An agent like Meridian Sourcing works across more than one buyer and more than one order type in the same week, which raises the number of facts an AI tool might be asked to help organise, and the number of chances for an unsupplied fact to be filled with a plausible guess. Knowing that a check is needed, from lesson 1.3, is not the same as having a method for doing it quickly and consistently.
This lesson gives you a short, repeatable method for checking an AI-generated answer, so that the check itself becomes routine rather than something skipped under time pressure. Without a method, "check it first" stays good advice that is easy to forget exactly when a hybrid mix of orders makes the check matter most.
You need: the ideas from lessons 1.1 to 1.3 (what an AI tool does, which tasks suit it, why it makes confident mistakes) are useful preparation but not required.
By the end of this lesson you can:
Not covered here: an automated fact-checking tool, or a claim that any specific AI product includes a built-in verification feature. Lesson 1.5 brings this method together with everything else in the course into one personal checklist.
| Term | Meaning |
|---|---|
| Fact check | Comparing a specific claim in an output against a source you know is correct. |
| High-stakes fact | A fact whose error would cost real money, time or trust if acted on before being checked. |
| Cross-buyer mix-up | An error where a fact belonging to one buyer's order is applied to another buyer's order. |
| Verification step | A defined point in a task where a person confirms a fact before it is used further. |
| Safe-to-use output | An AI-generated result whose high-stakes facts have been checked and confirmed against a source. |
Not every sentence in an AI-generated output carries the same risk. A polished turn of phrase costs little if it is imperfect; a wrong quantity, price, date or buyer name can cost real time and money if it is acted on. An efficient check identifies the few facts in an output that would be expensive if wrong, and compares only those against a source document, rather than re-reading every word with equal attention, which is slow enough that it tempts a busy person to skip the check altogether.
| What can differ | The shapes you will meet | Why it matters to you |
|---|---|---|
| Hybrid order handled through an agent | Meridian Sourcing works across Order A's Make to Order terms and other buyers' mixed order types in the same week, and must check which buyer's figures apply before using an AI-generated draft. | A cross-buyer mix-up, applying one buyer's fixed figure to another buyer's document, is a specific, checkable risk unique to working across several accounts. |
| Export order under Make to Order | A factory checks an AI-generated cost note against Order A's own fixed FOB and quantity, both single, unchanging facts once confirmed. | A single confirmed source (the purchase order) makes the check fast, since there is only one place to look. |
| Domestic replenishment under Make to Stock | A domestic brand checks an AI-generated reorder note against the current stock sheet, which changes weekly. | The source itself changes over time, so the check must use the current version of the sheet, not last week's copy. |
| A new agent compared with an experienced one | A new agent may not yet know which buyer's terms are commission-based or which figures are confidential to one buyer only. | Knowing which fact belongs to which source is itself part of the skill; without it, even a careful check can compare against the wrong document. |
| Different product families | Checking a sock order's gauge or a towel order's pile height against its own source sheet follows the identical method used for Order A's fabric weight. | The method does not change by product family; only the specific fact and source being compared do. |
Meridian Sourcing is an agent in Egypt working for two European buyers on a 5% commission. In the same week, Meridian is handling Order A (Northline Apparel, Make to Order, agreed FOB USD 3.20 per piece, 15,000 pieces) and a hybrid-model order for a second buyer. Meridian asks an AI tool to draft a short commission summary covering both orders for an internal record.
The AI-generated draft states Meridian's commission on Order A correctly, but for the second buyer's hybrid order, it states a commission rate that was never confirmed by that buyer in writing, a plausible-sounding figure the tool filled in because Meridian's own request did not specify it. Before filing the note, Meridian checks each stated commission figure against its own written agreement for each buyer, finds the mismatch on the second buyer's line, and corrects it before the record is saved.
The worked numberThis saving is only possible because Meridian identified the two figures that mattered before checking anything, rather than reading the whole note with equal attention, and then went to each buyer's own written agreement in turn to confirm them.
