Text → Detect
AI-generatedSummary
Submits text to Truthscan's AI-text detector, polls until the result is ready, and routes each item to an AI, Real, or Raw Output branch with an added classification object.
Inputs
- text (required) — Text to analyze. Must be under 30,000 words; 200+ words recommended for best accuracy.
- generateAnalysisDetails — Default false. When true, requests deep analysis (agreement, confidence, key indicators, reasoning) and keeps polling until the analysis status is no longer pending, which adds time.
- options.model — Detection model; default 'xlm_ud_detector'.
- options.retryCount — Retries Truthscan should make if processing fails; default 0. Sent to the API as retry_count.
- options.pollIntervalMs — Wait between status checks in ms; default 2000, minimum 250.
- options.maxPollAttempts — Maximum status checks before giving up; default 30, minimum 1.
Output shape
Three outputs: 'AI' and 'Real' contain the Truthscan result JSON plus a classification object ({classification: 'AI'|'Real', isAI, score, label}). 'Raw Output' contains the unmodified result for every item.
Classification uses the API's
label(anything other than 'human' counts as AI), falling back to a score >= 50 if no label is present. Fails with an error if no document id is returned, if the document status is 'failed'/'error', or if polling times out before any 'done' status. If a 'done' result was seen but analysis was still pending when attempts ran out, that result is returned. With Continue On Fail, errors are emitted on Raw Output as {error: message}. Requires Truthscan API credentials.
Examples
Example 1: Screen incoming blog submissions for AI-written content and route them differently.
Text = {{ $json.body }}, defaults for other settings; connect the AI output to a review queue and Real output to publishing.
Example 2: Get an explanation of why a text was flagged.
Enable Generate Analysis Details and raise Max Poll Attempts to 60 to allow for the longer analysis.