Why is AI's first draft always so generic?
Everything AI knows about your task is what you typed.
"Summarise this report." It's too long.
"Shorter." Now the key point is missing.
"Add the bit about budgets." Long again.
UX researchers at Nielsen Norman Group have a name for this: accordion editing. The answer keeps stretching and shrinking while you try to steer it towards something you never actually described.
Everything you typed. That's it.
AI doesn't know who the summary is for, or that your boss hates bullet points, or that the budget section is the only part anyone will read, or what was said in yesterday's meeting.
When something is missing, it fills the gap with the most average choice. And average reads as generic.
Imagine handing the same request to a smart new hire on their first day. They've never met your team and don't know the project.
Would they know what you want?
If they wouldn't, the AI won't either.
Before:
Summarise this report.
After:
Summarise this report for our CFO. She has two minutes and cares mostly about cost. Start with whether we're over budget, then the top two risks. Under 150 words, no bullet points. You can skip the methodology section.
Same model, very different answer.
Anthropic studied how people actually brief Claude. Only about 1 in 6 conversations said who the result was for. Fewer than 1 in 3 said what format they wanted.
A quick checklist:
If you're not sure what to include, end with:
Before you start, ask me whatever you need to know to do this well.
Hardly anyone does this. In the same study, only about 1 in 10 people talked through the approach before letting it start.
Spend 30 seconds on the brief. It saves five rounds of edits.
Give an example of something good, if you have one.
When in doubt, let it ask you questions first.