After twenty sources, the team knows more about leftover food, but not yet what it should do differently at the next party. Thorough research can still miss its actual task. The brief needs a decision that the information is meant to feed into.
OpenAI’s documentation on research in Work describes Deep Research for investigations across several sources and a report that can be checked. The question, the scope and the desired result belong in the brief. Access differs by account and interface. The following example of our own explains the research contract; no corresponding run was carried out.
Name three courses of action explicitly
A fictional club wants less leftover food at a party. Under discussion are more binding advance registrations, smaller serving sizes and shopping planned more closely around demand. At this stage these are alternatives, not recommendations that have already been proven.
Describe the setting: a single event, a small volunteer team and limited means of collecting data. An approach from a large canteen may offer interesting pointers without fitting the club unchanged. The question of transferability therefore belongs in the research from the start.
Set the criteria before searching
A comparison needs the same questions for every alternative. Our example looks at the information needed, the preparation effort, possible drawbacks and the state of the evidence. These criteria prevent one approach from winning on a single attractive figure while its preconditions remain invisible.
Compare advance registration, smaller serving sizes and shopping planned closer to demand
for a small, fictional club party.
Use original sources that are as accessible as possible.
For each approach, set out: preconditions, effort, drawbacks, evidence, open questions.
Explain the limits of transferring findings from other institutions.
State which of our own information most strongly influences the decision.
No blanket ranking of winners.This is a starting brief of our own. You can narrow the scope further, for instance to one particular catering process. A smaller question is often easier to follow than a report that is meant to explain purchasing, nutrition, logistics and behaviour all at once.
Keep observation and application apart
A source may examine one specific serving situation. It does not automatically follow that the same effect will occur at a different event. So separate what the source says from your idea for applying it and from the assumption that idea requires.
A helpful comparison line contains: “The source deals with …”, “For our party this could mean …” and “For that we would need to know …”. This lets the team discuss a plausible transfer without reading it as a result that has already been measured. Conflicting findings, too, should remain identifiable by their respective settings.
Open the original passages for the central statements. Check whether drawbacks, exceptions and conditions have been preserved in the report. A detailed summary can repeat the same source indirectly several times; the number of links alone does not yet describe an independent body of evidence.
Spot the missing information on your side
Perhaps every approach depends on a more reliable number of participants. In that case, the next step towards insight is not necessarily more general literature. The team could first establish which confirmed registrations or quantity figures it already has.
Note which additional information would change a decision. “Expected number of participants unclear” is more precise than a general need for more research. It also shows when an investigation has done enough for now: the remaining gap is specific and can be closed outside the literature.
A decision paper rather than a final verdict
The report may set out different priorities. One team may accept more preparation; another needs a process that works with only a few helpers. The AI does not decide on its own what effort is reasonable for the club.
So keep the alternatives, preconditions and open points together in a clear final version. Also name the person responsible for the decision and the authoritative version of the sources. Whether Deep Research is available in a specific account and which sources can be reached is something you check in your actual Work access. The practical yardstick remains a decision that can be answered, not the length of the research delivered.
Sources and documentation status: 25 October 2026.
