Findings
Measured from the outside, not marketed.
Pre-publication figures from the program’s frozen analyses. Both papers are in preparation; the published versions take precedence over anything shown here.
Key result
Price steering by seed-price band.
How much pricier are recommendations than the product being viewed? Mean price delta of recommendations, % above the viewed product, by the price band of the seed product.
Seed price band
Shopify · wave 1
BigCommerce · census
Blocks · Shopify
Under $25half-open band · winsorized [−100%, +200%]
+50.0%
+35.7%
174 blocks
$25–75half-open band · winsorized [−100%, +200%]
+19.1%
+24.1%
222 blocks
$75–200half-open band · winsorized [−100%, +200%]
+1.8%
+8.7%
119 blocks
Over $200half-open band · winsorized [−100%, +200%]
+12.8%
+3.0%
106 blocks
Blocks with comparable prices: Shopify 621 · BigCommerce 2,648. Pre-publication figures from the program's frozen analyses; both papers are in preparation.Cross-checked with article drafts · Sep 2026
Central finding
The cheaper the product a shopper starts from, the harder the algorithm pushes them toward pricier recommendations.
Replicated across two platforms: the direction and shape reproduce; between-category differences do not.
Spec sheet
Metrics first. Adjectives never.
Price steering
Mean price delta · seed under $25+50.0% · +35.7%
Seed $25–75+19.1% · +24.1%
Seed $75–200+1.8% · +8.7%
Seed over $200+12.8% · +3.0%
Blocks with comparable prices621 · 2,648
Upsell share0.37 · 0.35
Popularity loop
Top-20 overlap (POP_OVERLAP)0.26 · 0.21
Stores with a reliable best-selling sort205 · 881
Brand lock-in
Single-brand blocks (BRAND_HHI = 10,000)48% · not exposed
Blocks · stores389 of 806 · —
Representation
Coded blocks · fashion & beauty214 · 214
Blocks with more than one age group1 of 118 with an index
Accessibility
Stores with cards lacking alt text22.6% · 5.8%
Cards lacking alt text13.6% · 3.7%
Exposure
Live stores with no observable recommendation layer25% · —
Blocks covered by marketing pop-ups14% · 2%
Questions
Questions we get asked.
Do you work with the platforms?+
No. We audit from the outside — the way a shopper meets the shelf. No API access, no partnerships, no permission required.
Can I rerun the measurement?+
Yes. Sampling rules, seed selection, capture procedure, metric definitions and the visual-representation codebook are published on GitHub under CC BY 4.0. Any researcher can rerun it on the same platforms or on one we have not reached.
Are individual stores named?+
No. Stores are anonymized by category and catalog size, and datasets are published without URLs. Extreme cases are reported as examples, never as identifiable storefronts.
Where are the data?+
Anonymized datasets are deposited on Zenodo with DOIs alongside each publication — September 2026. Capture losses are reported, never silently dropped.
Are the figures on this site final?+
They are pre-publication figures from the program’s frozen analyses. Both papers are in preparation; the published versions take precedence over anything shown here.
Open science
Rerun the shelf yourself.
The protocol, codebooks and datasets are open. Take the measurement to a platform we have not reached yet — or write to the lab.
Protocol v2.1 · CC BY 4.0Datasets with DOIs · ZenodoNo platform cooperation