/detect — AI Detection

Check if text is AI generated The /detect command scans a passage and returns a human probability score plus a per sentence breakdown, powered by Winston AI. Use it to check whether text was likely...

Check if text is AI-generated

The /detect command scans a passage and returns a human probability score plus a per-sentence breakdown, powered by Winston AI. Use it to check whether text was likely written by an AI model.

How to use

/detect [paste the text you want to analyze]

Paste the text directly after the command. One scan accepts 300 to 150,000 characters, so essays, articles, and reports fit in a single run instead of chunks. Scans run on English text.

You can also skip the command and ask in plain language — “check if this was written by AI” — and the agent runs the same analysis.

Results

The result card shows:

Human probability — a 0–100 score for the whole passage; higher means more likely human-written, with the complementary AI percentage shown beside it
Verdict — Likely Human (75%+), Mixed Content (50–74%), or Likely AI (under 50%)
Sentence analysis — every sentence scored on the same scale, with likely-AI lines highlighted so you can quote, rewrite, or rescan just the flagged parts
Text stats — word and sentence counts

What it detects

Winston’s model is tuned for mainstream generators like ChatGPT, Claude, Gemini, and Llama. It scores sentence structure, vocabulary distribution, and writing-style consistency to estimate how much of the text reads as machine-written.

Limits

Detection is probabilistic, not proof — treat the score as one signal, not a verdict on a person
Passages under ~600 characters produce less reliable scores; paste whole paragraphs rather than single sentences
Heavily edited or mixed human/AI text is harder to flag
English text only

Credits

3 credits per 1,000 characters, minimum 3 credits. If the scan fails, the reserved credits are refunded.

Input length	Cost
Up to 1,000 chars	3 credits
2,400 chars	9 credits
10,000 chars	30 credits
50,000 chars	150 credits

/plagiarism — scan the same text for copied sources
/humanize — rewrite flagged passages into natural phrasing
Understanding credits — how per-character billing works
Detection is also available through the Krater API (POST /v1/text/detect) and the Krater MCP server.

More in Slash Commands

See Krater pricing or contact support.