Every week, a new software startup promises to “revolutionize academic research with generative AI.” They bolt an OpenAI API wrapper to a text box, market it as an “AI citation assistant,” and ask scholars to trust it with their careers.
The consequence has been an unprecedented crisis across scholarly publishing: the silent epidemic of hallucinated citations.
Large Language Models do not know truth; they predict statistical token sequences. When an LLM drafts a literature review, it routinely stitches together real author surnames, plausible journal titles, and completely fabricated DOIs. The resulting citations look authentic to the naked eye, slip right past hurried peer reviewers, and poison the permanent scientific record.
- — Predicts probable words instead of verifying databases.
- — Fabricates plausible DOIs & phantom volume numbers.
- — Requires sending full manuscripts to commercial cloud models.
- — Every identifier resolved against the agency that issued it.
- — Zero hallucinations: if a DOI is not in the registry, it fails.
- — Your manuscript is read in your browser and never uploaded.
Academic integrity demands deterministic ground truth.
That is why we engineered the StrictCite Engine. It contains zero LLM wrappers. It does not guess, synthesize, or autocomplete. Instead, it executes deterministic cross-database reconciliation across the canonical infrastructure of science—Crossref, DataCite, OpenAlex, DBLP, and doi.org.
Every reference string is segmented with strict regular grammar, queried directly against authoritative registries, and reconciled byte-for-byte. If a DOI does not resolve in official publisher databases, it is flagged. If a publication year differs by one digit from the indexed metadata, it is highlighted with an exact discrepancy chip.
Science was founded on the motto of the Royal Society: Nullius in verba—“Take nobody's word for it.” We believe citation verification must operate by that exact same principle: mathematical certainty, immutable databases, and zero hallucinations.