Shopping Feed Testing: A/B Test Your Way to Better Google Shopping Performance

Most ecommerce advertisers set up their product feed once and forget it. Meanwhile, their competitors are systematically testing and improving every element. The difference compounds over time.

The Testing Advantage

Feed testing isn't about making random changes and hoping for the best. It's about systematically identifying what drives performance and scaling those insights across your catalog.

What Can Be Tested?

Testable feed elements include:

• Product titles (structure, keywords, length)
• Product descriptions (content, format, length)
• Images (main image, lifestyle vs. product shots)
• Pricing (price points, sale pricing)
• Custom labels (segmentation effectiveness)
• Product attributes (colors, materials, categories)

The Compounding Effect

Consider these modest improvements:

• 10% better CTR from title optimization
• 15% better conversion from image testing
• 10% better ROAS from price testing

Combined: 1.10 × 1.15 × 1.10 = 39% overall improvement

Small wins in each area multiply into major gains.

Why Most Brands Don't Test

Common barriers:

• "It's too complicated"
• "We don't have the tools"
• "We don't know what to test"
• "It takes too long to see results"

This guide solves all of these.

Valid feed testing requires proper experimental design. Here's how to run tests that produce actionable insights.

The A/B Testing Framework

1. Hypothesis Formation State clearly what you're testing and why: "Changing title format from [Brand + Product] to [Product + Benefit + Brand] will increase CTR because shoppers see the benefit first."

2. Variable Isolation Test one element at a time:

• Bad: Change title AND image AND price
• Good: Change only title structure, keep everything else constant

3. Statistical Significance Ensure enough data for valid conclusions:

• Minimum: 100 clicks per variant
• Better: 500+ clicks per variant
• Ideal: 1,000+ clicks per variant

4. Time Control Run variants simultaneously to control for:

• Day-of-week effects
• Seasonal variations
• Competitive changes

Traffic Splitting Methods

Method 1: Random Product Assignment Split products randomly between control and test groups.

• Pro: Simple to implement
• Con: Product differences can skew results

Method 2: Custom Label Segmentation Use custom labels to route traffic.

• Create label: "title_test_v1" vs. "title_test_v2"
• Route to different campaigns
• Compare performance

Method 3: Time-Based Rotation Alternate between versions.

• Week 1: Version A
• Week 2: Version B
• Week 3: Version A
• Week 4: Version B
• Con: Time-based factors can confuse results

Recommended Approach

For most tests, use custom label segmentation:

1. Assign products to test groups randomly
2. Apply different treatments via supplemental feed
3. Create separate campaigns or ad groups per label
4. Run simultaneously for 2-4 weeks
5. Analyze results with statistical significance calculator

Product titles are the highest-impact element to test. Small changes can yield significant CTR improvements.

Title Structure Tests

Test 1: Brand Position

• Control: "Nike Air Max 90 Running Shoes Men's"
• Variant A: "Men's Running Shoes Nike Air Max 90"
• Variant B: "Running Shoes Air Max 90 by Nike"

Hypothesis: Leading with product type may increase CTR for non-branded searches.

Test 2: Benefit Inclusion

• Control: "Vitamix 5200 Blender"
• Variant: "Vitamix 5200 Blender - Makes Hot Soup in 5 Minutes"

Hypothesis: Benefit-focused titles increase CTR for problem-aware shoppers.

Test 3: Specificity Level

• Control: "Running Shoes"
• Variant: "Men's Cushioned Road Running Shoes Size 10-13"

Hypothesis: More specific titles attract more qualified clicks.

Title Length Tests

Short vs. Long

• Control: "Wireless Headphones" (18 chars)
• Variant: "Wireless Bluetooth Headphones with Active Noise Cancellation 30hr Battery" (72 chars)

Hypothesis: Longer titles capture more long-tail queries.

Mobile Optimization Test

• Control: Full 150-character title
• Variant: Front-load key info in first 70 characters

Hypothesis: Mobile-optimized titles improve mobile CTR.

Keyword Placement Tests

Test: Primary Keyword Position

• Control: "Men's Nike Air Max 90 Sneakers"
• Variant: "Sneakers Nike Air Max 90 Men's"

Measure impact on impression share for "sneakers" queries.

Common Title Testing Insights

From thousands of tests, these patterns emerge:

• Primary keywords in first 70 characters: +5-15% CTR
• Benefits included: +10-20% CTR for consideration-stage shoppers
• Specific sizes/colors: Higher conversion rate, lower total impressions
• Brand-first: Better for branded searches, worse for generic

Images determine whether shoppers click. Yet most brands never test them.

Image Type Tests

Test 1: White Background vs. Lifestyle

• Control: Product on white background
• Variant: Product in lifestyle context

Hypothesis: Lifestyle images increase emotional connection and CTR.

