Sara, a buyer at Northline Apparel, often needs a document written quickly: a confirmation email to Delta Garments, a summary for her own manager, or a short brief restating what was agreed on a call. An AI tool can turn her rough notes into a clean, well-organised draft in seconds. That speed is genuinely useful, but it changes where her attention needs to go. When Sara wrote the whole document herself, every sentence passed through her own judgement as she typed it. When an AI tool writes the first version, that judgement has to happen afterwards, on purpose, or it may not happen at all.
This lesson is about keeping that judgement in the loop. Drafting with an AI tool does not mean handing over the decisions inside the document; it means handing over the typing, while Sara still decides what is true, what is agreed, and what should not be said yet.
You need: basic comfort using an AI tool to turn notes into a draft, and Order A's own fixed facts from earlier lessons.
By the end of this lesson you can:
Not covered here: which specific AI product to use, negotiation tactics, or contract law. This course teaches evergreen judgement, not legal advice.
| Term | Meaning |
|---|---|
| Draft | A first version of a document, written to be reviewed and changed before it is final. |
| AI output | The text an AI tool returns after an instruction. |
| Professional judgement | The decision a person makes about what a document should say, based on the real facts and their own accountability for it. |
| Confirmed fact | Something that has actually been agreed or verified, as opposed to something that sounds agreed. |
| Iteration | Asking an AI tool to revise a draft again after checking it, rather than accepting the first version. |
| Document hand-off | The point where a document leaves one person's control and reaches someone else who will act on it. |
An AI tool is good at producing organised, readable prose from rough notes, because that is a pattern it has seen many times in the text it learned from. It is not good at knowing which of Sara's notes are firm and which are still under discussion, unless she tells it. Left alone, it tends to smooth everything into confident, settled-sounding language, because confident language is the common pattern for a business document. That tendency is exactly what makes an AI draft risky to send unchanged: a maybe can quietly become a definitely.
| What can differ | The shapes you will meet | Why it matters to you |
|---|---|---|
| Order type | An export confirmation to Delta Garments states shipment and payment terms; a domestic buying note to a local supplier states shop delivery windows instead. | The facts needing review differ, but every document still needs a check for what is confirmed versus what is hoped for. |
| Business model | An MTO confirmation email names one order's own dates; an MTS reorder note may state a recurring quantity that changes weekly; a hybrid order needs the confirmed portion and the forecast portion kept clearly separate in the draft. | Letting an AI tool merge a confirmed figure and a forecast figure into one sentence is a common source of a misleading document. |
| Who receives the document | An internal summary for Sara's own manager can stay informal; an email to Delta Garments becomes a record the factory may rely on. | The more the recipient will act on the document, the more the review has to focus on substance, not just tone. |
| How much the draft changes tone versus content | Sometimes an AI draft only needs a tone fix (too blunt, too long); sometimes it needs a substance fix (a stated fact that is not actually agreed). | Fixing only the tone while missing a substance error is worse than not drafting with AI at all, because the document reads as more reliable than it is. |
| Product family | The specific facts change (a sock order's delivery lot size, a bedding order's fabric approval status), but the review question stays the same: does every stated fact match what is actually confirmed? | The judgement pattern for reviewing a draft is the same across every product family Northline Apparel buys. |
Sara gives an AI tool her rough notes: Order A is 15,000 pieces at USD 3.20 per piece, total USD 48,000, and Northline Apparel pays by letter of credit at shipment. She asks for a short, formal confirmation email to Mona at Delta Garments.
The AI tool returns a well-written email. One sentence says: "We confirm the full USD 48,000 will be transferred to your account before production begins." That sentence is fluent and sounds reasonable, but it is wrong: Order A's actual, agreed term is payment by letter of credit at shipment, not a transfer before production. Sara's own notes never said "before production begins"; the AI tool filled that detail with a common payment pattern from other business emails it has seen, not from Order A's real terms.
Because Sara reads the draft against what she actually knows is agreed, she catches the error before sending. She corrects the sentence to state the real term, keeps the rest of the fluent draft, and sends the corrected version.
The worked numberSara's email had five sentences. She checks each one against Order A's own agreed facts.
