POST /api/analyze is the core endpoint of the Rhetoric Audit API. You send article text and optional metadata; the FME V19.1 engine returns a structured forensic evaluation covering ideology classification, manipulation risk, logical fallacies, emotional resonance, source reliability, and more. Responses are cached by URL and content hash — repeated submissions of the same article return instantly at no cost.
Endpoint
Request
The full text of the article to analyze. Must be between 100 and 100,000 characters. Strip HTML before submitting — the engine expects plain text.
The article title. Optional but recommended — improves classification accuracy and appears in the cached scan record.
The canonical URL of the article. Used to deduplicate identical content across requests. If omitted, deduplication falls back to a hash of the first 100 characters of
text.Your Rhetoric Audit user ID. Required to access the
signed_free or paid tier. See Authentication.A stable client-generated UUID. Required when no
userId is present. At least one of userId or deviceId must be included.Where the request originates. Accepted values:
extension, web, api, share, popup. Pass api for third-party integrations. This value is stored in the scan event log.Response
A successful200 response includes all V18.6 core fields plus optional V19 additive fields when the V19 pipeline is active. The fields below are always present.
Identity and metadata
ISO 8601 timestamp of when the analysis was run (or when the cached scan was originally created).
Ideology classification
A short label describing the dominant philosophical framing of the article, e.g.
"Nationalist Realism".The numerical gap between the
winner and second_place scores. A low gap combined with ambiguity_flag: true indicates weak classification.Manipulation and logic
Emotion and narrative
A prose description of information the article omits that would materially change its framing.
A detailed analytical narrative written at doctoral level, synthesizing all FME dimensions into a single interpretive summary.
Source and credibility
How transparent the article is about its intent and perspective,
0 (fully obfuscating) to 100 (fully informing).Publisher reliability score,
0–10. Derived from a known-publisher lookup table; for unknown publishers the LLM estimate is pulled toward the mean.Cache and quota
Short alphanumeric ID for this scan. Use it to retrieve the scan later via
GET /api/scans/{shortcode} or share it at https://www.rhetoricaudit.com/s/{shortcode}.Code examples
Error responses
The
422 response includes a reason field with a short string describing why the content was rejected, and rejected: true. This is normal for pages that are not news articles — for example, the Chrome extension gates on this check client-side before sending to the API.