K: Key Output: Define What Good Looks Like

Start with an example

A school counselor asks AI to turn her notes from a parent workshop into a handout.

Without Key Output: She gets four long paragraphs in a formal tone, including a statistic about college acceptance rates that wasn’t in her notes. She has to reformat it, soften it, and fact-check it.

With Key Output:

• Format: One page. A short intro, then 5 numbered tips, then a “Questions? Contact” line.
• Tone: Warm, encouraging, plain English. Parents should feel supported, not lectured.
• Quality check: Use only information from my notes. If a tip needs a fact I didn’t provide, flag it as [CHECK] instead of filling it in.
• Include: My office hours and email at the bottom.
• Avoid: Statistics, jargon like “postsecondary,” and anything that sounds like a sales pitch.

The result is a handout she prints the same day. The one fact gap is clearly flagged, so she fixes it in 30 seconds instead of discovering it after parents read it.

Key Output is a quality standard

This is the most misunderstood part of R.I.C.K. Most people think “output” just means format: a list, a paragraph, a table. That’s part of it. But Key Output tells the AI what good looks like for this specific task. You’re not only telling it what to produce; you’re telling it how to produce it well.

Five fields live here.

1. Format: how the answer should be structured

Bullet list, numbered steps, table, paragraph, email, memo, script, summary, comparison chart. Name it. Without a format, the AI makes its own decision, which often doesn’t match what you need.

You need…	Ask for…
Something to scan quickly	Bullets, most important first
A comparison	A table with named columns
Steps to follow	A numbered checklist
Something to send	An email with a subject line
Something to say out loud	A script with short sentences
Data for a spreadsheet	A table or CSV with specific columns

Add limits too: “under 120 words,” “exactly 5 bullets,” “one sentence per point.”

2. Tone: how it should sound

Formal. Conversational. Direct. Plain English. Encouraging. Urgent. Warm but professional. Matter-of-fact.

Tone isn’t decoration. It decides whether the reader trusts the message, keeps reading past the first line, or feels talked down to. Name the tone that fits your audience and purpose, and combine words for precision: “confident, not salesy,” “warm but direct.”

3. Quality check: what makes this response actually good

This is the field most people skip entirely, and it’s the one that separates a useful output from a great one.

A quality check is a standard the AI should hold itself to before finishing. It answers: How will I know I did this well?

• “Before finishing, check that every action item has an owner and a deadline.”
• “Flag anything that needs human verification before use.”
• “If anything is unclear or unsupported, say so explicitly instead of filling the gap.”
• “Review your output for any claims that could create legal or compliance problems.”
• “Read it back as a first-time reader. If anything requires knowledge they don’t have, simplify it.”

The quality check is an instruction to think before delivering, not just produce. It builds self-review into the prompt itself.

4. Include: the must-haves

What can’t be missing? A call to action. A risk flag. A summary sentence. A specific section. A recommendation. Contact information. Name what must be there, and the AI is far less likely to leave it out.

5. Avoid: the must-nots

What should not appear? Corporate jargon. Passive voice. Legal claims. Filler phrases. Invented statistics. Exclamation points. Opinions stated as facts. Clichés like “game-changing” or “in today’s fast-paced world.”

The must-nots matter as much as the must-haves. Every AI has habits, like overused phrases, too many adjectives, and a love of “Great question!”, and the Avoid list is how you switch them off.

Think about it: If you had to write an Avoid list for your own writing, what would be on it? What phrases do you lean on when you’re not sure what to say?

A worked example: the full Key Output

Jordan volunteers with a food pantry and needs a volunteer sign-up post for the community Facebook group.

KEY OUTPUT:
- Format: A Facebook post, 3 short paragraphs, then the sign-up details as a 4-item list.
- Tone: Warm, community-minded, direct. Like a neighbor asking for help, not an organization making an announcement.
- Quality check: Confirm the post answers: what we need, when, where, how long a shift is, and how to sign up. If any of those is missing from my notes, mark it [MISSING].
- Include: The sign-up link and a thank-you to last month's volunteers.
- Avoid: Guilt-tripping language, statistics about hunger, hashtags, and exclamation points.

The AI’s draft flags [MISSING: shift length], because Jordan forgot to include it. Without the quality check, the AI would likely have invented “2-hour shifts.” Jordan adds the real answer, 3 hours, and posts it.

Think deeper: why “check your work” changes the answer

When you ask a model to run a quality check, you’re changing what it predicts. A prompt that ends with a list of standards makes it more likely that the answer is shaped to meet them, and it gives the model a natural moment to compare its draft against your requirements before it finishes.

But notice the limit. A quality check helps the AI catch things it can see: a missing section, a word count, an action item without an owner. It can’t reliably catch things it doesn’t know are wrong, like a fact it believes is true. That’s why the best quality checks often ask the AI to flag uncertainty rather than resolve it: “mark it [CHECK],” “say so explicitly,” “note anything that needs human verification.” You’re turning the AI’s blind spots into a visible to-do list for you.

Questions to sit with:

• Teachers give rubrics before assignments, not after. How is a Key Output like a rubric, and what happens to students (and AI) when the rubric is missing?
• If you can’t describe what “good” looks like before the work starts, is the problem the AI, or that you haven’t decided yet?

Practice: upgrade one prompt

1. Pick a prompt you’ve already used.
2. Add a Format and Tone that match your real audience. Run it.
3. Add a Quality check that asks the AI to flag anything unsupported. Run it.
4. Add Include and Avoid lines. Run it once more.
5. Compare all versions. Which field saved you the most editing?

KEY OUTPUT:
- Format: [bullet list, numbered steps, table with named columns, email with subject line, script, memo]
- Tone: [plain English, direct, warm but professional, encouraging, formal]
- Quality check: [what the AI must verify before finishing, e.g. 'every action item has an owner and a deadline']
- Include: [the must-haves, e.g. a clear next step, a risk flag, a summary sentence]
- Avoid: [the must-nots, e.g. jargon, invented statistics, exclamation points]