Think of OpenAI's ChatGPT as a sharp new coworker on their first day. They are quick and willing, but they were not in your meeting, they do not know your team, and they will fill any gap you leave with a reasonable guess. A one-line request gets you a one-line-quality guess.
OpenAI's own prompting guide lists three steps: outline the task (what you want, who it is for, and why it matters), give helpful context, and describe your ideal output by telling ChatGPT how you'd like the response. In practice that means pasting in the real material (your notes, the email, the draft) and saying what done looks like, such as a bulleted list grouped by owner with due dates.
Here is the difference in one example. A one-line request: "Summarize these notes." A full request: "Turn these meeting notes into an action list for my team channel, grouped by owner, with due dates where the notes give one, and mark anything without an owner as unassigned." The second one tells ChatGPT the output, the reader and what to do with gaps.
If the first answer misses, say what's wrong in the same chat and what you want instead. That is a normal part of the work, and starting over throws away the context you already gave.
The honest limit: ChatGPT only knows what is in the chat. If your notes say nobody took a task, it may still invent an owner to make the list look complete. That is why the last step of every task in this course is a check against something outside the chat, here your own notes.
Give ChatGPT the real material, say who it is for and what done looks like, and correct it in the same chat. Then check the result against your own source, here the meeting notes, because ChatGPT will fill gaps with confident guesses.