AI inventory reorder comparison: all three got 12 units, but the handoff needed edits

Our starting point: Claude’s table

Claude keeps on-hand stock, committed orders, damaged units, available stock, incoming stock, the gap and the proposed order in one table. D and E are explicitly pending, distinct from the no-order rows. For checking this six-product snapshot, we prefer that traceable layout.

We would not copy its handoff unchanged. It groups A, C and F as not meeting the replenishment condition. C has only 10 available units, so it does meet the check threshold. Its 30 incoming units cover the gap; that is why it needs no new order. A parenthetical explanation gets the reason right, but the main sentence should too.

The rules were fixed before the answers

We submitted the same Chinese question once to each website, with no length limit. The English and Korean editions localize that run. Each case contains 12 units. Available stock is on hand minus committed orders minus damaged stock; the last two categories do not overlap.

Only availability strictly below 20 triggers the check. If triggered, the new-order gap is 40 minus available minus confirmed incoming. Round a positive gap up to full cases. Unknown incoming requires checking; negative availability requires reconciliation first. These are fictional store rules, not universal inventory targets.

Input in on-hand/committed/damaged/incoming order: A cups 50/20/5/0; B notebooks 30/15/5/24; C storage bags 20/8/2/30; D stickers 15/10/1/unknown; E folders 8/10/0/0; F cards 20/0/0/12. Every number is in units.

A six-unit gap produces a 12-unit order

A has 25 available, so no check is triggered. F has exactly 20, also outside the trigger. B has 10 available and a gap of 40−10−24＝6. Rounding to a 12-unit case produces a new order of 12, without buying the 24 already on order again.

C has 10 available and triggers the check, but 40−10−30＝0, so no new order. D has 4 available but unknown incoming: pending, not zero. Substituting zero would create an unsupported 36-unit order. E has−2 available: reconcile first instead of calculating a 42-unit gap and rounding to 48.

An independent Python calculator checked all six rows. We also compared each original against seven precommitted criteria,21 checks in total. All three match the final quantities or pending states. The handoff wording has separate issues. We did not connect to a warehouse, place an order or measure time savings.

ChatGPT: a correct table, an inconsistent opening

ChatGPT leads with the decision, then supplies a calculation table, checks and a sendable message. It distinguishes B’s six-unit gap from the 12-unit order, explicitly rejects treating D’s unknown value as zero, and holds E for reconciliation.

Its opening nevertheless says A, C and F do not trigger replenishment. The table marks C as triggered, and later text correctly says incoming stock covers the gap. Copying only the first sentence would lose that distinction.

This is not an incorrect order quantity. The edit is specific: A and F do not trigger the check; C does, but needs no additional order.

ChatGPT, translated: A, C and F do not trigger replenishment.

Gemini: detailed arithmetic, an added collective wait

Gemini works through each product before repeating the result in a table and handoff. C’s trigger and zero gap remain distinct. It offers more calculation explanation than Claude, which is useful when the reader needs to follow the method.

The handoff adds that arrangements should be made together after the checks are complete. The input requires checks for D and E; it does not require B to wait for those unrelated records. We would separate B’s 12-unit proposal from the two pending recalculations.

E is marked yes in the trigger column but pending in the order column. Gemini does not actually order against negative stock. Still, a data-error status would be clearer than grouping an invalid record with normal threshold checks.

Gemini, translated: arrange together after the pending checks are confirmed.

Claude: useful pending labels, a summary worth editing

Claude labels both D and E pending and states what needs checking. F explicitly says equality with 20 does not trigger. Those distinctions make the table a useful review document instead of a list of unexplained zeros.

Its handoff then groups A, C and F as untriggered, much like ChatGPT’s opening. It correctly explains C’s incoming coverage in parentheses, so the quantity is not wrong. We would make the main sentence reflect the two-stage decision.

The conclusion also calls D and E incomplete data. More precise labels help: D is missing a value; E has inconsistent recorded quantities. Both are on hold, but require different follow-up.

Claude, translated: pending—first reconcile inventory/order-allocation records.

WekeyLab AI’s rewritten purchasing handoff

This is our revised artifact using the same input, not a fourth service response or proof of an order.

Purchasing review: propose 12 new B notebooks, one case. Availability is 10, confirmed incoming is 24, and the six-unit gap rounds up to 12. A cups have 25 available and F cards 20; neither triggers the check, so no new order. C storage bags have 10 available and do trigger the check, but 30 incoming cover the gap: no additional order this round. Current availability remains 10. D stickers have 4 available and unknown incoming; the new quantity is pending. Check already ordered, unreceived stock and recalculate. E folders show 8 on hand against 10 committed, yielding−2; reconcile both records before setting a new quantity. This is a purchasing proposal, not an order placed. No arrival date or stockout-prevention guarantee is assumed.

Keep the decision order when using your own data

Validate the record first, check the threshold second, subtract confirmed incoming third, and round to case size last. Do not fill an unknown cell with zero just to make every cell numeric. Pending verification and a confirmed zero new order are different states.

Reusable prompt: “Apply my stock definitions and reorder rules. Distinguish available, committed, damaged and incoming units. Handle missing or inconsistent records before checking the threshold. Show the new quantity, calculations and checks, then write a handoff without adding approval rules or delivery promises.” Append your own rules and inventory snapshot.

Shopify’s official inventory-state documentation also distinguishes incoming from available stock. It does not supply this example’s thresholds or case size. Chinese search volume was not measured. For adjacent tasks, see our vendor-quote comparison and order-CSV reconciliation comparison.