Test 2: Single Product vs. Bundle Display

• Control: Single product image
• Variant: Product shown with accessories/complementary items

Hypothesis: Bundle images increase perceived value and CTR.

Test 3: Model vs. No Model

• Control: Clothing on mannequin/flat lay
• Variant: Clothing on human model

Hypothesis: Model images help shoppers visualize fit.

Image Angle Tests

Test: Primary Angle

• Control: Front view
• Variant A: 45-degree angle
• Variant B: Multiple angles in one image

Measure which angle drives higher CTR.

Image Quality Tests

Test: Resolution and Clarity

• Control: Standard quality image
• Variant: High-res, professionally lit image

Hypothesis: Quality images signal quality products.

Image Text Overlay Tests

Test: Text on Images

• Control: No text overlay
• Variant A: "Best Seller" badge
• Variant B: "Free Shipping" overlay
• Variant C: Price displayed

Note: Check Google's image policies—some text is prohibited.

Implementing Image Tests

Using Supplemental Feeds

1. Upload alternate images for test products
2. Use image_link override in supplemental feed
3. Split products into control/test groups

Using Multiple Product Variants

1. Create color/style variants with different images
2. Compare performance across variants
3. Roll winning image style to all products

Image Testing Best Practices

• Run tests for at least 14 days
• Need 500+ clicks per variant for reliable results
• Test during stable periods (not during sales/holidays)
• Document winning patterns for future uploads

Price significantly impacts Shopping performance. Strategic testing reveals optimal price points.

Price Point Tests

Test 1: Psychological Pricing

• Control: $50.00
• Variant A: $49.99
• Variant B: $49.97

Hypothesis: .99 endings increase perceived value and conversions.

Test 2: Round Numbers

• Control: $49.99
• Variant: $50.00

Hypothesis: Round numbers may signal quality in premium categories.

Test 3: Threshold Testing

• Control: $52.00
• Variant: $49.00

Hypothesis: Breaking below $50 threshold increases conversions enough to offset lower price.

Sale Price Testing

Test: Sale Price Display

• Control: Regular price only ($80)
• Variant: Sale price with strikethrough ($80 → $65)

Hypothesis: Visible discount increases CTR and conversion.

Test: Discount Percentage

• Control: 10% off ($90 → $81)
• Variant A: 15% off ($90 → $76.50)
• Variant B: 20% off ($90 → $72)

Find the optimal discount that maximizes total profit.

Promotional Messaging Tests

Test: Merchant Promotion Types

• Control: No promotion
• Variant A: "Free Shipping"
• Variant B: "10% Off First Order"
• Variant C: "Free Gift with Purchase"

Measure which promotion type drives highest conversion.

Price Testing Implementation

Method 1: Product Segmentation Split catalog into groups with different pricing.

Method 2: Geographic Testing Test different prices in different states/regions.

Method 3: Time-Based Testing Alternate prices weekly (less reliable due to time effects).

Price Sensitivity Analysis

Build a price-volume curve:

• Test price A: 100 units sold, $50 each = $5,000
• Test price B: 80 units sold, $60 each = $4,800
• Test price C: 60 units sold, $70 each = $4,200

Optimal price = maximum revenue (not always lowest price).

Promotion Fatigue

Monitor for:

• Decreased effectiveness over time
• Customers waiting for sales
• Brand perception impact

Rotate promotion types to maintain effectiveness.

Descriptions influence query matching and conversion. Here's how to test them effectively.

Description Format Tests

Test 1: Paragraph vs. Bullet Points

• Control: Traditional paragraph format
• Variant: Bullet-point feature list

Hypothesis: Scannable bullets improve matching and conversion.

Test 2: Length Optimization

• Control: Short description (200 characters)
• Variant A: Medium (500 characters)
• Variant B: Long (1,500+ characters)

Hypothesis: Longer descriptions capture more long-tail queries.

Description Content Tests

Test: Feature-Focused vs. Benefit-Focused

• Control: "Made from 100% organic cotton, 180 GSM weight"
• Variant: "Soft, breathable comfort all day. No itching, no shrinking."

Hypothesis: Benefit-focused descriptions improve conversion.

Test: Technical vs. Lifestyle Language

• Control: Product specifications emphasis
• Variant: Use-case and lifestyle emphasis

Measure impact on different audience segments.

Keyword Integration Tests

Test: Keyword Density

• Control: Natural keyword use
• Variant: Strategic keyword inclusion (every 100 words)

Hypothesis: Strategic keywords improve query matching without sounding spammy.

