An AI tool only has what you give it in the conversation: your words, any file or text you paste in, and whatever it has learned from patterns in text during its training. It does not know your order, your factory or your buyer unless you tell it. At work you use an AI tool for small, repeated jobs: turning notes into an email, drafting a short status update, rewriting a message in plainer language, or summarising a long document. This lesson covers the flow of admin work an order itself creates, such as writing a shipment-readiness note for management before Order A ships.
Without a clear instruction, an AI tool still produces confident, well-organised text, but it fills any gap you left with something that merely sounds plausible. It does not tell you which of its sentences are guesses and which are facts you gave it. A clear instruction closes that gap before it opens: it states the task, gives the known facts, and says what shape the answer must take, so the output can be checked against what you already know is true.
You need: basic comfort typing into a chatbot-style AI tool, and Order A's own fixed facts (quantity, fabric, price) from earlier tracks.
By the end of this lesson you can:
Not covered here: which specific AI product to use, how any one tool is built internally, or a general claim about how accurate AI tools are. This course stays vendor-neutral and evergreen.
| Term | Meaning |
|---|---|
| AI tool | A chatbot-style program that produces text, drafts or summaries in reply to written instructions. |
| Instruction | The written request you give an AI tool, stating the task, the facts and the required format. |
| Context | The facts, documents or numbers you give an AI tool so it works from your real situation instead of guessing. |
| AI output | The text an AI tool returns after an instruction. |
| Known fact | A fact you already have from an approved source, such as Order A's own fixed quantity or price, used to check an AI output. |
| Fact-check | Comparing a claim in an AI output against a known fact before you rely on it. |
An AI tool predicts likely, well-formed text from patterns in language it has seen before. That is why it can turn a short set of notes into a readable paragraph very quickly. It is also why an instruction with a missing fact does not come back empty: the tool produces a smooth sentence anyway, and a smooth sentence is not the same thing as a correct one. The fix is not caution about using the tool. The fix is putting the three things it needs into the instruction itself: what you want done, the facts it should use, and the format the answer must come in.
| What can differ | The shapes you will meet | Why it matters to you |
|---|---|---|
| Order type | An export order like Order A needs facts for a buyer-facing update; a domestic Make to Stock program needs facts about shop replenishment instead of a single shipment. | The task and the facts change, but the instruction still needs a task, facts and a format. |
| Business model | A Make to Order (MTO) instruction usually names one order and its dates; a Make to Stock (MTS) instruction may ask for a recurring weekly note; a hybrid order may need facts from both a confirmed portion and a forecast portion stated separately. | Mixing a confirmed fact with a forecast fact in one sentence, without saying which is which, is a common source of a misleading AI output. |
| Who reads the output | An internal note for a production manager can use trade terms freely; a note that will reach a buyer or a factory outside your own company needs the format and tone stated in the instruction. | The required format is part of the instruction, not something to fix afterwards. |
| Amount of context given | A one-line instruction with no facts; an instruction with two or three key facts pasted in; an instruction with a full data table attached. | More relevant context lowers the chance the tool fills a gap with a guess, but it does not remove the need to fact-check the result. |
| Product family | The facts change with the product (a sock order's yarn count and gauge, a towel order's GSM and pile), but the instruction still needs the task, the facts and the format. | The pattern for writing a clear instruction is the same across every product family Delta Garments makes. |
Mona, the merchandiser at Delta Garments, needs a short internal note summarising Order A's shipment readiness before a Friday production meeting. Order A is Northline Apparel's confirmed T-shirt order: 15,000 pieces, single jersey fabric at 180 GSM, agreed FOB USD 3.20 per piece.
Her first attempt is one line: "Write a shipment readiness note for the T-shirt order." The AI tool returns a fluent paragraph, but it invents a quantity and a date, because neither was in the instruction. Mona did not lie to the tool; she simply left a gap, and the tool filled it the way it fills every gap: with the most plausible-sounding continuation of the sentence.
Her second attempt states the task ("write a five-line internal note for the Friday production meeting"), gives the known facts (15,000 pieces, agreed FOB USD 3.20, fabric 180 GSM, current stage: sewing complete, inspection pending), and sets the format (five short bullet lines, no invented dates). The output now only contains facts Mona already gave it, so she can check each line against Order A's own approved figures in seconds.
The worked numberMona lists four facts a usable shipment-readiness note must contain: quantity, agreed FOB, fabric, and current production stage.