One sentence out of five, twenty percent of the draft, stated something that was not actually agreed. That single sentence, if sent unchanged, would have told Delta Garments the wrong payment term for an order worth USD 48,000, not a small error.
A well-organised, polite draft feels finished, so it is tempting to fix a clumsy sentence and send. A common mistake is treating a tone edit as if it were a fact check. The two are different tasks: one makes the document read better, the other confirms the document is true. A draft can read perfectly and still contain one invented sentence, in the middle of four correct ones, that changes what the recipient believes was agreed.
An AI tool can turn rough notes into a clean, readable document quickly, but it tends to make everything sound settled, even a point that was only a hope or a guess. The person sending the document still has to decide what is really true and really agreed, and put that back into the draft before it goes anywhere. Fast writing is not the same as fast deciding; the deciding part still belongs to the person, every time. A document that reads smoothly can still be wrong in a way that only someone who knows the real, agreed facts will notice, so reading for tone is never a substitute for reading for truth.
Sara has the AI-drafted email in front of her. It reads smoothly and would be ready to send within a minute if she only fixes one slightly long sentence. She has not yet checked every stated fact against Order A's own agreed terms.
Should Sara send the email once it reads well, given the Friday deadline? Decide before opening the answer.
No. A document that reads well is not the same as a document that is correct; the deadline changes when Sara sends it, not whether she checks it first.
1. The draft already contains one known risk pattern: an AI tool tends to state an uncertain point as settled, and Sara has not yet compared every sentence to Order A's own USD 48,000 terms and payment method. 2. The recipient will act on the email. Delta Garments may plan production or payment expectations around a wrong stated term, at real cost to correct later. 3. The check takes a few minutes, not a few hours: five sentences, checked against facts Sara already knows are agreed. 4. If the deadline is genuinely at risk, the message to send is a short, honestly incomplete one ("confirming details, full email to follow shortly"), not an unchecked draft dressed up as final.
The key point is that a deadline is a reason to work efficiently, not a reason to skip deciding what the document actually says.
Order A's full commercial value is fixed and known:
If Sara had sent the uncorrected AI draft, Delta Garments would have read that the full USD 48,000 was to be transferred before production began, instead of the real, agreed term of payment by letter of credit at shipment. Correcting that misunderstanding after the fact, once Delta Garments has already planned around the wrong term, costs far more than the few minutes Sara spent checking five sentences before she sent the email.
Use the T18-C02 Day-to-day AI-use checklist workbook. For each AI-drafted document, record the task, the facts you supplied, whether you fact-checked the output, and whether a human check happened before it left your own team. Add Sara's confirmation email as one row, noting the payment-terms correction, then repeat the exercise for one document you draft with AI this week.
[Download the Day-to-day AI-use checklist workbook](../../templates/T18-C02-day-to-day-ai-use-checklist.xlsx)
1. Why did the AI-drafted email state the wrong payment term for Order A?
2. A domestic buyer at Cedar & Co. asks an AI tool to draft a reorder note for one of its shops. The draft states a delivery date as 'confirmed' that was actually only proposed. What should happen before the note is sent to the shop?
| EN | AR | FR | TR | Meaning |
|---|---|---|---|---|
| Draft | مسودة | Brouillon | Taslak | A first version of a document, written to be reviewed and changed before it is final. |
| AI output | مخرج الذكاء الاصطناعي | Sortie de l'IA | Yapay zeka çıktısı | The text an AI tool returns after an instruction. |
| Professional judgement | الحكم المهني | Jugement professionnel | Mesleki değerlendirme | The decision a person makes about what a document should say, based on the real facts and their own accountability for it. |
| Confirmed fact | حقيقة مؤكدة | Fait confirmé | Doğrulanmış gerçek | Something that has actually been agreed or verified, as opposed to something that sounds agreed. |
| Iteration | التكرار | Itération | Yineleme | Asking an AI tool to revise a draft again after checking it, rather than accepting the first version. |
| Document hand-off | تسليم | Transmission | Devir | The point where a document leaves one person's control and reaches someone else who will act on it. |
Next you will learn how to decide, task by task, whether starting with an AI-drafted first attempt actually saves you time. Back to the course map