Web search and Deep Research

When a question needs current information or real literature, the assistant searches for it, reads the pages and papers it finds, and shows you every source under the answer.

Web search is always on

There’s no switch to flip. If a question calls for it, the assistant searches. Ask about anything recent, name a website, or paste a link, and it will go and look.

In the reply you’ll see:

• a Web search card listing what it found, and a Fetched pages card when it read a page in full;
• a Sources button under the answer — click it for every page used, and click a source for the exact passage.

Web search comes in two depths. The default is a thorough, multi-step search; for a quick fact the assistant uses a faster one. You don’t need to choose.

Useful prompts:

• “What’s the latest guidance on GLP-1 agonists and cognitive decline?”
• “Read https://example.com/report and summarise the methodology section.”

Academic search

Ask for papers and the assistant searches the scholarly databases directly — CrossRef, OpenAlex, Semantic Scholar, arXiv, PubMed and DOAJ by default, with specialist sources for biomedical topics (Europe PMC, ClinicalTrials.gov), education (ERIC), patents (USPTO) and grants (NIH, NSF). You can ask for a year range or peer-reviewed results only.

Each paper appears as a card labelled with its type (peer-reviewed, preprint, clinical trial, patent…) and three buttons:

• + Cite — add it to this chat’s citation list.
• + Add — save it to your Library.
• + KB — add it to the current project’s knowledge base.

The assistant can also read a paper’s full text (where an open-access copy exists) and check that a citation is real before it gives it to you.

Useful prompts:

• “Find peer-reviewed papers on microplastic bioaccumulation since 2022.”
• “Do a literature review on CA1 NMDAR knockouts and spatial memory, and put it in a table.” (This creates a literature review table — one row per paper, verified against the registries.)

Citations

The header of a research chat has a Citations panel: every paper cited in the conversation, with DOI links and the claims each one supports. From there you can export to BibTeX, RIS or CSV, or add papers to your Library.

A source review badge next to a reply’s actions summarises how well the answer is backed up: how many claims were corroborated, how deep the sources went, and which claims were flagged. Use Improve with feedback to turn a flagged claim into a follow-up. It’s an automated check — verify anything important yourself.

Every citation is verified against the publisher registries. Free and Starter plans have a daily allowance of verifications; past it, citations are still recorded, just marked as not yet verified. See Plans & limits.

Deep Research

For a question that deserves a full investigation, turn on Deep Research — the telescope beside the tools button — before you send.

The assistant answers as usual. Then, once the answer is on screen, it runs a second, independent research pass over the whole question: it must open primary sources rather than rely on snippets, and it then compares what it found with the answer it already gave. Anything the second pass adds is edited into the report; nothing already written is removed or rewritten.

While this happens the status line reads Verifying the findings with an independent research pass. It can take several minutes after the visible answer has finished, and it uses considerably more of your budget than an ordinary turn.

Deep Research is available on Starter and Pro plans.