AI Citation Auditor
Pricing
from $5.00 / 1,000 claim verifieds
AI Citation Auditor
Verify every citation in AI-generated text: fetches each cited source and returns supported/contradicted/unsupported verdicts with quoted evidence spans. Catches hallucinated citations before you publish.
Pricing
from $5.00 / 1,000 claim verifieds
Rating
0.0
(0)
Developer
Dominik Spacek
Maintained by CommunityActor stats
0
Bookmarked
2
Total users
1
Monthly active users
5 hours ago
Last modified
Categories
Share
Verify every citation in AI-generated text against the source it actually cites.
LLMs routinely produce citations that look real — the URL resolves, the paper exists — but the source doesn't say what the text claims it says. In June 2026 KPMG had to pull a flagship report after most of its citations turned out to be fabricated or misattributed. This Actor catches that before you publish.
What it does
For every citation in your document, the Auditor:
- Extracts claim + source pairs (markdown links, footnotes, DOIs, bare URLs)
- Fetches the cited source — dead links get an automatic archive.org fallback
- Reads the actual page content (boilerplate stripped, PDFs supported)
- Judges whether the source supports the claim, using an LLM that sees only the fetched source text — never its own world knowledge
Output
One row per claim:
| Field | Meaning |
|---|---|
verdict | supported / partially_supported / contradicted / unsupported / uncertain / could_not_fetch |
evidence | Verbatim quote from the source backing the verdict — verify it yourself in seconds |
confidence | Judge confidence 0–1 |
reasoning | One-sentence explanation |
httpStatus, resolvedUrl, fromArchive | Source liveness details |
Plus a REPORT in the key-value store with a citation integrity score (0–100) and per-verdict summary.
Input
Three ways to use it:
What it is not
- Not an AI-text detector — it doesn't care who wrote the text.
- Not a truth oracle —
supportedmeans the cited source says this, not this is true. - Never guesses — unreachable sources are reported as
could_not_fetch, not judged blind.
Use cases
- Pre-publish gate for AI-drafted reports, blog posts, whitepapers
- CI check for docs and marketing content
- Auditing research summaries and literature reviews
- Due diligence on any citation-heavy document
Open-source core: github.com/Franksterino/citeguard