The count does not measure the AI tool's general skill. It measures something Mona controls directly: how many of the facts the note needed were actually in her own instruction. When she supplied all four, all four came back correctly, because the tool had no gap left to fill with a guess.
A fluent, well-organised paragraph feels like it came from someone who understands Order A. It did not. The tool has no memory of Delta Garments or Northline Apparel beyond what was typed into that conversation. A common mistake is treating fluent output as verified output, then passing an invented quantity or date into a real meeting or, worse, a real document. The instruction is the only place those facts can come from reliably, so a gap in the instruction is a gap in the output, dressed in confident language.
An AI tool writes smooth sentences from patterns in language, not from knowledge of your factory or your order. If you tell it the task, the facts and the shape you want, it uses exactly those facts and that shape. If you leave any of that out, it does not stop and ask; it fills the gap with something that reads well but may not be true. Writing a clear instruction is how you decide what goes into the answer, instead of leaving that decision to a guess.
Mona's second, clear instruction produces a five-line note that lists 15,000 pieces, USD 3.20 agreed FOB, 180 GSM fabric, and "sewing complete, inspection pending." A colleague, in a hurry, suggests sending the note straight to Sara at Northline Apparel to save time, since the instruction was clear this time.
Should Mona send the note to Northline Apparel without reading it against Order A's own facts first? Decide before opening the answer.
No. A clear instruction lowers the chance of a wrong fact; it does not remove the need for a human check before an external hand-off.
1. A clear instruction reduces guessing, but it does not guarantee accuracy. The tool can still misstate a fact, especially if Mona's own wording was ambiguous in one place. 2. The note is going outside Delta Garments, to the buyer. Any wrong figure that reaches Northline Apparel becomes a business fact someone else may act on, at higher cost to fix than an internal note. 3. The check takes under a minute: does the note say 15,000 pieces, USD 3.20, and 180 GSM, and does it correctly describe the current stage as sewing complete with inspection pending? 4. Only after that check does the note leave Delta Garments. A clear instruction is step one of a two-step habit; the human check is step two, and it never gets skipped for output a buyer will read.
The key point is that a good instruction changes how likely the output is to be correct; it does not change who is accountable for what leaves the building.
Order A's total FOB value is a fixed, approved figure:
If an unchecked AI-drafted note quoted a wrong quantity to Northline Apparel, for example implying part of the USD 48,000 order was short-shipped when it was not, the resulting phone calls, corrected emails and lost trust would cost far more than the minute Mona spends fact-checking a five-line note before she sends it. The saving from skipping the check is measured in seconds; the cost of a wrong fact reaching a buyer is measured in hours of correction and a dent in how much the buyer trusts the next note.
Use the T18-C02 Day-to-day AI-use checklist workbook. For each AI-assisted task, record the task in one line, the facts you supplied, the format you asked for, and whether you fact-checked the output before using it. Start with Mona's shipment-readiness note for Order A, then repeat the exercise for one instruction you write this week.
[Download the Day-to-day AI-use checklist workbook](../../templates/T18-C02-day-to-day-ai-use-checklist.xlsx)
1. Why did Mona's first, one-line instruction produce a wrong quantity and a wrong date?
2. A merchandiser at Cedar & Co. needs a short weekly restock note for its own shops. Which instruction is most likely to produce a checkable note?
| EN | AR | FR | TR | Meaning |
|---|---|---|---|---|
| AI tool | أداة ذكاء اصطناعي | Outil d'IA | Yapay zekâ aracı | A chatbot-style program that produces text, drafts or summaries in reply to written instructions. |
| Instruction | تعليمة | Instruction | Talimat | The written request you give an AI tool, stating the task, the facts and the required format. |
| Context | السياق | Contexte | Bağlam | The facts, documents or numbers you give an AI tool so it works from your real situation instead of guessing. |
| AI output | مخرج الذكاء الاصطناعي | Sortie de l'IA | Yapay zeka çıktısı | The text an AI tool returns after an instruction. |
| Known fact | حقيقة معروفة | Fait connu | Bilinen gerçek | A fact you already have from an approved source, used to check an AI output. |
| Fact-check | التحقق من الحقائق | Vérification des faits | Gerçek kontrolü | Comparing a claim in an AI output against a known fact before you rely on it. |
Next you will learn how to use an AI tool to draft a document quickly while still deciding, checking and standing behind everything it says. Back to the course map