Check the car-wash example step by step. Follow Carla's demonstration, pause to make your own attempt, and compare your check with the result.
Your activity / Right here
Bring a notebook. Begin below.
The activity is included on this page. Work on paper or in your own document; keep your first attempt and the evidence you checked.
AI Lab: The Walk Me Through Move
Use only an AI tool your instructor approves and that allows users your age. Check its minimum age and parent-permission rules first; an adult's permission cannot override an 18+ limit. Follow your school's rules, and do not share personal information. About 25 minutes.
Car Wash Fundraiser (Made-Up Example. The numbers are illustrative.)
| Row | Item | Amount |
|---|
| 1 | Saturday washes (40 cars at $10) | $400 |
| 2 | Sunday washes (25 cars at $10) | $250 |
| 3 | Tips jar | $85 |
| 4 | Total money in | $735 |
| 5 | Soap and sponges | $48 |
| 6 | Hose and bucket rental | $30 |
| 7 | Total costs | $78 |
If the AI gets everything right the first time, that still counts. An audit that finds no exceptions is still evidence, as long as you did the testing.
1. On paper, estimate the profit before you open the AI.
2. Type the table into the AI and ask, "What was the profit?" Write down its answer.
3. Use the move: "Walk me through exactly where you got that. Show every number you used, which rows you included, and your formula."
4. Put a tickmark next to each number the AI used: T (traced to the table), R (recalculated, and it matches), or X (exception).
5. Source round. Ask the AI one factual question from a class you are taking, and add, "Give me the source for your answer." Open the source. Mark the result Match, Mismatch, or Source not found.
6. Verify: Calculate the profit by hand. For the source round, find the exact sentence in the source, or check the claim in your textbook. Write down what you found and where you found it.
7. Record your AI use: which tool you used, what you asked, and what you kept from the answer.
Own It
No AI. About 15 minutes.
Portfolio piece: My AI Audit Log. Auditors document what they tested and what they found. Your AI Audit Log does the same thing, all semester. Copy this template into a notebook or a document:
| Date | Tool | What I asked | What the AI said | What I checked (my evidence) | Match or exception? | What I did about it |
|---|
| | | | | | |
Write two entries by hand: one for the car wash and one for the source round. Then finish this sentence:
"The kind of claim I will always verify myself is __________ because __________."
Context File Upgrade
Add this section to your my_context.md file:
## Verify Rules for My AI
- When you give me a number, show every step.
- For every fact, name your source, or tell me you do not have one.
- Say "I am not sure" when you are not sure.
- Never invent a source, a quote, or a number.
- When something matters, remind me to check it myself.
These rules can lower the error. They do not replace your check. An AI can still break a rule you gave it.
Skeptic's Corner: What Kind of Statement Is This?
These statements are exercises to classify and check. Do not assume every statement is true.
Use the six labels from Chapter 1: Fact, Idea, Rob's Story, Made-Up Example, Rule, and Prediction.
1. "Professional skepticism is an attitude that includes a questioning mind and a critical assessment of audit evidence and other information that is obtained to comply with PCAOB standards and rules."
2. "President Reagan used the phrase 'trust, but verify' at a treaty signing in 1987."
3. "Mike's group reported $1,240 in profit."
4. "A student described in Rob's slides used AI to prepare tickmark text but did the testing."
5. "Checking one good source is better than checking none."
6. "By 2030, AI will never make mistakes." (A claim you might see online.)
Read the video transcript
Read along, pause, or return to an idea. This transcript follows the selected lesson video.
I'm Carla, an AI avatar.
Today I'm going to show you the most useful sentence in
this course.
Then I'll show you what it caught.
This move comes from Rob's Audit Midyear slides.
There, step four of a prompt sequence is one sentence:
walk me through exactly where you got those numbers.
The note under it says that in a simulation,
this is where the user catches a mismatch and corrects it.
Here's a made-up table from a car wash fundraiser.
Seven rows: money in, money out,
and two rows of totals.
Before I open the AI, I think first.
Money in was seven hundred thirty-five dollars.
Costs were under a hundred.
So the profit should be around six hundred fifty.
I write that down.
Why estimate first?
An estimate gives me something to compare against.
If the AI's answer lands far from my estimate,
that gap is a signal to look closer.
Now I type the table into a teacher-approved AI and ask a
simple question.
What was the profit?
Thirteen hundred fourteen dollars.
That's about twice my estimate.
My estimate might be wrong.
The AI might be wrong.
I don't know yet, so I don't use either number.
This is the moment most people skip.
The answer looks official.
It has a dollar sign and a comma.
A clean number on a screen is still just a claim until
someone checks it.
Here's the move.
Walk me through exactly where you got that.
Show every number you used and which rows you included.
Now I can see its work, which means I can audit it.
I put a tickmark next to every number.
T means I traced it to the table.
R means I recalculated it.
X means exception: something that doesn't match.
Two X's.
Row 4 is total money in, and the AI added it as if it were more money.
Row 7 is total costs, and it added that as more costs.
It counted both totals twice.
Notice what I skipped: I never asked the AI,
"Are you sure?"
That would be asking it to check itself.
I checked it against the table, the original document.
So I push back with evidence.
Rows 4 and 7 are totals.
Recalculate without counting them twice.
Six hundred fifty-seven dollars.
That's close to my estimate, and close still needs proof.
So I verify it myself.
Seven hundred thirty-five minus seventy-eight is six
hundred fifty-seven.
Now the answer has evidence behind it:
my own math from the original table.
If the AI had been right the first time,
I would still do every step.
An audit that finds no exceptions is still evidence,
as long as I did the testing.
Now the source round.
I ask one factual question from a class and add:
give me the source for your answer.
Then I open that source and look for the exact sentence.
Match, mismatch, or source not found.
Last, I write it all in my AI Audit Log:
the date, the tool, what I asked,
what it said, what I checked, and what I did about it.
Then I record my AI use.
Here is my entry.
Tool: the one my teacher approved.
What I asked: the profit.
What it said: thirteen fourteen.
What I checked: the table and my own math.
Exception: totals counted twice.
What I did: pushed back and verified six fifty-seven.
The AI did the arithmetic.
I did the audit.
Your turn.
Open the AI Lab and try the move.
Before you use an AI tool
Follow your course or institution's rules. Use an instructor-approved tool that permits your age, and meet any required consent conditions. Keep names, contact details, school records, and other private information out of prompts. Use fictional examples when you can.
Learning without an AI account
Follow the demonstration, pause before each answer, and do the reasoning on paper. Discuss what you would ask, how you would check it, and which decisions would remain yours.
About the characters and voices
This course uses AI-created characters and synthetic voices. Carla is an AI avatar. They help tell the story; they are not real students sharing testimonials.
Designed for high-school and community-college learners. Choose examples from your classes, work, and everyday life. Keep your completed work in your course portfolio.
