The useful result
Store one checked classification for each synthetic submission ID, even when the same input arrives twice.
- Store one checked classification for each synthetic submission ID, even when the same input arrives twice.
- Inspect the supplied example and identify a result that fails the task.
n8n combines AI components with ordinary workflow logic. The important trick is to keep generated classification separate from deterministic validation and duplicate handling. This lesson uses synthetic leads and a test store; it does not send marketing messages, enrich personal data or connect a payment workflow.
Reviewed on 9 October 2026 against the linked official sources. These instructions teach a bounded practice task; account features and limits can differ. Expected outputs are illustrative rather than captured AI runs. Keep original inputs and verify the result before using it.
Before you start
- A test n8n workspace and model connection you are authorised to use
- A test store that can enforce unique submission IDs
- Three synthetic lead records; no real contact data
Review evidence
Documentation-based review · 2026-10-09
Official capabilities checked; exercise prompts and expected outputs are original illustrative teaching material.
Define the input contract
Prepare three records with a submission ID and a service request. Define exactly three allowed labels: design, development and other. Keep an empty or ambiguous request on a review route rather than forcing a confident classification.
Example / expected result
Illustrative example (not a recorded model result): demo-104: redesign a homepage; demo-105: fix a login bug; demo-104 delivered again.
Check your result
Confirm duplicate records share the same ID and that no email address or private customer data is needed.
Use AI only for the interpretation step
Configure a documented model or agent step to classify the service text. Keep output small and structured. Provider charges are separate from workflow access; use the smallest test batch.
Classify this supplied service request as design, development or other. Return submissionId, label and evidence phrase. If insufficient information, use other and mark needsReview true. Do not send messages or invent a budget.Example / expected result
Illustrative example (not a recorded model result): demo-104 → design, with evidence “redesign a homepage.”
Check your result
Parse the output and check the exact ID, allowed label and evidence phrase against the input.
Enforce validation and uniqueness outside the model
Use ordinary workflow conditions to reject malformed output. Write to a test store with a durable uniqueness rule on submission ID; an in-memory prompt instruction cannot prevent concurrent duplicates. Route a conflict to already processed or review.
Example / expected result
Illustrative example (not a recorded model result): First demo-104 inserts one draft row; the repeated ID does not insert another.
Check your result
Inspect the actual store: exactly one demo-104 row. Test concurrent duplicate delivery if your integration can retry in parallel.
Exercise the failure routes
Run the normal, duplicate and ambiguous examples. Add a malformed label in a controlled test to confirm it is rejected. Inspect execution data and keep downstream send or publish actions disabled.
For a request containing only “help,” return other and needsReview true. Do not guess a service or customer intent.Example / expected result
Illustrative example (not a recorded model result): The ambiguous request waits for review; a label outside the allowed set is rejected.
Check your result
Check counts, stored IDs, review flags and logs. Confirm no email or external business write was enabled.
When it goes wrong
A prompt asking for no duplicates does not enforce database uniqueness. Use durable storage and an actual unique-ID rule. If a provider times out, preserve the ID across retries. If the model returns an invalid label, reject it rather than silently mapping it to a business action.
Make it your own
Send five synthetic submissions with two repeats and one ambiguous request. Confirm the store contains one row per valid unique ID and a visible review route.
Useful next resources
Sources and review
Last checked Oct 9, 2026. Features and account limits can change. Verify important results against your original material.
Did this help you complete your task?
Tell us what needs improving. No account or private output required.
Report incorrect information or a broken linkGenius Forges Editorial
Practical, documentation-reviewed AI exercises with explicit examples and independent result checks.
