> This page is for Product.

> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.feedonomics.com/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.feedonomics.com/_mcp/server.

# A/B testing enriched content

> How to run a controlled split test of enriched content on your feed, including group splitting, custom labels, and the caveats that protect your results.

The most convincing case for enriched content is a controlled split test on your own feed. This playbook walks through running one with the tools already in your Feedonomics account. You own the test and the analysis; this is the method.

> **Before you start**
>
> You'll need your paid media data connected and your products tagged, since group segmentation relies on custom-label reporting in Google Ads — see [Analytics onboarding](/product/analytics/onboarding) and [Measuring impact in Analytics](/product/enrichment/reports). If you're not ready for a full split test, the simpler baseline-and-holdout approach in [Proving enrichment works](/product/enrichment/measuring-impact) is a reasonable first step.

## The method at a glance

![Diagram of the split test: test products are hashed per product and split fifty-fifty into a control group with original content and a test group with enriched content. The group is written to a custom label in the export, and Google Ads reporting compares CTR, conversion rate, and ROAS per group over two to four weeks.](/_fern-files/feedonomics.docs.buildwithfern.com/10dd3082b2b1bc688a57d14f87b4b2a7bff4fda9db95e466d7add23f54dafbd8/docs/assets/enrichment/ab-test-split.svg)

### Form a hypothesis

One variable, one expected outcome: "Enriched titles will improve CTR on our Google Shopping traffic." Titles are the classic first test — they're the highest-impact field.

### Select your test products

Big enough for statistical significance, similar enough to isolate the variable. One healthy category beats a random scatter across your whole catalog.

### Split 50/50

Assign each product a random group with transformers: generate a stable hash per product, then split on it — half to Control (A, original content), half to Test (B, enriched content). Keep variants of the same product in the same group so a single item group never straddles both arms.

### Label the groups

Write the group assignment to a custom label in your export so you can segment by it in Google Ads reporting.

### Run for 2–4 weeks

Track CTR, conversion rate, and ROAS per group per week. Don't call it early — wait for significance.

### Decide and repeat

Winner ships to the full catalog. Then move to the next hypothesis: descriptions are the natural second test after titles.

## Caveats that save tests

* **Mind seasonality.** Slow weeks depress everything; sale weeks inflate ROAS. Note anything unusual in the window.
* **Analyze Shopping campaigns, not Performance Max.** PMax blends surfaces at the campaign level and will muddy your read.
* **Freeze campaign settings.** Don't change budgets or ROAS targets mid-test — you'd be testing two things at once.
* **Check your conversion tracking** is revenue-based before you start, or ROAS comparisons are meaningless.

> **This is not Google Experiments**
>
> Google's native Experiments feature tests *campaign settings*. This playbook tests *feed data*. They're complementary — just don't run both on the same traffic at the same time and expect a clean read.

## After the test

A winning test tells you which field to enrich next, and in what order. Titles first, then descriptions, then the detail fields — [Where enriched fields go](/product/enrichment/field-mapping) shows what's available per channel. To expand a winner across your catalog, run a larger batch with the same config: see [Re-running enrichments](/product/enrichment/re-running).