Product review Shorts per SKU: vary the focus, not just the name

YouTube's channel monetization page lists "a character series or review channel where each video has a distinct storyline, focus, or concept" among allowed content, and lists minimal variation across videos, and the same situation and outcome repeated, as not allowed (read 2026-10-03). A per-product Short series therefore passes or fails on one thing: whether each Short has its own focus, not whether it has its own product name.

The October 1 recommendation change adds pressure on template-based bulk changes. Search Engine Journal reports YouTube named them, with voice-over descriptions and minor technical edits, as not making a clip original (read 2026-10-03). That was said about other creators' clips, but the same logic is the sensible bar for your own batch.

What to vary per row

If you review 30 products, the row data should contain more than a name, a price and a photo. Put a finding in each row, something you actually did with the product, and let that finding decide the angle.

Running the batch with Formats

Format bulk runs queue up to 100 runs of one Format at a concurrency of 1 to 16. Each entry in items is the same body as a single run, so every row can carry its own instruction and input. Send a fresh Idempotency-Key for each new batch, and poll GET /v1/format-run-queues/{id}.

One detail from the docs saves a bad publish: a queue with status completed only means every item reached a terminal state, so look at the failure count before treating the batch as done. There is no queue-level webhook, so poll.

The generator below builds the items list from rows that carry a finding and refuses a row without one. Whatever Format you point it at, the instruction text per row is yours; what to put in the Format body is covered in the Formats call docs.

def build_items(rows):
    items = []
    for r in rows:
        if not r.get("finding"):
            raise ValueError(f"{r['sku']}: add a finding before queuing")
        items.append({
            "instruction": (f"Short about {r['sku']}. Lead with: "
                            f"{r['finding']}. Verdict for: {r['buyer']}."),
            "input": {"url": r["photo"]},
        })
    return items

rows = [{"sku": "Kettle A", "finding": "boiled 1 L in 3m10s",
         "buyer": "small kitchens", "photo": "https://example.com/a.jpg"}]
body = {"concurrency": 2, "items": build_items(rows)}
print(len(body["items"]))

Before you publish any of them

Sume does not guarantee reach. A well-built row gives you a better chance of a Short with its own purpose, but YouTube has published no threshold, start date or dashboard for the update (Search Engine Journal, read 2026-10-03).

• Read every output; a batch can return fluent text that states a finding you did not make.
• Remove claims you cannot back with your own test or a source you checked.
• Disclose paid or gifted products as the platform and your local rules require.
• Space the Shorts out instead of posting all at once.

An example of two rows that differ

Row one: a desk lamp. Finding: the dimmer steps are uneven below 20 percent. Verdict: fine for reading, poor for video calls. Comparison: the other lamp in your list that dims smoothly. Row two: a webcam. Finding: autofocus hunts in dim light. Verdict: pick it only for a lit room. Comparison: a model with a larger sensor.

The two Shorts will not share a hook, an evidence line or a closing. They may share a visual frame, which YouTube's page explicitly allows, since a consistent intro and outro with different content between is listed as acceptable (read 2026-10-03).

If a row has no finding, do not queue it. A review of a product you did not touch is a summary of a listing, and a batch of those is what a template looks like.

Cost and review

A bulk queue runs one Format run per item, and each run is billed as that Format is. Check the Format's price on its own page before queuing 100, and try two rows first. After the queue finishes, open each result; a completed queue status does not mean every row succeeded.