Google Sheets AI Function: Classify Order Notes for Review
A short order note can sound more definite than it is. A customer asking to cancel has not necessarily received a cancellation, and printing a shipping label does not confirm courier pickup. Google Sheets AI can suggest a consistent category for these notes, provided a person checks the wording before accepting the result.
This exercise classifies what a note says. It does not establish the live state of an order or update a fulfillment system. Keep that distinction visible when sharing the sheet.
1. Check access and prepare the columns
The native AI function requires an eligible Google Workspace or Google AI plan. Administrator and language settings can also affect access. Check Google’s current eligibility guidance and supported languages before starting.
In a new sheet, enter these headers: A1: Note_ID, B1: Note, C1: AI_suggestion, D1: Human_final, and E1: Checked. Keep B unchanged so reviewers can compare every suggestion with its source. Leave D empty and write Pending in E until review is complete.
The four labels describe evidence in the note. Shipped requires explicit carrier handover or dispatch. Delayed requires an explicit postponement or stock delay. Cancelled requires a confirmed, completed cancellation. Use Needs review for missing, unclear, conflicting, unsupported, or pending information.
2. Try five fictional notes first
Copy the IDs and raw notes below into A2:B6. The final column contains original human reference answers for this exercise; these are editorial examples, not observed Gemini results. Keep them outside the input range when generating suggestions.
| ID | Raw note | Human reference |
|---|---|---|
| DEMO-001 | Handed to the carrier this morning. | Shipped |
| DEMO-002 | Waiting for stock; dispatch postponed until Friday. | Delayed |
| DEMO-003 | Cancellation completed before dispatch. | Cancelled |
| DEMO-004 | Customer asked to cancel; no approval yet. | Needs review |
| DEMO-005 | Label printed; courier pickup not confirmed. | Needs review |
On a narrow screen, swipe the table sideways to see every column.
The last two rows test whether the classifier distinguishes a request or preparation step from a completed event. A reasonable-sounding label is insufficient if the note does not support it. Needs review is a useful outcome, not a failed attempt to choose another label.
3. Add the shared prompt and generate suggestions
Open the shared order-note classification prompt. Copy its full prompt text. In C2, use the formula template below, replacing the entire uppercase placeholder with that text inside the straight double quotation marks. Keep those quotation marks around the inserted prompt. The placeholder itself is not a runnable classifier.
=AI("PASTE_SHARED_PROMPT_HERE",B2)This example uses the English function name and a comma separator. Use the argument separator your sheet requires if needed. B2 supplies the note as range context; the function does not automatically receive the entire sheet. The requested result is one exact text label.
Copy the completed formula down through C6, checking that each row refers to its own B cell. Select C2:C6 and click Generate and Insert. Google’s AI function instructions describe this generation step and Refresh and Insert for regeneration. Treat generation as an explicit action, not continual automatic recalculation. Keep AI as its own formula rather than nesting it inside IF.
4. Review the source before accepting a label
Read all five raw notes alongside their suggestions. Check both the meaning and the exact label: extra explanation or a different spelling needs correction. Enter your accepted label in D only after checking the note, then replace Pending with Checked in E. The human reference column above lets you inspect disagreements without assuming the AI suggestion is correct.
If a note or the prompt changes, write Pending in E, use Refresh and Insert, and review the final label again. A previous check does not validate a revised input. Neither C nor D should be treated as live fulfillment data or used to update orders automatically.
If the native function is unavailable, use the same notes and definitions for manual classification in D. For this small practice set, compare your choices with the human references and write Checked in E after checking. For a different task involving themes across feedback, see analyzing customer feedback with ChatGPT.
Photo: Bench Accounting / Unsplash.
