The research you'd do — if you had the hours.
A background research agent that reads what you approve — specific documents or your whole knowledge base, plus the live web if you allow it — and comes back with a cited, structured report.
Know what's supported, see what's disputed, remember what matters.

Watch a research run, start to finished report.
- 1Set it upPick an outcome and a depth, type your question, choose what it reads. Web search is off until you switch it on.
- 2It runs in the backgroundThe run works through its stages. You can leave: the bell shows when the report is ready.
- 3A report with its sourcesA TL;DR, sections with citation badges, the sources used, and the report filed in your library.
Pick an outcome and a depth, type your question, choose what it reads. Web search is off until you switch it on.
- What you setAn outcome (Understand, Decide, Compare, Verify or Create), your question, a depth, and what it may read: specific documents or the Entire Knowledge Base, plus current web sources if you switch them on for that run.
- Three depthsQuick (up to 8 minutes, up to 25 sources), Focused (up to 20 minutes, up to 50 sources), Thorough (up to 45 minutes, up to 100 sources). Credits are set aside when a run starts and given back if it fails, times out or is cancelled.
- The reportAn "As of" date, a TL;DR, folding sections, inline citations — E1-style labels for your documents, numbers for web pages — and a Sources used list. Once you open a finished report, it is filed into your library automatically — into the collection My Research keeps in My Documents — and trained, so later chats can use it.
What is My Research?
My Research is where CorpGPT researches for you in the background. You ask a real question, tell it which sources it may use — specific documents or your whole knowledge base, plus the live public web if you approve it — and it works for minutes, then delivers a structured report with a TL;DR, evidence with inline citations — E1-style labels for your documents, numbers for web pages — and the methodology and limitations sections it is instructed to close with. It's the only feature that searches the web; everything else works from the documents in your library. Once you open a finished report, it is filed into your library automatically — into the collection My Research keeps in My Documents — and trained, so later chats can use it. Over time, your reports build into a personal, searchable research bank.
Say what you actually need
Pick one, and the whole report is shaped to it.
Understand
A grounded explainer — plus what's genuinely contested, not just the tidy consensus.
Decide
A recommendation, an options-and-trade-offs table, the risks, and how confident to be.
Compare
Alternatives weighed against your own criteria in a real table, then a verdict.
Verify
The verdict up front, then the evidence for it and the evidence that pushes back.
Create
Need it written? Get the deliverable itself, built from the evidence and written to cite its sources.
Choose how deep to go
Depth decides how hard it works and how wide it searches — not which answer you get. Times and source counts are upper limits.
Quick
The core answer and the essential evidence, fast.
Focused
DefaultCompares credible viewpoints and weighs the trade-offs.
Thorough
Tests assumptions, surfaces conflicts, and hunts for the gaps.
You decide what it reads
Specific documents or your whole knowledge base — not both: switch Entire Knowledge Base off first to pick documents — plus the live web if you approve it. At least one source is required, and the live web stays off until you switch it on for that run.
Specific documents
Pick individual trained documents. The button stays greyed out while Entire Knowledge Base is on, so switch that off first. The report is instructed to treat the documents you pick as the primary evidence.
Your whole knowledge base
Every trained document (Entire Knowledge Base in the app). On by default, so a first run never lacks a source; switching it on clears any documents you picked.
The live public web
Off by default, approved one run at a time. The only part of CorpGPT that searches the web.
A report that shows its work
Instructed never to invent or pad a source, and to close the same honest way.
- A TL;DR up top, then clean, collapsible sections
- Evidence with inline citations — E1-style labels for your documents, numbers for web pages
- Each [E1] label links to the document's name in the Sources used list; each number links to the web source
- Comparison tables in Decide and Compare reports
- Instructed to end with Methodology, Limitations, Open questions and Follow-ups
Each report feeds the next question
Once you open a finished report, it is filed into your library automatically — into the collection My Research keeps in My Documents — and trained, so later chats can use it. It isn't a one-off answer but searchable, chattable evidence you keep (and can delete). Ask a question in chat weeks later and it can draw on what you've already researched. Over time, My Research becomes your own research bank — and the more you use it, the more your workspace has to draw on.
Why My Research
Say what you actually need
Understand, Decide, Compare, Verify or Create — the report is shaped to the answer you asked for.
You choose the sources
Specific documents or your whole knowledge base, plus the live web if you approve it. Live web is off until you do.
Claims you can trace
Inline citations — E1-style labels for your documents, numbers for web pages; each label links to the document's name in the Sources used list. It is instructed never to pad or invent a source.
Honest by design
Every report is instructed to close with its methodology, limitations, open questions and what to look at next, and to point out web sources older than 18 months.
Walk away while it works
Runs continue in the background — close the tab and the report still finishes. The notification bell shows it when you come back (and, if CorpGPT is open in another tab and you allow it, so does a desktop alert).
Credits you can't lose
Credits are set aside when a run starts and given back if it fails, times out or is cancelled.
Your research bank grows
Once you open a finished report, it is filed into your library automatically — into the collection My Research keeps in My Documents — and trained, so later chats can use it. Delete it from My Documents if you don't want to keep it.
How it works
Write the question
Pick your outcome and depth. Plain language is fine — it preserves your intent.
Approve the sources
Specific documents or your whole knowledge base, plus the live web if you approve it — for that run only.
Get a cited report
Read the TL;DR and follow the citations. Once you open a finished report, it is filed into your library automatically — into the collection My Research keeps in My Documents — and trained, so later chats can use it.
Runs in the background — leave the page, and the notification bell shows you when your report is ready.
Frequently asked questions
Does My Research search the internet?
Only if you approve it for that particular run. Live web research is off by default, and My Research is the only part of CorpGPT that searches the web — chat and generated content work from the documents in your library (which can include web pages, YouTube links and RSS feeds you choose to add).
How long does research take?
It depends on the depth you pick: up to 8 minutes and 25 sources for Quick, up to 20 minutes and 50 sources for Focused, up to 45 minutes and 100 sources for Thorough. You don't have to wait on the page — it runs in the background, and the notification bell shows you when the report is ready.
What happens to my credits if a run fails?
You get them back. Credits are set aside when a run starts and released if the run fails, times out or you cancel it — you only pay for research that produces a report.
Can I trust the answer?
The report is instructed to cite every claim that matters with inline citations — E1-style labels for your documents, numbers for web pages; each label links to the document's name in the Sources used list — so you can check it yourself. It is also instructed to close with its methodology, limitations, open questions and follow-ups, and to point out web sources older than 18 months as possibly out of date. A run that hits its depth's length limit stops early with no report, and the credits set aside for it come back.
What if the research can't find enough?
The report is instructed to say so plainly rather than guess, and to show what's supported and what's disputed. If a run fails, times out or is cancelled, no report is produced and the credits set aside for it come back.
Does my research end up in my knowledge base?
Yes. Once you open a finished report, it is filed into your library automatically — into the collection My Research keeps in My Documents — and trained, so later chats can use it; you can delete it from My Documents at any time. Reports are also kept in My Research, and OpenAI, which runs the research, keeps each request and response under its own data-retention terms (see our Privacy Policy).
Can I research my own documents and the web together?
Yes. Choose specific documents or your whole knowledge base, then approve the live web for that run. The report is instructed to treat documents you picked as the primary evidence.
Put a real question to it
Free to start, no credit card. Ask something that matters and get a cited report you can stand behind.
Start free