The methodWorking across more than one buyer or order raises a specific trap: comparing a stated fact against a familiar source out of habit, rather than the correct one for that particular line. A common mistake is confirming a commission rate or a fabric figure against Order A's own terms because Order A is the one most recently reviewed, when the fact in question actually belongs to a different buyer's agreement. The cost of this mistake is a false sense of having checked something, when the comparison itself was against the wrong record. A check performed against the wrong source can feel identical to a correct one, since both involve opening a document and comparing a number, which is why naming the correct buyer or order first, before opening anything, matters as much as the comparison itself.
Think of checking an AI-generated answer the way you would check a delivery note against a purchase order: you do not reread the whole warehouse's inventory, you check the specific items and quantities that matter for this delivery, against the specific order they belong to. Do the same with an AI-generated answer: find the few facts that matter, find the right source for each one, and compare them, one by one, especially when more than one buyer's own paperwork is involved in the same task.
A colleague at Meridian Sourcing asks for a quick confirmation of the fabric weight on a different, hybrid-model buyer's order, and pastes in an AI-generated summary to save time. The summary states "180 GSM," which is correct for Order A but was never confirmed for this particular hybrid buyer's own order.
Should Meridian confirm 180 GSM for the hybrid buyer's order? Decide before opening the answer.
No. Meridian must open the hybrid buyer's own tech pack and confirm the figure from that document before answering.
1. 180 GSM is Order A's own fixed fact, not automatically the hybrid buyer's own figure, and reusing an AI-generated summary across two different orders is exactly the cross-buyer mix-up this lesson warns against. 2. The source document for this specific question is the hybrid buyer's own tech pack, not the earlier note about Order A, however similar the two might look. 3. Confirming an unchecked figure to a buyer costs far more to correct, once a fabric order is placed on the wrong weight, than the few minutes needed to open the correct source document.
The AI-generated summary was not wrong for the task it was originally written for; it became a risk only when it was applied to a different buyer's question without checking its source against the order actually being discussed.
Order A's total order value stays fixed at:
A commission or specification mix-up caught before it reaches a buyer costs only the few minutes of a targeted check. Caught after it reaches a second buyer's own record, such as a wrong commission rate stated on an internal note that is later shared, it can cost hours of correction and, more seriously, a buyer's confidence that Meridian Sourcing keeps each account's own figures separate and correct, a reputation cost far larger than any single order's value.
Use the T18-C01 AI use worksheet. In the Verification log sheet, record each AI-generated output, the high-stakes facts you identified in it, the source document you checked each one against, and whether a correction was needed. Start with Meridian Sourcing's own commission-note example, then repeat the exercise for a task from your own week.
[Download the AI use worksheet](../../templates/T18-C01-ai-use-worksheet.xlsx)
1. Why does Meridian Sourcing check the commission figure for the second buyer's hybrid order separately from Order A's own commission figure, rather than checking the whole note as one block?
2. A trader at Atlas Trading uses an AI tool to draft a summary covering a Make to Order shipment and a separate Make to Stock reorder for two different retailers in the same message. What is the most efficient way to check this summary before sending it?
| EN | AR | FR | TR | Meaning |
|---|---|---|---|---|
| Fact check | تدقيق حقيقة | Vérification de fait | Gerçek kontrolü | Comparing a specific claim in an output against a source you know is correct. |
| High-stakes fact | حقيقة عالية المخاطر | Fait à forte incidence | Yüksek riskli gerçek | A fact whose error would cost real money, time or trust if acted on before being checked. |
| Cross-buyer mix-up | خلط بين مشترين | Confusion entre acheteurs | Alıcılar arası karışıklık | An error where a fact belonging to one buyer's order is applied to another buyer's order. |
| Verification step | خطوة تحقق | Étape de vérification | Doğrulama adımı | A defined point in a task where a person confirms a fact before it is used further. |
| Safe-to-use output | مخرَج آمن للاستخدام | Résultat prêt à l'emploi | Kullanıma güvenli çıktı | An AI-generated result whose high-stakes facts have been checked and confirmed against a source. |
Next you will bring everything from this course together into one short, personal checklist you can use on any AI-generated task, seen through Cedar & Co.'s own domestic buying work. Back to the course map