Research mode
For questions that need sources rather than an opinion. It gathers material, tracks where each claim came from, and shows the sources beside the answer.
Turn it on
/research on- Switches the mode on in a chat.
/research <url>- Adds a page you want it to read.
The rule
The rule underneath it is mechanical, and that is the point.
A claim counts as supported only if its quote is found verbatim in the copy of the page we fetched and stored. Not “a second model agreed”, but a string comparison against the bytes retrieved.
The quote
tracks where each claim came from
found in
The stored copy
For questions that need sources rather than an opinion. It gathers material, tracks where each claim came from, and shows the sources beside the answer.
- A model can
- write a confident citation for a page that says no such thing.
- A model cannot
- make an invented quote appear in text it did not write.
What follows
- Check it instead of trusting it. Claims are linked to their sources, so you can go and read what they rest on.
- A fetch made for you follows your policy. Pages are fetched under the same network rules as your container’s outbound traffic, and every fetch is audited.
What it costs
Slower and dearer than a normal answer. Use it when the citation matters.
When you want a whole report Deep research