How many ad variants per product? Google says test gradually

Google's October 1, 2026 YouTube ads post says to test gradually and to favor quality over quantity. It gives no number, so any variant count is your call. A defensible start is 3 variants per SKU: enough to compare, few enough that a person reviews each one. Sume's bulk run takes up to 100 items per request, so 33 SKUs at 3 variants fit in one.

Google's advice is from its Ads & Commerce post, read 2026-10-01. Sume facts are from Bulk runs.

How do the numbers work?

A bulk request is 1 to 100 items and the arithmetic is simply SKUs times variants. Each item is an ordinary Format run, so you pay per run, not per request.

How do I label variants so I can compare them?

Put the SKU and the angle in each item's instruction, so the finished run tells you what it was. Poll GET /v1/format-run-queues/{queue_id} and read counts.failed before you trust a completed status.

curl -sS -X POST "https://api.sume.com/v1/formats/YOUR_HANDLE/product-promo/bulk-runs" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $(uuidgen)" \
  -d '{
    "concurrency": 4,
    "items": [
      { "instruction": "SKU-100 angle A: close-up", "input": { "url": "https://example.com/100.jpg" } },
      { "instruction": "SKU-100 angle B: in use", "input": { "url": "https://example.com/100.jpg" } },
      { "instruction": "SKU-100 angle C: gift", "input": { "url": "https://example.com/100.jpg" } }
    ]
  }'

What does gradually mean in practice?

Ship the first batch to a small budget, read the results in Google's reporting, then queue the next round for the winners only. Sume makes the files; it does not see how an ad performs. Confirm Google's current advice on its page. For the per-ad pipeline see one ad into six variations.