Meta's Sept 3 advertiser results: test the claims on your own ads

Meta's September 3 page gives five results from advertisers using its AI ad tools, including a 64% lower cost per purchase and a 4.4% lift in vehicle sales, and every figure is stated by Meta on behalf of the advertiser. To find out whether the claim holds for you, build matched creative variants, run them against each other in Meta's own A/B test tooling and compare cost per result, not Meta's headline numbers.

The figures come from Meta's How Businesses Are Driving Results with Meta's AI-Powered Ads, dated September 3, 2026 and read on 2026-10-03. Sume has no data on these advertisers and cannot confirm the numbers.

What did Meta actually publish?

The page names Advantage+ Sales Campaigns, Advantage+ Creative with generative AI, Partnership Ads with creators, Meta Business Agent and Omni for Automotive, and lists one result for each group of tools.

Why can't I copy those numbers?

The page is a marketing summary. It does not give the test design, the spend, the length of the test or the baseline, so a 64% lower cost per purchase could start from a high or a low cost. The advertisers also work in different industries and regions. Treat each line as a direction to test, not a forecast.

How do I build a fair creative test?

Change one thing at a time. If you want to know whether AI-generated creative helps, keep the offer, audience, budget and placement fixed and swap only the creative.

A simple setup is two or three variants of the same message. One is your current creative. The others are generated or edited versions, such as a new hook or a different 9:16 clip made with the Video Router. Sume's bulk runs queue up to 100 Format runs with a concurrency window, which is useful when you want the same treatment applied to a list of products.

curl -X POST https://api.sume.com/v1/video-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: variant-b-hook-01" \
  -d '{
    "model": "seedance-2.5",
    "prompt": "A vertical UGC-style product clip on a desk, natural light",
    "resolution": "720p",
    "duration": 8,
    "aspect_ratio": "9:16",
    "mode": "async"
  }'

Where does the test itself run?

In Meta. Meta's Ads MCP Server page lists seven tool categories, including one for A/B tests and conversion lift studies and one for comprehensive reporting. Sume does not run ad tests, measure lift or read your ad account.

What Sume does is produce the variants and keep the job records. Poll each job as described in Jobs and results, and note which job id made which file so that a winning ad can be rebuilt later.

What would count as a result?

Decide the measure before the test: cost per purchase, cost per subscription or purchases per spend. Run it long enough for your account to collect a stable number, and keep the losing variant's file. If your own result differs from Meta's page, that is normal; the page describes five advertisers, not yours.