A grounded, cited conversation#
Ask answers questions in plain language and grounds every answer in the corpus. As an answer streams in you see the stages it moves through - interpreting the question, retrieving sources, writing and checking - then the finished answer with numbered citations you can click straight through to the source passage. It also shows how it read your question, as an "Interpreted as..." line above the answer.
Follow-up questions keep the context of the conversation, so you can drill in without restating everything each time.
Reading the answer#
Under each answer you get the full picture of what it stands on:
- The sources it used, and a note of the years they span (for example "Cited sources: 2016-2019"), so you can see how current the material is.
- An evidence list - each source with its matched passage and a relevance score.
- Also retrieved - relevant passages the answer did not lean on, kept visible so nothing is hidden.
Want to see the working? Journey through the context walks you through the passages the answer was built from, and adds a short AI verdict on each source - whether it Supports, is Partial or is Not relevant to your question. (That judgement is only worked out when you open the journey, so it is never generated for answers you do not choose to dig into.) Administrators additionally see Show the pipeline, a developer view of the retrieve, write and check stages behind it.
When an answer comes back thinly grounded, Ask offers to re-answer it deeply - re-running your question against the full text of the matching documents rather than the retrieved passages alone - so a weak first pass has a one-tap path to a stronger one.
Sessions and your research trail#
Each conversation is saved as a session in the sidebar. Start a new session, rename one, reopen an earlier one, or delete one you no longer need. You can Export a session as a Word-compatible document to keep the whole research trail: questions, answers, sources and the quality scores. Ask exports to Word only; for a PDF, see Exporting your work under Getting started.
Deep research#
Turn on Deep research to have the portal first map your question into focused sub-questions, research each of them, and then answer with full-document grounding. It is slower but more thorough for broad or multi-part questions. Questions about risk, safety, effects or comparisons are broken down automatically so decisive passages are not missed.
The list of sub-questions shown above a deep answer is the multi-step view of that research: each one was retrieved and answered on its way into the final answer, so you can see how the question was broken down and which parts the corpus covered. There is no separate "agentic" surface - the old /agentic address simply opens Ask.
Feedback and watches#
Mark an answer Helpful or Not helpful to signal how well it landed. Watch this question to keep an eye on it - the portal re-checks it daily and flags it in Search when new results turn up. See Watch a search for how watches work.