Test: Synonym Variation

• Control: "Shoes" used throughout
• Variant: Mix of "shoes," "sneakers," "footwear," "runners"

Hypothesis: Variation captures more search queries.

Description Testing Process

1. Select 50-100 products for testing
2. Create alternate descriptions via supplemental feed
3. Split into control/test using custom labels
4. Run for 3-4 weeks minimum
5. Measure: Impressions, clicks, conversions, revenue

Measuring Description Impact

Since descriptions affect matching more than clicks:

• Monitor impression volume changes
• Track search query breadth
• Measure long-tail query performance
• Compare conversion rates

Custom labels enable sophisticated testing and optimization. Here's how to test segmentation strategies.

Segmentation Strategy Tests

Test 1: Performance-Based Segmentation

• Control: All products in single campaign
• Variant: Products segmented by ROAS tier (high/medium/low)

Hypothesis: Segmented campaigns with tailored bids improve overall ROAS.

Test 2: Price Tier Segmentation

• Control: Unified campaign structure
• Variant: Campaigns by price tier (luxury/premium/value)

Hypothesis: Price-appropriate bid strategies improve efficiency.

Test 3: Margin-Based Segmentation

• Control: ROAS-optimized bidding
• Variant: Profit-optimized bidding using margin labels

Hypothesis: Margin-aware optimization improves actual profit.

Label Attribution Testing

Test: Which Labels Predict Success? Run correlation analysis:

• Does "bestseller" label predict ROAS?
• Does "high-margin" label predict profit?
• Does "new-product" label predict conversion rate?

Refine labels based on predictive power.

Campaign Structure Tests

Test 1: Granularity Level

• Control: Single Shopping campaign
• Variant A: Campaign per category
• Variant B: Campaign per brand
• Variant C: Campaign per performance tier

Measure: Management efficiency vs. performance gains.

Test 2: Bid Strategy by Segment

• Control: Max Conversions for all
• Variant: Target ROAS for bestsellers, Max Clicks for new products

Hypothesis: Segment-appropriate strategies outperform one-size-fits-all.

Testing Label Criteria

Test: Label Thresholds

• What ROAS threshold defines "top performer"?
• Test: Top 10% vs. top 20% vs. top 30%
• Measure which cutoff creates most actionable segments

Test: Recency of Data

• Should labels be based on 7-day, 30-day, or 90-day data?
• Test different lookback windows
• Find balance between recency and stability

Automation Testing

Test automated label updates:

• Daily updates vs. weekly updates
• Simple rules vs. ML-based classification
• Measure: Accuracy of predictions, management overhead

The right tools make feed testing manageable at scale.

Feed Management Platforms

DataFeedWatch

• A/B testing features built-in
• Rule-based feed modifications
• Performance analytics
• Best for: Mid-size catalogs (100-10,000 SKUs)

Feedonomics

• Enterprise-grade testing
• Advanced analytics
• Multi-channel support
• Best for: Large catalogs (10,000+ SKUs)

Channable

• Visual feed builder
• Easy A/B test setup
• Strong automation
• Best for: Multi-marketplace sellers

Google Merchant Center Native

Feed Rules

• Built-in transformation rules
• No additional cost
• Limited but functional
• Best for: Simple tests, small catalogs

Supplemental Feeds

• Override main feed attributes
• Free and powerful
• Best for: Title/description testing

Spreadsheet-Based Testing

For smaller operations:

1. Export product data
2. Create test variations in spreadsheet
3. Upload as supplemental feed
4. Track performance manually

Tools: Google Sheets + Google Ads Scripts

Analytics and Statistical Tools

Google Ads Reports

• Segment by custom labels
• Export to analyze in spreadsheet
• Limited statistical testing

Optimizely/VWO

• Professional A/B testing calculators
• Statistical significance determination
• Bayesian analysis options

Custom Analytics

• Build dashboards in Looker Studio
• Connect Google Ads API data
• Create automated significance testing

Essential Testing Stack

Minimum viable:

• Google Sheets for test planning
• Supplemental feeds for variations
• Google Ads reports for results
• Online significance calculator

Recommended:

• Feed management tool (DataFeedWatch/Feedonomics)
• Looker Studio dashboard
• Google Ads Scripts for automation

Running tests is half the battle. Proper analysis determines whether you get actionable insights or misleading data.

Statistical Significance

Why It Matters Without statistical significance, you can't know if results are real or random chance.

The Basics

• 95% confidence = 5% chance results are random
• Minimum: 95% confidence level
• Better: 99% for major decisions

Sample Size Requirements

For a test comparing 3% CTR (control) vs. 3.3% CTR (variant):

• Minimum sample: ~5,000 impressions per variant
• For conversion tests: More volume needed due to lower rates

Use a sample size calculator before starting tests.

