Aster Zhao writes: “Across all three platforms, the strongest use case for these AI features is early-stage research: learning unfamiliar vocabulary, assessing whether a document is relevant, or locating a useful passage in a long text. Database embedded tools can lower the barrier for novice researchers because they operate in databases libraries already license. Their limitations are equally clear. None of these tools are well suited for expert, transparent, or reproducible searching. The tools generally support one document, chapter, or viewed result set at a time—not synthesis across a full literature. Output quality is another challenge.”