AI research assistants and citations: common questions answered

Researchers are right to be careful with AI tools. These are the questions that come up most often, with direct answers.

Can I trust an AI research assistant's citations?

Trust should come from checking, not from the tool. A good assistant links each claim to the paper and page it came from so you can read the passage yourself. Constella is built around that step: verify before you cite.

Where do the answers come from?

From the sources in your library, not whatever ranks well online. You choose what goes in, and you can tell a project assistant to use only papers in that project.

How does it find the right paper in a large library?

It searches by meaning rather than only matching keywords, then drills into the exact passage. See how reasoning search works.

What can I save to the library?

Papers and preprints, PDFs and slides, web pages and threads, screenshots and figures, plus typed or spoken notes. Google Drive import is supported.

What if the evidence is mixed?

You can instruct the assistant to say so. Writing guardrails in plain English, such as flagging conflicting results, keeps answers honest.

What is the difference between Instant and Thinking modes?

Instant handles quick lookups, like which paper used actigraphy. Thinking handles synthesis, like comparing study designs across three papers.

Does the AI make decisions for me?

No. It surfaces sources and drafts answers. You verify and decide. Details are on the trust and controls page.

Can my lab share a project?

Yes. A shared project gives everyone the same memory and assistant, which helps keep the method consistent. See Constella for researchers.