03Arc 1 / Control / GROUND / 120 MIN

Work With Context, Files & Sources

I can identify what AI needs to know, choose which source should be trusted, and inspect whether AI actually used the right evidence.

Session / includes prompting practice Prompting exercise · use AI/GPT
What you'll learn

I can identify what AI needs to know, choose which source should be trusted, and inspect whether AI actually used the right evidence.

Bring / required

Required: computer or phone with AI/GPT for the live prompting route. Bring your own small work task or use the starter case. Your course playbook should be open for notes.

What you'll do

Follow the session exercise, run only the text explicitly marked as a prompting activity, inspect the result and record evidence in your session card.

Take away

A completed session card, a checked result or decision, and an update to your Personal AI Work Playbook.

Explanations and workbook templates are not prompts. Prompting activities are labelled explicitly.

The session

THREE BITES. ONE PRACTICAL MOVE.

03AGive AI the Right Reality
03BAvailable Does Not Mean Authoritative
03CAsk for Evidence, Conflicts & Unknowns

What you'll practise

  • Identify the minimum useful context for a real task.
  • Choose a source of truth when sources conflict.
  • Inspect evidence, conflicts and unknowns rather than trusting fluent output.
Keep this idea

A better instruction cannot compensate for the wrong reality.

Try it with the supplied case

Prompting exerciseYou write or revise the AI instruction as part of this exercise. Only the instruction for that step plus the referenced case material goes into AI; the worksheet remains your notes.

Question: Which information should I trust? Input: S2/S3.

  1. In 4 minutes choose the Q-104 price privately and give your reason.
  2. In 12 minutes rank P-SEP, P-JUN, the chat note and EX-103 by authority, date and scope. Prepare a tiny context pack for Q-104.
  3. In 12 minutes inspect a draft's important price claim. Open the original card, point to the exact row/exception, and revise unsupported claims.
  4. Change the enquiry to Q-103. Explain why a different applicable source can change the price.

Keep/check: Q-104 RM4,000; Q-103 RM1,500 under its explicit exception. Save one claim, exact supporting passage/value and limitation. AI's own “supported” label is not proof.

Use this now

ANCHOR TASK GROUNDING PACK

Use this as your session notes and evidence. Do not paste the whole card into AI. Only text explicitly labelled Prompting exercise belongs in an AI chat.

ANCHOR TASK
______________________________________________________

CONTEXT PACK

Task:
______________________________________________________

Facts needed:
______________________________________________________

Source material:
______________________________________________________

Examples to imitate:
______________________________________________________

Rules to obey:
______________________________________________________

Ignore:
______________________________________________________

Unknown:
______________________________________________________

SOURCE-OF-TRUTH RULE

For fact ______________________, trust ______________________
because _________________________________________________.

Do not trust __________________ because ____________________.

EVIDENCE PATTERN

Mark claims as:
SUPPORTED / INFERENCE / CONFLICT / UNKNOWN

One claim I checked:
______________________________________________________

PROOF
What changed after grounding?
______________________________________________________

RETURN TO MY QUESTION
Status: answered / demonstrated / needs more evidence / outside this course
Evidence or gap:
Next application:

Class rhythm

Feel itExperience the problem or contrast before naming the method.
Name itMake the useful pattern explicit.
Use itApply it to a real work object and inspect the result.
Keep itCapture the part worth reusing after class.

Field-guide bridge

Aha

Session 03 teaches how to ground real work. The Context Engineering field guide is where I go deeper into minimum useful context, source authority, conflict, freshness and how to prove that an important claim actually came from the right reality.

Open Context Engineering →

← Previous / 02Brief & Steer AI for Better Results Next / 04 →Analyse, Compare & Audit Consistently