Product page tests: a full example report

A walk through a real product page test report: a foldable shoe cabinet on TikTok Shop US, the original page against a benefit-led version, with 25 virtual consumers and 5 sessions per variant.

AI simulation for pre-testing, not a sales forecast.

What you test

Upload screenshots of 2–6 versions of a product page, with one as the control. Versions can be different products, not just page designs, such as a 2-tier and a 3-tier shoe rack. Give each variant its own name, size, material and price, so each variant’s shoppers see the right facts.

In this example

  • A · Original page (control): foldable shoe cabinet, white, 3 tiers, plastic, 17.3 W × 13.8 D × 20.87 H in, $37.69 with free shipping
  • B · Benefit-led: the same product and price, with a free-shipping badge and 3 benefit bullets on the page
  • Shoppers browsing TikTok Shop US, not already looking to buy

Variants can be different products

A product page test can also compare different products for the same need, to help choose which one to list. Each variant carries its own size, material, features and price.

Two different home storage products on TikTok Shop US: a white 3-tier foldable shoe cabinet (plastic, magnetic doors, 17.3 W × 13.8 D × 20.87 H in) at $37.69, and a black 5-tier 3-in-1 entryway organizer (metal frame, 8 hanging hooks, 31.10 W × 10.04 D × 67.13 H in) at $59.99. Both list free shipping.

The conclusion

Every report opens with a conclusion: a decision banner, up to three short reasons and one action, always framed as a simulation. The decision is one of five outcomes:

  • Proceed to a live A/B test with a variant: it is clearly ahead of the control.
  • Refine and re-test: a variant points ahead, but not clearly.
  • Leaning toward the control: a variant looks worse, but not yet clearly.
  • No meaningful difference: the variants and the control perform about the same.
  • Keep the control: a variant clearly does worse and none clearly does better.
Example report: Shoe cabinet, “No meaningful difference found”, with the reasons and one action. The report says so instead of declaring a winner.

Key numbers and what was tested

The key numbers as tiles, then the variants side by side and the buy rate for each with its range. When the ranges overlap this much, no winner can be declared.

Example report: Shoe cabinet. 59.1% (A) vs 61.0% (B): a difference of 1.9 points, with 25 shoppers simulated.
Example report: Shoe cabinet. A, original page, and B, benefit-led: the same product and price, with a free-shipping badge and three benefit bullets on B.
Example report: A, original page: 59.1% (45.7–72.5%). B, benefit-led: 61.0% (46.7–74.8%).

Segmentation analysis

Buy rates for each variant by age, gender and more, with a range and the number of virtual consumers in each group, and a one-line insight per chart. Groups with fewer than 5 virtual consumers are flagged as a small sample.

Example report: Shoe cabinet. Female shoppers 78.5% (A) vs 76.8% (B), male shoppers 38.2% vs 43.9%.

Audience differences, concerns and questions

Where groups differ, what stopped the shoppers who did not buy, and, further on in the report, the questions shoppers ask about the control page.

Example report: Shoe cabinet. Lack of need (82–85% of non-buyers) and price (73%) were the main concerns for both pages.

Recommendations and detailed findings

Recommendations in three groups: what to do, what to improve next (with the behavioural-economics principle behind each idea) and what to check after the change. The detailed findings follow, with the overall result, the comparison and next steps. They build on this test’s own numbers and do not forecast sales.

Example report: Shoe cabinet. Keep the current page; add weight capacity, durability proof and assembly details to the page, then watch questions and returns.
Example report: Shoe cabinet. Both pages convert around 59–61% in simulation, with no meaningful difference.

Representative personas

A sample of virtual consumers with their profile, their decision on the control, the main reason and the main concern, in their own words.

Example report: Shoe cabinet. 6 personas: 3 would buy, 3 would not.

Markets, audiences and set-up

Markets

Amazon, TikTok Shop, Temu, SHEIN, Walmart and more, or your own store, in the US, the EU and Asia.

Browsing or in-market

Choose how shoppers arrive: coming across the page while browsing, or already looking to buy. Browsing suits most traffic; in-market shoppers already want this kind of product, so buy rates run higher.

Audience filters

Narrow the audience with grouped filters, such as age, gender, income, education, home, pets, family and spending.

Reports in English and 中文

Read every report in English or Chinese. This page shows the version in the language you choose.

Re-tests

Re-test a changed version and keep the screenshots you did not change.

Assistant and API

Describe the test to the Assistant: it drafts a plan, and nothing runs until you click Start. Or create, run and read tests with the Product Tests API v1 and the Python SDK 0.3.0, which also talks to the Assistant (client.agent).

Limits

  • 10–100 virtual consumers per test without a plan, up to 500 on Base and 2,000 on Pro; 3–10 sessions per variant
  • Free tests: up to 100 virtual consumers, 2 variants and 5 sessions per variant; larger tests are paid from your balance
  • 2 free tests in total, shared by both test types
  • A product page test usually takes 5–15 minutes
  • Reliable results need larger persona samples. Larger tests are paid from your balance: up to 100 personas without a plan, 500 on Base and 2,000 on Pro.

Find your stronger page before launch

2 free tests, no card needed. Add a card for a $25 credit.