Veridicta

Verification, not AI-detection

Catch citations that don’t exist, before your report leaves the building.

Veridicta checks every citation in a finished document against real bibliographic databases, Crossref, OpenAlex and arXiv, and reports with evidence exactly what it could and could not verify. Either a DOI resolves or it does not. No guessing, no “AI-written” scores.

Run the 60-second demoSee how it works

No sign-up. The demo scans a sample report and visibly catches a citation that does not exist.

Source check
12 citations

Karpathy et al. (2024), Nature Machine Intelligence

Exists

DOI resolves; title, authors and year match.

Halloran, J. (2023), “Latent audit drift,” J. Appl. Finance

Not found

No match in Crossref, OpenAlex or arXiv.

Okafor & Lin (2022), Review of Economic Studies

Minor issues

Source found; cited year is off by one.

Trust score
9 of 12 verified

Illustrative result.

Why this matters now

AU$440K

refunded by Deloitte Australia over non-existent references in a government report.

Deloitte Australia, 2025

5 of 45

citations real in a pulled KPMG report. The rest could not be found.

Reported 2026

$145K+

in US court sanctions for AI-invented citations in Q1 2026 alone.

US courts, 2026

The problem

AI now drafts the work. Nobody checks the citations.

Consultants, analysts, expert witnesses and auditors increasingly draft citation-heavy work with AI assistance. The prose is fluent. The footnotes look right. But language models invent sources that never existed, attach real authors to the wrong paper, and quote passages that are not in the document cited.

These mistakes are now reaching clients, regulators and courts, with real financial and reputational cost. Veridicta is the check that runs before you publish, so a fabricated citation never makes it out the door.

What it checks

Three checks, one verdict you can defend.

Each check produces evidence you can open and inspect. Nothing is a black box.

Existence

Every citation is resolved against Crossref, OpenAlex and arXiv, with per-component matching on title, authors, year and DOI. Either the source is there or it is not.

Quotes

Every quotation is checked word for word against the source you cite, with a side-by-side diff when the text does not match.

Grounding

An assistive, opt-in check on whether the cited source actually supports the claim, flagging statements a source may not back up.

The Source-Verification Certificate

A tamper-evident record of what was checked and what was found, that you can hand a client, reviewer or court. Shareable, signed, and re-verifiable by anyone, with no account.

How it works →

How it works

From upload to evidence in under a minute.

01

Ingest

Upload a PDF or DOCX, or paste your text. We extract every citation from the body, footnotes and reference list.

02

Resolve

Each citation is checked against Crossref, OpenAlex and arXiv, with per-component matching on title, authors, year and DOI.

03

Review

Get a verdict per citation with the evidence behind it: the closest real source, a component-match table and a live link. You decide.

See the full pipeline and verdict states →

Why not just ask an AI?

Verification beats asking a model to check itself.

The obvious alternatives, side by side. The model that may have invented a citation has no way to verify it, and a style detector was never built to.

Checks each citation against real databases
Tell an AI to verify it
No. It re-reads with the same model that may have written it.
AI-detectors
No. They score writing style, not sources.
Veridicta
Yes. Crossref, OpenAlex and arXiv.
Has ground truth to check against
Tell an AI to verify it
None. It generates plausible-sounding text.
AI-detectors
A statistical model, not a source.
Veridicta
A retrievable source behind every verdict.
Catches a citation that does not exist
Tell an AI to verify it
Unreliable. It can confirm a fake.
AI-detectors
Not what they do.
Veridicta
Yes. Names what resolved and what did not.
Verifies a quote is real and verbatim
Tell an AI to verify it
No. It can hallucinate the match.
AI-detectors
No.
Veridicta
Yes, with a side-by-side diff.
Never flags honest work as AI-written
Tell an AI to verify it
Does not judge authorship.
AI-detectors
Flags it, and is often wrong.
Veridicta
Does not judge authorship.
Produces shareable proof for a client or court
Tell an AI to verify it
Nothing to hand over.
AI-detectors
A probability score.
Veridicta
A tamper-evident certificate and link.
Shows its work with auditable evidence
Tell an AI to verify it
No.
AI-detectors
No. A black-box score.
Veridicta
Yes. Evidence on every citation.
Catches problems before you publish
Tell an AI to verify it
Only if you happen to re-check.
AI-detectors
Flags style, not facts.
Veridicta
Yes. Pre-publication, by design.
Read the full comparison →

The distinction

Why “verification, not detection”

AI-detection guesses whether text was machine-written, and is scientifically unreliable. Veridicta does the opposite: it checks verifiable facts. Either the cited paper exists in a real database or it does not. Every verdict is backed by an artifact you can open and inspect, so you are never asked to trust a black box.

Built to be fair

  • Neutral language: a missing source reads “Unverified, not in the databases searched,” and names them, never an accusation.
  • An “Exists, minor issues” class separates a wrong year from a source that cannot be found at all.
  • Evidence-first: no verdict without a one-click path to its source.

Who it is for

For anyone who signs their name under cited work.

Management & strategy consultants

Client decks and reports that lean on market data, studies and benchmarks.

Expert witnesses & litigators

Filings and expert reports where a fabricated citation can sink the case.

Research & financial analysts

Research notes and due-diligence memos that cite filings and studies.

Audit & advisory firms

Assurance work where citation integrity is part of the brand promise.

Explore use cases →

Trust & security

Designed for work that carries liability.

Credibility you can check, not claims you have to take on faith.

  • We do not train on your documents.
  • Deterministic: a DOI resolves or it does not.
  • 0.0% false “not found” on our held-out benchmark.
  • Every verdict names the databases searched.
Read about trust & security →

Questions

Frequently asked

Is Veridicta an AI detector?
No. AI detectors guess whether text was machine-written, and that is scientifically unreliable. Veridicta does the opposite: it checks verifiable facts. It looks up each citation in real databases and reports what it found. It never judges who or what wrote your document.
Which databases do you check against?
For scholarly and scientific work: Crossref, OpenAlex and arXiv, open and authoritative sources covering tens of millions of records. Case-law checks use CourtListener. Every verdict names the databases that were searched.
What if a source is not in those databases?
We label it “Unverified, not in the databases searched” and name them. That is not an accusation. Books, paywalled reports and grey literature do not always appear in open databases, so we tell you exactly what we could and could not check rather than guess.
Do you train on my documents?
No. We do not train any model on your content. Uploaded documents are processed to produce your result and kept briefly (about 24 hours) so you can reopen a scan, then deleted. See our Privacy page.
Can I share the result?
Yes. Pro issues a tamper-evident Source-Verification Certificate with a shareable link. Anyone you send it to can confirm it was not altered, without creating an account.
See all questions →

See it catch a citation that does not exist.

The demo runs on a sample report in under a minute. No account needed.

Run the demoSee pricing