A research task usually arrives as a single sentence. “Find out what the current rate limits are.” “Compare X and Y.” “Figure out why Z is happening.” The fastest way to feel productive is to translate that sentence into a search query and start reading. That is the move we have been trying to suppress.
The discipline is small but it has a name we use inside the team: write the question first. Not the search query, not the keywords, but the actual research question we are trying to answer. We write it down, in full sentences, before we open the first source.
It looks like overhead. It saves more time than we expected.
What unframed search produces
When we skip the question-writing step, the failure mode is consistent. We open a source, find something interesting, follow a citation, find something else, and an hour later we have a folder of tabs and a vague sense of progress. The findings, when we try to write them, sprawl. We have evidence for things no one asked about and no evidence for the thing they did.
Part of the problem is that good sources are seductive in their own right. A well-written paper or a thorough comparison post is genuinely interesting to read, and reading it feels like research. Sometimes it is research. Often it is reading-adjacent-to-research, which is different. When we do not have the question fixed in our heads, we cannot tell which one we are doing.
The other part is that the original sentence was almost always too compressed. “Compare X and Y” is not a question. It is a topic. It does not say compared on what dimensions, for what purpose, against what success criteria. If we do not unpack it before we start, we end up comparing whatever the sources happened to compare, which is a different question altogether.
What writing the question forces
When we sit down and write the question in full, three things tend to fall out.
The dimensions become explicit. “Compare X and Y” turns into something like: “How do X and Y differ in setup time, ongoing cost, and operational risk for a team of our size?” We have named what we are measuring against. Sources that compare X and Y on other dimensions, however well-written, are now off-topic. We can skim them and move on.
The success criteria appear. What does it look like to be done? Sometimes the answer is “a one-sentence recommendation with two supporting sources.” Sometimes it is “a table of features plus one paragraph of caveats.” Knowing the shape of the answer before we start tells us what to look for. We collect to fill a structure, not to fill a folder.
The unknowns get named. Writing the question forces us to identify what we do not know yet but need to. “We do not know whether Y supports our auth model” is an actionable unknown. We can search for that specifically. “We do not know if Y is good” is not actionable. The framing of the question is what converts vague uncertainty into specific gaps we can close.
How we phrase a research question
Over time we have noticed that the same patterns help and the same ones hurt.
Questions that ask “what is X?” almost always produce sprawl. “What is Kubernetes?” returns the entire internet. We rephrase to “what does our team need to know about Kubernetes to make decision Y?” The decision pins the question down.
Questions that ask “is X good?” are unanswerable. We rephrase to “is X better than the alternatives we are already considering, on the dimensions we care about, for the use case we have?” That is longer, and it forces us to enumerate the alternatives, the dimensions, and the use case. All of which we needed anyway.
Questions that include “best practices” are usually traps. There is no canonical list. We rephrase to “what are the failure modes other teams have reported for this approach, and how do they say they avoid them?” Now we are looking for specific evidence, not consensus.
A good question, by the time we are done writing it, is one that someone could read and tell us we got the answer wrong. If the question is too vague to be answered wrongly, it is too vague to be answered at all.
The cost of the discipline
Writing the question takes ten minutes. It feels like ten minutes of not researching. The pull is to skip it, especially under time pressure, because we can always tell ourselves we will clean up the framing later when we draft the finding.
That has not worked for us. The framing produced under deadline at the end of a research session is shaped by what we found, not by what was asked. Findings written that way drift toward whatever the sources had a lot of, which is usually not the same thing as what the requester needed. We have learned that the question has to be written before the sources have a chance to bias it.
The cost is also that some questions, once written down, look hard. Sometimes impossible. That is information. A question we cannot even phrase precisely is a question we should escalate or renegotiate before we burn hours on it. The discipline of writing the question forward reveals the unanswerable ones early, which saves time even when the project ends differently than we planned.
What this changes downstream
The clearest signal that the discipline is working is that our findings have gotten shorter. When the question is precise, the answer can be too. We do not need three paragraphs of context to set up a single-sentence answer, because the question already named the context. The reader reading the finding does not need a tour through what we considered. They need the answer to the question we both agreed on at the start.
The other change is that we are more likely to come back and say “the question was wrong” early. We used to discover that at the end, when the finding would not quite write. Now we discover it on day one, when we can still rework with the requester. The cost of an hour at the start has been worth the cost of half a day at the end, every time we have measured it.
The start of a research task is the part that determines how it ends. Treating it as a real step, instead of the moment before the real work, is the change that changed the rest of our process.