how to brief claude like a colleague.

elisabeth hitz · june 18, 2026 · 5 min read

when claude does the work instead of just answering, the way you ask changes. it is less like prompting a chatbot and more like briefing a smart colleague who is about to disappear into a task and come back with it done. the better the brief, the better what comes back. here is what a good one names.

this picks up where what claude cowork actually is left off: you know what to delegate, now here is how to ask.

name three things

a strong brief covers all three. leave one out and you are usually fixing it later.

• the deliverable. "a one-page brief," "a single slide for the qbr," "a ranked list with notes." being specific about format and length saves you a regenerate.
• the inputs. which folder, which channels, which date range, which file. it is only as good as the context you point it at.
• the nuance only you know. the one or two things your expertise sees that it can't guess. "this is for the exec team deciding the new pricing tier, lead with the recommendation, and flag anything we can't verify." that sentence is the difference between generic and useful.

if you are coming from a chatbot, this feels like more work upfront. it is. the trade is that the back-and-forth happens before claude starts, instead of across five rounds of "actually, can you also."

answer the questions specifically

most non-trivial tasks open with a clarifying question or two. that is not friction, it is claude closing the gaps before it commits, so it does not return an hour later with the wrong thing. most are easy: it offers a couple of options, you pick one. when none fit, answer in your own words. the more specific your answer, the less rework at the end.

steer while it runs

watch the plan as it works. if it picks the wrong source, the wrong format, the wrong tone, jump in. you do not have to wait for it to finish and then regenerate, the way you would with a chatbot. correct it mid-task and it picks up from where it was. the cost of a redirect is low, so use it early.

review it like a draft from someone new

when it is done, the deliverable is the artifact, not the chat, and it is sitting in your folder. read it the way you would read a draft from someone you trust but do not yet fully know, with your judgment on. three checks, especially the first few times:

• does it meet the actual objective, or something subtly different from what you asked?
• are the facts right? ask claude which docs it pulled from, then check them yourself.
• does anything sound made up? a date, name, or quote you can't trace to an input is a flag, not a feature.

if it is mostly right, tell claude what to change rather than starting over, it edits faster than it regenerates. if it is wrong in a load-bearing way, the brief was missing a load-bearing piece of context. point it at the missing piece and let it adjust. and keep the approval gates on for anything that sends or shares, the same discipline from using cowork safely.