Why AI gives you mush

Narrated by a synthetic voice, not by me — said here because you’d rather know than find out. The narration covers the chapter itself; the copy-paste prompt library at the end is written-only.

By the end of this lesson you will be able to turn a weak, watery AI answer into a sharp, useful one — using three simple moves.

You open ChatGPT or Grok or Claude. You type a question. You get back a wall of polite text that could have come from a brochure.

That is mush.

We measured this in our own work. Before we fixed our prompts, most answers needed heavy editing. Today the AI produces about 95% of our work product — but only when we give it the right instructions.

The difference is not which AI you use. The difference is what you type in.

A quick word before we start. A prompt is the message you type to the AI. That's it. This whole lesson is about writing better ones.

Why You Get Bland Answers

AI is trained to be helpful. To a machine, "helpful" usually means safe, balanced, and average.

The AI has read millions of web pages, books, and forum posts. It knows what the average answer looks like. So that's what it gives you — unless you force it not to.

We saw this in our own early work sessions. We typed "help me with my marketing." We got back a list of ideas that would fit any business on Earth. None of them fit ours. We wasted hours rewriting.

The fix is not a better AI. The fix is three levers. A lever is a small move that makes a big difference. You control all three, every single time you ask for something.

Lever 1: Give the AI a Job

Tell the AI who it should act as. (Some people call this "role framing." We just call it giving the AI a job.)

Don't say: "Give me marketing ideas."

Say: "You are the head of marketing for a company that has run hundreds of logged work sessions with AI. You have built real systems that put out work every day."

The job changes what counts as a good answer. A helper with no job gives everyone the same advice. A head of marketing with a track record gives you specific moves, in order, that fit the limits you're about to give it.

That is why every one of the seven prompts waiting for you in Chapter 2 begins by giving the AI a job.

Lever 2: The No List

The No list is a short list of things you tell the AI NOT to do. (The formal name is a "refusal clause." Forget that. It's a No list.)

Add lines like these:

• No advice that could fit any business but mine.
• No lists without proof behind them.
• Never use hype words.

Here's why it works. The AI is trained to avoid saying anything that might be wrong. So it hides in the safe middle. The No list blocks the safe middle. Now the only way out is to be specific.

In our own work, this one lever cut fix-up time the most. One of our early prompts had a weak No list, and the AI kept telling us how great we were doing. We added two rules — "Audit from evidence, not self-report" (judge us only on real results, not on how we describe ourselves) and "No inference presented as observation" (no guesses dressed up as facts). The answers went from a pep talk to a real check-up.

Lever 3: The Answer Format

Tell the AI exactly what the answer should look like. (The formal name is "forced output shape." We call it the answer format.)

Don't say: "Tell me what to do."

Say: "Give me exactly these things:"

• The ripple effects I have not thought of.
• The one small change that would fix several problems at once.
• The safe bet — the option with a small downside and a big possible upside.
• What could prove you wrong.

Named parts force real thinking. Loose paragraphs invite mush. The answer format works like a checklist the AI has to finish.

Every one of the seven prompts in Chapter 2 ends with an answer format plus the prove-it line. That combination is the pattern behind them all.

The Prove-It Line

The prove-it line is the single most powerful thing we ever added to our prompts. It is one line at the end:

"Now tell me what could prove you wrong."

This line turns a confident-sounding answer into an answer you can test. It forces the AI to show what it's assuming. In our own decision records, this line stopped several overconfident plans before anyone acted on them.

We now require this line in every prompt where the stakes are real.

Session Story: The Viral Post That Was Not a Secret Mode

Early on, a post crossed our desks. The title: "ChatGPT has a secret mode. 7 prompts that feel illegal to know."

It was getting saved and shared all over the feed.

But the title was marketing. There is no secret mode. Those prompts work because of the three levers you just learned. That post became the first lesson in this course — the one you are reading right now. We swapped the hype for the levers.

A Note About New AI Models

AI models change all the time. The levers do not.

