Mistakes, Iteration, and Your Prompt Library

Start with a familiar story

“I tried it. It didn’t work. AI is useless.”

You’ll hear this a lot. Almost every time, what actually happened is that someone wrote one prompt, got one weak answer, and quit. They never found out what a second or third round would have produced.

Strong prompts often take two or three iterations to land. This lesson covers the mistakes that cause weak results, how to fix them quickly, and how to save what works so you get it right on the first try next time.

The six most common mistakes

Mistake 1: One-line prompts for big jobs

“Write my business plan.”

A business plan is 20 to 30 pages. One line gives the AI no industry, no model, no audience, no goals, so you get a generic template that fits every business and serves none.

Fix: Break it down. Use R.I.C.K. on one section at a time. A strong, specific chapter is more useful than a generic full draft.

Mistake 2: Skipping the audience

“Write a marketing email about our new service.”

To whom? A new prospect? A loyal client? A skeptical board member? Each needs a different message, tone, and call to action.

Fix: Name the Audience in Context: “This goes to current clients who’ve been with us at least a year and asked about adding services.”

Mistake 3: No format, no consistency

“Give me feedback on this report.”

Without a Format, the AI decides: sometimes paragraphs, sometimes bullets, sometimes praise first. It’s hard to act on.

Fix: “Give feedback as a numbered list. For each item: name the issue, explain why it matters, and suggest one specific fix.”

Mistake 4: Trusting the output without reviewing it

This is the most expensive mistake on the list. AI output is a draft, not a finished product. It can sound completely confident while being wrong, outdated, incomplete, or subtly biased. It fills gaps with what’s plausible, not what’s true.

Fix: Review every output before you use it. Verify any claim that matters. Cut anything you can’t confirm. The review step is where your judgment does the work AI can’t.

Mistake 5: Pasting sensitive information into a public tool

Client files, patient records, student data, and employee information: once pasted, it has left your control.

Fix: Ask, “Would I be comfortable if this appeared somewhere I don’t control?” If not, don’t paste it. Use placeholders and approved tools.

Mistake 6: Giving up after one bad output

Fix: Reread your prompt. Go through R.I.C.K. section by section. What’s missing? What’s vague? Add it and try again. Treat the prompt like a draft, because it is one.

Think about it: Which of these six mistakes have you made? Most people have made at least three. Which one do you think costs organizations the most, and why?

Iterate: turn a weak result into a strong one

Diagnose with R.I.C.K.

When an answer misses, match the problem to the missing field:

The answer is…	Probably missing
Generic, written for anyone	Role or Audience
Doing the wrong thing	Action
Inventing facts	Source and a Quality check that flags gaps
Off-target or pointless	Goal
The wrong length or shape	Format
Full of jargon or clichés	Tone and Avoid

Give feedback with Keep / Change / Because

“Keep the opening and the list of dates. Change the closing to one sentence with the sign-up link. Because most people will read this on their phones and won’t scroll.”

The Because is powerful: once the AI knows readers are on phones, it may tighten other sections too.

Ask the AI to review itself

“Review your draft against every field in my R.I.C.K. prompt. List anything you missed, assumed, or couldn’t support, then give me a revised version.”

This catches skipped instructions and hidden assumptions. Then check it yourself, because self-review can’t catch a fact the AI doesn’t know is wrong.

Know when to start over

If quality stops improving after three or four rounds, stop patching. Rewrite the original R.I.C.K. prompt with everything you’ve learned and start a fresh chat. Long, tangled conversations can carry old mistakes forward.

Save what works: your prompt library

When a prompt works, don’t just close the window. Turn it into a template by replacing details that change with {{blanks}}:

Task: Write a reminder email to {{audience}} about {{event}} on {{date}}…

Then save it as a library entry:

Field	Example
Name	Volunteer Reminder Email
Use when	2–3 days before any volunteer shift
Prompt	Your full R.I.C.K. prompt with {{blanks}}
Check before using	Date, time, location, and sign-up link
Last updated	Oct 2026, your name

Organize your library by task: emails, summaries, plans, social posts, and learning. Keep private details out of saved templates; fill them in only at the moment of use.

A library of 10 to 15 proven prompts can save hours every week, and it lets anyone on your team produce consistent, high-quality results from day one.

A worked example: four rounds to done

Tre is captain of his school’s debate team and needs a recruiting post for the school’s Instagram.

Round 1: His R.I.C.K. prompt produces a solid post, but it’s 180 words and sounds like an adult wrote it.

Round 2: “Keep the line about traveling to tournaments. Change it to under 70 words and make it sound like a student talking to students, because 9th graders scroll past anything that sounds like an announcement.”

Round 3: “Review your draft against my R.I.C.K. prompt. What did you assume?” The AI flags that it said “no experience needed” and “practices are twice a week,” but neither was in his context. Tre confirms the first is true and corrects the second: practices are Tuesdays only.

Round 4: Tre reads it aloud, cuts one word, and posts it.

Then he saves it as “Club Recruiting Post” in his library, with {{club}}, {{meeting day}}, and {{selling point}} as blanks, and shares it with the student council.

Think deeper: why your second prompt is smarter than your first

Something interesting happens when you iterate: you learn what you actually wanted. Many people start a prompt with only a fuzzy idea of the goal. Seeing a wrong answer is often the fastest way to discover what a right answer would need to include. In that sense, a “bad” first output isn’t a failure. It’s information.

This is also why prompt libraries become so valuable over time. Each saved template holds lessons learned from earlier rounds: the constraint you forgot the first time, or the Avoid list you built after seeing the same cliché three times. Your library becomes a record of your judgment, not just a set of shortcuts.

Questions to sit with:

• Is it a weakness of AI that it needs several rounds, or is that just how all good work with any collaborator happens?
• What would a prompt library tell a new employee about how your team thinks, what it values, and what it refuses to do?

Practice: four rounds and a library entry

1. Run a R.I.C.K. prompt for a real task.
2. Give one Keep / Change / Because follow-up.
3. Ask the AI to review its draft against your R.I.C.K. prompt.
4. Check the result yourself with the 5-question check.
5. Save the final prompt as a library entry with {{blanks}}.

You’ve finished Prompting Basics

You can now build complete prompts with Role, Instructions, Context, and Key Output, recognize and fix the most common mistakes, iterate to a strong result, and save your best work for reuse. Those skills work with every AI tool, now and in the future.

Want the full story behind the R.I.C.K. Model? It comes from Chapter 5 of Rage with the Machine by Rick Pride.

Next, put AI to work on repeat tasks in Automation with AI.

FOLLOW-UP (Keep / Change / Because):
Keep: [what's working]
Change: [exactly what to fix]
Because: [why, so the AI applies the same thinking everywhere]

SELF-REVIEW:
Review your draft against every field in my R.I.C.K. prompt. List anything you missed, assumed, or couldn't support, then give me a revised version.

LIBRARY ENTRY:
Name: [short, searchable name]
Use when: [the situation]
Prompt: [your R.I.C.K. prompt with {{blanks}} for details that change]
Check before using: [what a person must verify]
Last updated: [date] by [name]