API
Last updated 2 October 2026
GerviGrein has a public API to look up a news article's AI score and to ask for an article to be scored. No key or sign-up is needed. GerviGrein's browser extension uses the same API.
- Base URL:
https://gervigrein.is. HTTPS only. - Responses are JSON (UTF-8).
- Ten outlets are tracked:
ruv.is,mbl.is,visir.is,dv.is,heimildin.is,vb.is,nutiminn.is,mannlif.is,frettatiminn.isandfrettin.is.
Look up an article
GET /api/lookup?url=<article URL>
Returns the article's AI score if it has been scored, and the outlet's estimated share of AI-written articles
over the last 30 days. Add &l=en to get links to the English version of the article's page
and card.
GET https://gervigrein.is/api/lookup?url=https://visir.is/g/20262942330d/bjoda-ut-sorphirdu-i-breidholti-til-sex-ara&l=en
{
"tracked": true,
"outlet": "visir",
"url": "https://visir.is/g/20262942330d/bjoda-ut-sorphirdu-i-breidholti-til-sex-ara",
"article": {
"id": 3396,
"title": "Bjóða út sorphirðu í Breiðholti til sex ára",
"words": 972,
"published": "2026-10-01T23:00:45+00:00",
"page": "/a/3396?l=en",
"card": "/a/3396/card.png?v=629eedbe00&l=en"
},
"status": "scored",
"score": 0.07,
"detector": "blend",
"level": "human",
"detectors": {
"icebert": {"score": 0.05, "line": 0},
"claude": {"score": 0.09, "line": 0}
},
"band": 0,
"can_request": false,
"requests": 0,
"outlet_stats": {
"name": "Vísir", "share": 0.275, "text": "28%", "scored": 534,
"rank": 5, "of": 10, "estimate": true, "period": "30d"
}
}
Fields
| Field | Meaning |
|---|---|
tracked | false if the URL isn't on one of the ten outlets' sites; no other fields follow. |
outlet | The outlet's id: ruv, mbl, visir, dv, heimildin, vb, nutiminn, mannlif, frettatiminn, frettin. |
url | The article's normalised URL. Missing, along with status, when the URL is a front page or listing (article is then null). |
article | The article, if it's in the database: id, title, words, published (ISO 8601), page (its page on GerviGrein) and card (a 1200×630 PNG whose URL changes when the score does). Paths are relative to the base URL. Otherwise null. |
status | See below. |
score | AI score from 0 to 1, the same score as on the article's page; higher means more traits typical of AI-written text. It is on a common scale measured against human-written news: 0.1 is the 5% line (about 1 in 20 such articles score higher) and 0.5 the top line (higher than every human-written article the detector was compared with). Only when status is scored. |
detector | Which estimate score is: blend (Mixed, both detectors), else icebert (Language detector) or claude (Presentation detector). |
level | The article's verdict, from every detector: all Both detectors find AI signals (both past their top lines), one One detector finds AI signals, single Rare for human-written news (the only detector, past its top line), some Some unusual traits (past the 5% line), human Reads like human-written news. |
detectors | Each detector's own score on the same scale; line is 0 below its 5% line, 1 past it, 2 past its top line. |
band | Deprecated, for older clients: the old verdict 0–4, now worked out from score and level. Use level. |
can_request | Whether POST /api/request-scan could add the article. |
requests | How many times the article has been requested. |
outlet_stats | The outlet over the last 30 days: share is the estimated share of AI-written articles (0–1, corrected for the detector's error rates; text is the same number formatted, or "≥X%" when the correction reaches 100%; null and "—" when no articles are scored), scored the number of scored articles, rank of of (1 = highest share). estimate is false when the correction can't be applied; share is then the share of flagged articles. |
Values of status
scored | The article has been scored. |
queued | The article is in the database and will be scored automatically soon. |
summary | Only an RSS summary is stored so far; the full text is awaited. |
short | The text is too short to score. |
not_auto | The article is stored but won't be scored automatically (e.g. older than 14 days). |
requested | The article was requested and is being fetched. |
failed | Fetching the article failed. |
unknown | The article isn't in the database. |
Request a scan
POST /api/request-scan
Content-Type: application/json
{"url": "https://www.ruv.is/frettir/innlent/..."}
Send a link only, never text. GerviGrein fetches the article itself from the outlet's site (mbl.is: only an
archive.org copy, which usually appears within a day) and scores it within its daily limits. The response is
the same as from /api/lookup, plus accepted (whether the request was taken).
202: the request was taken. Look the article up again in a few minutes.200: the article was requested before, or is already stored.400: the URL isn't an article on one of the outlets' sites.429: too many requests; try again later.
Limits and fair use
- Up to 120 requests a minute per IP address, and 10 a minute for
/api/request-scan. - New scan requests: 20 a day per IP address and 200 a day in total.
- Send a descriptive
User-Agentwith a way to contact you, and keep answers for a few minutes instead of asking again and again. - If you plan to send many requests, write to [email protected].
Using the results
- Credit GerviGrein and link to the article's page (
page) when you show a score. - A single article's score is uncertain. Don't present it as proof that an article was written by AI; the outlet's share is the more reliable measure. See "About" on the home page for how it's calculated.
- The API may change. It's provided as is, without warranty; see the terms of use.
- What's stored about requests: see privacy.