The job, the No list, and the answer format work on every major model we have tested. When you save a result, write down which AI model you used. The levers are a method, not a tool.

Do this now

1. Open Claude (or whichever AI you use). You will see a box at the bottom where you type, like texting.
2. Pick one real decision you face this week. Write it down in one sentence.
3. Write a prompt that uses all three levers plus the prove-it line. Use Main Prompt 1 below as your guide, or the Strategist prompt from Chapter 2. Type it into the box and hit send.
4. Now click "New chat" to start over fresh. Ask the same thing as a plain, one-line question. Compare the two answers.
5. Make a document in Google Docs or Word. Name it My AI Notebook. Copy both answers into it (highlight the answer, right-click, choose Copy — then right-click in your document and choose Paste). You will add to this notebook in Lesson 3.

You know it worked when the new answer gives you at least one specific next move and at least one thing you could actually check that would prove it wrong. If it doesn't, make your No list stronger and run it again.

Prompt Library - Chapter 1

Four prompts you can paste today. Each one puts this chapter's levers to work on your own business or your own life.

Main Prompt 1: Your First Three-Lever Prompt

You are a sharp, honest business advisor. You give specific, usable advice, and you are helping me with [YOUR BUSINESS].

The No list:
- No advice that could fit any business but mine.
- No hype words.
- If you are not sure about a fact, say you are not sure.

My decision: [THE DECISION YOU'RE FACING]

Give me exactly these four things:
1. The ripple effects I have not thought of.
2. The one small change that would fix several problems at once.
3. The safe bet — the option with a small downside and a big possible upside.
4. The first step I should take this week.

Now tell me what could prove you wrong.

Make it yours — do this before you paste:

1. Copy everything in the gray box above (highlight it, right-click, choose Copy — or use the Copy button).
2. Change every word in [BRACKETS] to your own information. Use this table:

Replace this	With	Example
[YOUR BUSINESS]	what you do, in a few words	"my bakery in Flint"
[THE DECISION YOU'RE FACING]	the real choice in front of you, in one or two sentences	"Should I start taking wedding-cake orders, or stick to walk-in sales?"

1. Paste the whole thing into the AI chat — right-click in the typing box, choose Paste — and hit send.
2. What you should get back: four numbered answers about your exact situation, plus one thing you could check that would prove the advice wrong. If you don't, add this line to the No list — "Do not give me advice that would work for any business but mine" — and send it again.

Here's what it looks like filled in (Maria's version — Maria is a made-up bakery owner we use as an example):

You are a sharp, honest business advisor. You give specific, usable advice, and you are helping me with my bakery in Flint.

The No list:
- No advice that could fit any business but mine.
- No hype words.
- If you are not sure about a fact, say you are not sure.

My decision: Should I start taking wedding-cake orders, or stick to walk-in sales?

Give me exactly these four things:
1. The ripple effects I have not thought of.
2. The one small change that would fix several problems at once.
3. The safe bet — the option with a small downside and a big possible upside.
4. The first step I should take this week.

Now tell me what could prove you wrong.

Sub-prompt variations for Main Prompt 1 - Want more? Paste this next: "Take your best idea from that answer and turn it into a simple one-week plan: what I do Monday, and what it should look like by Friday." - Check it: paste this to test the answer: "Now be the toughest fair critic of your own answer. Which of your four points is the weakest, and why?" - Too long? Paste this: "Say all of that again in under 150 words, as a plain list I can read on my phone."

Main Prompt 2: Turn Any Question Into a Three-Lever Prompt

I am going to give you a plain question. Do not answer it yet.

First, rewrite my question as a stronger prompt that:
1. Gives you a job — who you should act as to answer this well.
2. Adds a No list — three things a lazy answer would do that you must not do.
3. Sets the answer format — the exact pieces the answer must include.
4. Ends with this line: "Now tell me what could prove you wrong."

Show me the rewritten prompt. Then answer the rewritten prompt.

