The evidence-grade research stack

All four tools here have a genuine free tier, so a working configuration costs nothing while you evaluate it. The 03 COSTS table above breaks down each tool's pricing. The meters that climb with use sit on the research side: Perplexity's report depth, Elicit's extraction across papers and NotebookLM's daily generation caps, with Claude's usage on top once synthesis turns heavy. Extraction volume tends to bite first as projects deepen, and each upgrade is an independent decision, not a bundle.

Rarely. Perplexity is the core: every engagement starts with sourced discovery, and many run on it alone. The numbered steps are the workflow order and double as the adoption order, though need sets the pace. Add Elicit when the question turns academic and structured extraction across papers is what scrutiny demands; NotebookLM once evidence accumulates and wants one interrogable home; Claude when the deliverable is a written argument that will be judged, not a summary.

Then you already own the discovery layer: sourced, cited answers and deep reports from the live web. Keep it there and defer the rest until the work demands it. What this stack adds is what Perplexity does not reach: Elicit's structured extraction across the academic literature, NotebookLM answering strictly from the corpus you have gathered with passage-level citations, and Claude's synthesis depth for a deliverable that will be judged. Discovery stays; the chain adds rigour behind it.

Perplexity and NotebookLM both answer research questions with citations, the one real overlap here. The dividing rule is the source. Perplexity searches the live open web: current answers and deep reports on what is out there. NotebookLM answers strictly from the corpus you have already gathered, with passage-level citations. Discovery leans on the first, grounded interrogation of your own evidence on the second. The Perplexity and NotebookLM comparison, linked above, covers the finer calls.

Two signals, both about scope rather than cost. The first: the corpus stops being assembled fresh per engagement and becomes a standing body of knowledge that a whole team must search. The second: retrieval has to reach across many systems and respect who may see what, with answers governed and owner-verified rather than uploaded by one analyst. Both point to the world-class tier, the governed knowledge ops stack.

The strictest member sets the floor. On personal tiers, treat these as semi-public: Claude's consumer plans may train unless you turn that setting off, and Perplexity's unconditional no-training guarantee sits on Enterprise (Pro subscribers can match it by switching off the retention toggle in settings). Client material belongs on Claude's Team or Enterprise plan and Perplexity's paid tiers. NotebookLM puts uploads on Google servers, so check your Workspace terms first; keep Elicit to published papers, not sensitive files.