 characters, 3 credits per 1,000.
Paste a passage and get a human probability score plus sentence-level flags, powered by Winston AI. Scans run on English text from 300 to 150,000 characters and cost 3 credits per 1,000 characters.
Krater pricing · AI model directory · AI Image Generator and Photo Editor
Each scan returns a 0 to 100 human score for the whole passage. Higher means more likely written by a person.
Every sentence gets its own score on the same scale, so you see exactly which lines read as machine written.
Scans accept 300 to 150,000 characters, so essays, articles and reports fit whole instead of in chunks.
3 credits per 1,000 characters with a 3 credit minimum. A 2,000 character scan costs 6 credits, shown before it runs.
Winston's model is tuned for mainstream generators like ChatGPT, Claude, Gemini and Llama.
The score card arrives as a normal chat result: review it, rerun a shorter excerpt, or share the conversation.
The passage goes right after the command. Anything from 300 to 150,000 characters works in one scan.
Cost is 3 credits per 1,000 characters. Under 600 characters the score is less reliable, so prefer whole paragraphs.
The result card shows a 0 to 100 human probability for the whole passage. Low scores point to machine writing.
Each sentence is scored on the same scale, so you can quote, rewrite or rescan the parts that triggered the flags.
Screen essays and assignments for machine-written passages before grading.
Check freelance submissions and guest posts for undisclosed AI drafts.
Verify that campaign copy reads human before it ships.
Score your own draft before submission to see what a detector flags.
Spot-check outsourced articles that are supposed to be human written.
Detection is probabilistic. Human writing can score as AI and AI text can pass as human, especially short or edited passages. Use the score as one signal among others.
Scans run on English input. Passages in other languages need a different check.
There is no free plan for detection. Scans start on Pro at $20 per month. For a single check, a free detector is enough.
Not fully. /detect runs on credits at 3 credits per 1,000 characters of text. Pro is $20 per month and includes monthly credits.
The result is a human probability, not a verdict. Scores on short, heavily edited or mixed human and AI passages are less reliable, and every detector can produce false positives.
Between 300 and 150,000 characters per scan. Under 600 characters the score is less reliable, so paste whole paragraphs rather than single sentences.
The detector is tuned for mainstream generators like ChatGPT, Claude, Gemini and Llama. Heavily rewritten or human-edited AI text is harder to flag.
Rewrite flagged passages into natural phrasing.
Scan the same text for copied sources.
Condense long documents before analysis.
Every model on Krater with credit pricing and capabilities.