My plain question: [YOUR PLAIN QUESTION]

Make it yours — do this before you paste:

1. Copy everything in the gray box above.
2. Change every word in [BRACKETS] to your own information. Use this table:

Replace this	With	Example
[YOUR PLAIN QUESTION]	the question you would normally type	"How do I get more customers?"

1. Paste the whole thing into the AI chat and hit send.
2. What you should get back: first the stronger, rewritten prompt, then the answer to it. If it skips straight to an answer, reply "Show me the rewritten prompt first, then answer it," and send that.

Sub-prompt variations for Main Prompt 2 - Want more? Paste this next: "Do the same thing with the next question I type — rewrite it first, then answer the rewrite." - Check it: paste this to test the answer: "Which of the three levers did your rewrite lean on the most, and what would the answer have missed without it?" - Too long? Paste this: "Shrink the rewritten prompt to five lines or fewer without losing the job, the No list, or the answer format."

Main Prompt 3: Fix a Mushy Answer

You are helping me fix a weak AI answer.

Here is what I asked for: [WHAT YOU ASKED FOR]

Here is the answer I got: [PASTE THE AI'S ANSWER HERE]

Tell me:
1. Which lever was missing — a job, a No list, or an answer format.
2. A rewritten prompt that adds the missing levers.
3. The new, better answer to that rewritten prompt.

Now tell me what could prove you wrong.

Make it yours — do this before you paste:

1. Copy everything in the gray box above.
2. Change every word in [BRACKETS] to your own information. Use this table:

Replace this	With	Example
[WHAT YOU ASKED FOR]	your original question, copied word for word	"Give me marketing ideas for my shop"
[PASTE THE AI'S ANSWER HERE]	the whole weak answer, pasted in	the wall of bland text you got back

1. Paste the whole thing into the AI chat and hit send.
2. What you should get back: a short diagnosis, a stronger prompt, and a sharper answer. If the new answer still feels bland, reply "Make the No list stronger and try again."

Sub-prompt variations for Main Prompt 3 - Want more? Paste this next: "Turn that rewritten prompt into a fill-in-the-blanks template I can reuse for any task like this one." - Check it: paste this to test the answer: "Put the old answer and the new answer side by side. Name the three biggest differences in plain words." - Too long? Paste this: "Give me only the rewritten prompt. Skip the explanation."

Main Prompt 4: Build Your Own Reusable Template

Help me build one prompt template I can reuse every day.

About me: [YOUR BUSINESS]

The kind of work I ask AI for most: [THE WORK YOU ASK AI FOR MOST]

Build me a fill-in-the-blanks template that:
1. Gives the AI a job that fits my work.
2. Includes a No list of three or four rules that block bland answers.
3. Sets an answer format with named parts.
4. Ends with this line: "Now tell me what could prove you wrong."

Leave a blank spot marked [MY TASK TODAY] where I will type each day's details. Then show the template used once, on a made-up example, so I can see how it works.

Make it yours — do this before you paste:

1. Copy everything in the gray box above.
2. Change every word in [BRACKETS] to your own information. Use this table:

Replace this	With	Example
[YOUR BUSINESS]	what you do, in a few words	"my bakery in Flint"
[THE WORK YOU ASK AI FOR MOST]	the task you hand to AI most often	"writing posts for my shop's Facebook page"
[MY TASK TODAY]	leave this one alone — the AI keeps it as a blank in your template	don't change this one

1. Paste the whole thing into the AI chat and hit send.
2. What you should get back: a fill-in-the-blanks template plus one worked example of it in use. If the template runs longer than one page, reply "Cut it so it fits on one page."

Sub-prompt variations for Main Prompt 4 - Want more? Paste this next: "Make two more versions of the template: one for quick daily tasks, one for big decisions." - Check it: paste this to test the answer: "Test the template on a task it wasn't built for. Where does it break, and what one line would fix it?" - Too long? Paste this: "Cut the template to under 120 words so I can paste it fast."