Most People Give AI Fewer Than 9 Words. The Prompts That Work Use About 21. | Edition 371
Edition 371 — AI at Work, Part 1 of 12. Brief AI like a new hire: the goal, who it's for, what good looks like, what to skip.

This is Part 1 of AI at Work, Simplified, a 12-part series. Each week, one skill you can use at your desk that day, in ChatGPT, Claude or Gemini.
Most people use AI at work the way they use a search box. They type a short question, get a bland answer, and decide AI is overrated.
Google has watched this happen at scale. In its prompting guide for Gemini in Workspace, it says the most fruitful prompts "average around 21 words with relevant context, yet the prompts people try are usually less than nine words."
Those extra words aren't filler. They're the brief: what you'd tell a smart new colleague before handing over the task.
Why a short question gets a generic answer
AI doesn't know your company, your customer, your boss or your deadline. When you leave all of that out, it fills the gaps with the most average guess it can make.
Anthropic's own prompting guide puts it plainly: "Think of Claude as a brilliant but new employee who lacks context on your norms and workflows. The more precisely you explain what you want, the better the result."
OpenAI's guide for ChatGPT at work says the same thing in different words: "Outline what you want, who it's for, and why it matters."
Anthropic has also measured it. Its January 2026 Economic Index report found that the sophistication of what people type and the sophistication of what Claude writes back are "nearly perfectly correlated." In its words: "While Claude is able to respond in a highly sophisticated manner, it tends to do so only when users input sophisticated prompts."
Put simply: the answer can only be as good as the brief.
The four-line work brief
You don't need a template or a course. You need four lines.
- The goal. What is this for, and what happens next? "I need to get three vendors to quote by Friday" beats "write an email."
- Who it's for. Your boss, a customer, a new hire, a board. Say what they already know and what they care about.
- What good looks like. Length, format and tone. "Five bullets I can paste into Slack" or "one page, plain English, no jargon." If you have a past example you liked, describe it.
- What to avoid. Words, claims or topics that are off-limits. "Don't promise a delivery date." "Don't mention pricing."
Here's the difference on an everyday task.
Before: "Write an email about the new expense policy."
After: "Write an email to our 40-person sales team about the new expense policy. Goal: get receipts submitted within 7 days so finance can close the month. Audience: busy reps who skim on their phones and dislike admin. Good looks like: under 120 words, the one rule in the first line, a friendly tone. Avoid: anything that sounds like a reprimand, and don't mention who was late last month."
The first version gets you a generic policy announcement. The second gets you something you could send after a light edit.
Before you send a brief, use Anthropic's "key test": "Show your prompt to a colleague with minimal context on the task and ask them to follow it. If they'd be confused, Claude will be too."
| Role | Typed this | The brief added | What changes |
|---|---|---|---|
| Sales manager | "Write a follow-up email" | Goal: book a second call. Audience: a CFO who raised price concerns. Avoid: discounts. | Speaks to cost, asks for one specific time, no discount offered |
| Office manager | "Make a meeting agenda" | Goal: decide the office move. Audience: 6 department heads. Good: 30 minutes, decision first. | Short agenda built around one decision, with time boxes |
| HR lead | "Write a job post" | Audience: experienced bookkeepers. Good: 200 words, real duties first. Avoid: "rockstar", "fast-paced". | Concrete duties up top, none of the cliché filler |
| Freelancer | "Summarize this report" | Goal: pitch a client. Audience: their marketing head. Good: 3 bullets tied to their goals. | A summary angled at the client's problem, not a neutral recap |
A better brief is not a guarantee
Briefing well makes AI more useful. It doesn't make it right.
The best evidence comes from a Harvard and Boston Consulting Group field experiment with 758 consultants. On tasks AI handles well, those using it completed "12.2% more tasks" and finished "25.1% more quickly on average," with higher-quality work. But on a harder task chosen to sit outside what AI does well, the AI users "were 19% less likely to produce correct solutions."
So the brief gets you a stronger first draft. Your judgment still decides whether it's correct.
Check it before you send it
- Did it answer the goal, or just the topic? Reread your first line of the brief, then the output.
- Is every fact, number and name real? Anything you didn't give it, verify.
- Would you sign it? If not, tell the AI exactly what's off and ask for another pass.
Try this today
Pick one task you'd normally type as a one-line question. Write the four lines first: goal, audience, what good looks like, what to avoid. Then compare the answer with what you usually get.
Next week, Part 2: what you can safely paste into AI at work, and what you can't.