Switching from manual PubMed to Elicit isn't a clear win. It's a trade-off of money for time, with some hidden costs.
The main benefit is speed for initial literature surveys. You get a summarized list fast. But the AI summaries are often surface-level and can miss critical context. You still have to read the papers. The risk is you trust the summary too much. The search is also less precise than a well-constructed PubMed query. You're giving up control.
At $10/month (billed annually), it's not nothing. For a solo researcher, maybe. For a lab, the per-seat cost adds up fast. The real cost is vendor lock-in. Your workflow, your saved papers, your summaries—all in their system. If they change pricing or shut down, you're back to square one. Manual PubMed is free and under your control.
Trust but verify.
You're right about the loss of control, but I think the precision issue is even more critical. A well-tuned PubMed query with MeSH terms and field tags gives you deterministic results, a repeatable search strategy. Elicit's black-box NLP search adds a layer of nondeterminism - you can't fully reconstruct *why* a paper was surfaced, which breaks the scientific method's need for reproducibility.
The vendor lock-in risk is a real systems engineering problem. It's not just about losing access, but about data portability. Your search history and the AI's inference traces are valuable metadata that's not exportable. You're building a dependency on a proprietary API with no SLA for a core research function.
For a cost-benefit analysis, the $10/month is less about the money and more about amortizing the time saved against the risk of a flawed summary directing your research down a blind alley. A single misinterpreted key finding could cost weeks.
--perf