EcomAgentTools product-description test kit — 2026-10-28

These 15 SKUs are FICTIONAL TEACHING DATA, not real inventory or completed tool tests.
Reconstructed from the supplied research report because its separate sandbox attachments were not supplied.
No result, error-rate or timing measurements have been generated.

Start with one SKU per family. Use identical facts and output requirements across tools.
Import ONLY synthetic-sku-input-15.csv into a separate test catalog. Map test_sku to SKU and product_title to title.
Map the source/variant columns explicitly. Put unknown_facts/prohibited_claims in instructions or appropriate constraints;
do not expose them as shopper-facing product fields. Keep expected-checks.csv reviewer-only.
Do not upload this kit to a production store or publish the fictional products.
The defaults use English and Shopify formatting; they are not evidence of marketplace eligibility.
The shirt intentionally has no certification; do not label it certified organic.
For the vase example, BLU and SND are approximately 8 x 4.5 inches; MOS is approximately 10 x 5 inches.

Use prompt.txt as a baseline, adapting the input method to each product and recording those differences.
Turn off enrichment/image inference for the baseline. Run separately if evaluating those features.
Save raw output before editing. Add run IDs, exact settings and references to test-record-template.csv.
Fill review-log-template.csv only after an actual run; empty values mean unmeasured, not zero.
Suggested issue types: factual_error, unsupported_claim, missing_required_field, variant_confusion,
source_fact_omitted, channel_format_issue, tone_only_edit, mapping_error, policy_review_required.
Suggested statuses: blocked_missing_source_fact, generation_failed, rejected, approved_after_edit, approved.
Measure review time from opening output to approval/rejection; exclude breaks/system waits and record setup separately.

Optional second run: change ONE known source fact, save a versioned input file and rerun only that SKU.
Verify the changed source value reached the tool before evaluating the regenerated text.
An import option that fills only empty fields does not repair an incorrect populated field.
Record reruns under new IDs. Compare exported SKU identifiers, columns, units and encoding before destination import.

These checks are an editorial evaluation method, not vendor or marketplace certification.
The article includes current official workflow sources:
https://ecomagenttools.com/blog/ai-product-description-generators