Analyzing CTR Tests

Metrics to Compare

• CTR: Primary metric for titles/images
• Impression share: Did the change affect matching?
• Click quality: Did better CTR lead to worse conversion?

CTR Analysis Process

1. Calculate CTR for control and variant
2. Check statistical significance
3. Look at absolute improvement (not just relative)
4. Verify no negative secondary effects

Analyzing Conversion Tests

Metrics to Compare

• Conversion rate
• Average order value
• Revenue per click
• ROAS

Common Pitfalls

• Higher CTR but lower conversion (bad clicks)
• Better conversion but lower volume (niche appeal)
• Short-term lift that doesn't sustain

Revenue Impact Calculation

For a title test that increased CTR from 2% to 2.4%:

• Monthly impressions: 100,000
• Old clicks: 2,000
• New clicks: 2,400
• Additional clicks: 400
• Conversion rate: 3%
• Additional conversions: 12
• Average order value: $75
• Additional revenue: $900/month = $10,800/year

Test Documentation

For every test, record:

1. Hypothesis
2. Test period and sample size
3. Control and variant definitions
4. Results with confidence levels
5. Decision (implement, iterate, abandon)
6. Learnings for future tests

Build a systematic testing program with this quarterly roadmap.

Quarter 1: Foundation Tests

Month 1: Title Structure

• Test brand position (beginning vs. end)
• Test keyword placement (primary keyword first)
• Winner: Roll out to 20% of catalog

Month 2: Image Optimization

• Test main image style (white vs. lifestyle)
• Test image quality (enhanced vs. standard)
• Winner: Update product photography guidelines

Month 3: Price Testing

• Test psychological pricing (.99 vs. .00)
• Test sale price display effectiveness
• Winner: Update pricing strategy

Quarter 2: Advanced Testing

Month 4: Description Optimization

• Test description length
• Test format (paragraph vs. bullets)
• Winner: Create description templates

Month 5: Segmentation Testing

• Test performance-based labels
• Test margin-based optimization
• Winner: Implement new campaign structure

Month 6: Promotion Testing

• Test merchant promotion types
• Test promotional title messaging
• Winner: Define promotion playbook

Quarter 3: Scaling Winners

Month 7: Full Catalog Title Rollout

• Apply winning title formula to all products
• Monitor for unexpected results
• Iterate on edge cases

Month 8: Image Refresh

• Apply image learnings to new photography
• A/B test new images vs. old
• Scale winning styles

Month 9: Advanced Segmentation

• Implement refined label strategy
• Test automated bid adjustments
• Measure efficiency gains

Quarter 4: Continuous Optimization

Month 10: Second-Order Tests

• Test combinations of previous winners
• Look for interaction effects
• Optimize the optimizations

Month 11: Seasonal Testing

• Test holiday-specific variations
• Measure seasonal effectiveness
• Build holiday playbook

Month 12: Year Review

• Document all learnings
• Calculate cumulative impact
• Plan next year's testing roadmap

Ongoing Habits

Weekly:

• Review active test results
• Check for statistical significance
• Document observations

Monthly:

• Start new test
• Close completed tests
• Update feed based on learnings

Quarterly:

• Review testing program effectiveness
• Adjust testing priorities
• Share learnings with team

Use this checklist to launch your feed testing program.

Week 1: Setup

• Choose feed management tool (or plan supplemental feed approach)
• Set up Looker Studio dashboard for test monitoring
• Create test documentation template
• Define custom labels for testing (test_group_a, test_group_b)
• Identify first test hypothesis

Week 2: First Test Launch

• Select 100-200 products for test
• Randomly assign to control/variant groups
• Create variant content (titles, images, etc.)
• Upload via supplemental feed
• Set up tracking in Google Ads
• Document test in tracking sheet

Week 3-4: Monitoring

• Daily: Check for anomalies or errors
• Weekly: Review preliminary results
• Track impressions and clicks accumulation
• Note any external factors (sales, competitor changes)

Week 5: Analysis

• Calculate results with confidence intervals
• Document statistical significance
• Analyze secondary metrics
• Make implementation decision
• Plan rollout of winners

Ongoing Test Management

• Maintain test calendar (one test at a time per element)
• Track cumulative improvements
• Share learnings with broader team
• Continuously generate new hypotheses

Test Prioritization Matrix

Score potential tests on:

• Impact potential (1-10)
• Ease of implementation (1-10)
• Risk level (1-10, higher = riskier)

Priority Score = (Impact × 2) + Ease - Risk

Focus on high-impact, easy, low-risk tests first.

Success Metrics

Track testing program health:

• Tests completed per month
• Win rate (% of tests with positive results)
• Cumulative performance improvement
• Time to implement winners