# The first thing your examiner reads is your bibliography.

> The quick check matches every entry against Crossref, OpenAlex, swisscovery and the German National Library.

Per entry the check returns: found (Crossref, DNB), a metadata mismatch (e.g. "Wrong volume"), or not found. 2 minutes · no sign-up.

[Check yours for free](https://app.acurio.ch/quick-check?lang=en)

## 01 // The problem

**Fabricated references are no longer a fringe case.**

| Finding | Value | Source |
| --- | --- | --- |
| Journal articles with fabricated references | 2023: 1 IN 2,828 → END OF 2025: 1 IN 458 (12× in two years) | Topaz et al. 2026 · The Lancet · 97m citations checked |
| fabricated citations across 53 papers at NeurIPS 2025. Despite peer review. | 100 | NeurIPS 2025 · GPTZero audit |
| of the 4,046 cases found drew no response from the publishers. | 98.4 % | Ansari 2026 · preprint, arXiv |

AI detection demonstrably does not work. You only recognise the fake against the original.

## 02 // What the quick check cannot see

**A source can exist and still prove nothing.**

**Quick check · the form · free** — Sweller, J. (2011). Cognitive load theory. Psychology of Learning and Motivation, 55, 37-76. → Found · Crossref · DOI resolves

**Acurio · the content** — Cognitive load has no effect on retention (Sweller, 2011). → Original passage says the opposite · p. 41

**3 in 10** citations have a weak spot: transposed numbers, wrong page, overstretched claim. Whether written by humans or by AI. (Observed across Acurio check runs)

## 03 // The content check

**Acurio reads your sources.**

- Original passage
- Claim
- Page number
- Verdict

Chapter 2 · page 14: High cognitive load reduces retention (Sweller, 2011), while adaptive feedback compensates for that effect (Weber, 2019).

**Two minutes. Then you know.**

[Check your bibliography now](https://app.acurio.ch/quick-check?lang=en) — Free · no sign-up · no credit card

---

Source: https://acurio.ch/en/bibliography-checker