No feature currently in ReferenceLib uses a generative language model. The question decoder matches command verbs and Bloom's taxonomy stages using fixed rules, not AI. Evidence search and reference verification match your question or claim against OpenAlex, Crossref and DOAJ's own published metadata. Citations are produced by a deterministic citation-formatting engine (citeproc-js), not generated text. Nothing you read on this site today has been written or paraphrased by a language model on your behalf.
A future feature is planned that would use a language model for one narrow purpose: checking whether a claim you have already written yourself is supported by evidence the free search has already found, and stating plainly what is missing for it to meet the required academic standard. Per the academic-integrity policy, it will not write or rewrite your argument, and will not be built to disguise its own use.
This page will be updated if and when that feature is built, so it always reflects what the tool actually does, not what is planned.