Recommendation Intelligence Research™ · Atom Foundry · June 2026
We captured 4,000 AI supplement recommendations and compared them against real e-commerce stores.
The question was simple:
Does being recommended by AI have anything to do with how AI-ready a store actually is?
Across another 4,000 recommendations, the answer is again: almost not at all.
This is the second category we have analyzed. Beauty showed the same pattern. Supplements replicate it.
Key Finding
Recommendation Frequency™ and AI Commerce Score™ show no measurable relationship.
Correlation: r = -0.015 (n = 39)
In other words:
Stores that AI recommends more are not more AI-ready.
Stores that score higher on AI readiness are not recommended more often.
Recommendation behavior appears largely disconnected from store quality.
Methodology
Everything below comes from captured model outputs.
Dataset
Category: Supplements
Model: GPT-4o-mini
Shopping intents: 20
Runs per intent: 20
Total prompt-runs: 400
Recommendations captured: 4,000
Distinct brands: 371
Metrics
Recommendation Share™
A brand's share of all recommendations captured.
Recommendation Frequency™
The percentage of prompt-runs where a brand appeared at least once.
Recommendation Position™
Average ranking position when a brand appeared.
Lower is better.
Important Note About Retailers
Retailers and marketplaces were excluded from brand analysis.
Sites such as:
Amazon
iHerb
GNC
sell hundreds of brands and would distort the results.
This study measures competition between individual brands and their own stores.
Supplement Recommendation Leaderboard
| Brand | Share™ | Freq™ | Pos™ | AI Commerce Score™ | Readiness |
|---|---|---|---|---|---|
| Garden of Life | 6.7% | 66.8% | 3.6 | 52 | Low |
| Optimum Nutrition | 4.3% | 42.8% | 1.4 | 81 | Moderately |
| Cellucor | 3.2% | 31.5% | 5.8 | 67 | Low |
| NOW Foods | 3.0% | 30.3% | 5.4 | 14 | AI Invisible |
| Nature Made | 2.5% | 25.3% | 3.3 | n/a | Off-index |
| Thorne Research | 2.3% | 22.8% | 6.2 | n/a | Off-index |
| BSN | 2.1% | 20.8% | 3.3 | n/a | Off-index |
| MyProtein | 1.9% | 19.3% | 6.0 | n/a | Off-index |
| Kirkland Signature | 1.8% | 17.8% | 7.4 | n/a | Off-index |
| Dymatize | 1.7% | 16.8% | 4.5 | n/a | Off-index |
| Vital Proteins | 1.7% | 16.8% | 5.0 | 61 | Low |
| MegaFood | 1.6% | 16.0% | 5.0 | n/a | Off-index |
| New Chapter | 1.6% | 15.8% | 4.6 | n/a | Off-index |
| Kaged Muscle | 1.6% | 15.5% | 5.2 | n/a | Off-index |
| MuscleMilk | 1.5% | 14.5% | 4.0 | n/a | Off-index |
| Nature's Way | 1.4% | 13.5% | 6.3 | 14 | AI Invisible |
| Vega | 1.3% | 12.8% | 6.0 | 52 | Low |
Recommendation Frequency Does Not Follow Readiness
If AI recommended the stores that are easiest for AI to read, recommendation frequency and AI Commerce Score™ would move together.
They do not.
Across all measured single-brand stores:
r = -0.015
Statistically, that is indistinguishable from zero.
Not weak. Not small.
Simply no measurable relationship.
Examples
Highly Recommended, Poorly Built
NOW Foods
Recommendation Frequency™: 30.3%
AI Commerce Score™: 14
Nature's Way
Recommendation Frequency™: 13.5%
AI Commerce Score™: 14
Both stores sit deep inside what we classify as AI Invisible Risk™.
Yet AI recommends them frequently.
Highly Built, Rarely Recommended
Some of the strongest stores in the category barely appear at all.
Examples include:
Liquid I.V. (90)
Moon Juice (81)
DripDrop (75)
Excellent store readiness.
Minimal recommendation visibility.
Beauty Showed The Same Pattern
This is not the first category.
Our beauty research found:
r = 0.17
Again, statistically insignificant.
Combined:
Beauty: 4,000 recommendations
Supplements: 4,000 recommendations
8,000 recommendations total.
The same conclusion appears twice.
Store readiness does not predict recommendation frequency.
Brands, Not Retailers
Supplement recommendations are overwhelmingly brand-driven.
Retailers and marketplaces represented only:
195 of 4,000 recommendations
Less than 5%.
Unlike beauty, however, supplements show a stronger marketplace dependency.
Only:
33.5% of recommendations
mapped to direct-to-consumer stores we measure.
Many supplement brands rely heavily on:
Amazon
iHerb
GNC
rather than their own storefronts.
That fragmentation is itself a signal.
Recommendation by Memory™
The pattern points in one direction.
AI appears to recommend brands it already knows.
Not stores it can necessarily read.
We call this:
Recommendation by Memory™
The model reaches into parametric memory and returns familiar names it has seen repeatedly during training.
Examples:
Garden of Life
NOW Foods
Optimum Nutrition
Nature Made
These brands win despite having average, weak, or unknown store readiness.
Why This Matters
Most AI search discussions focus on visibility.
Questions like: Was I mentioned? Was I cited? Did I appear?
Those questions matter.
But recommendation is a different layer.
A brand can appear in the candidate set and still never make the shortlist.
Our data suggests:
Visibility is not recommendation.
And:
Recommendation is not readiness.
At least not today.
What Happens Next?
Today commerce appears heavily memory-driven.
Tomorrow may be different.
As AI systems evolve toward:
live retrieval
browsing agents
autonomous purchasing agents
the advantage shifts from brands AI remembers to stores AI can actually:
read
trust
understand
transact with
The brands winning today because of fame may be the most exposed when that transition happens.
Conclusion
Across 4,000 supplement recommendations:
Recommendation Frequency™ did not correlate with AI Commerce Score™
Famous brands dominated recommendations
Poorly optimized stores frequently outperformed better-built stores
Recommendation behavior appears driven more by memory than readiness
The question is no longer: Can AI see your brand?
The harder question is: Why did AI choose your brand instead of someone else's?
That is the question Recommendation Intelligence Research™ is designed to answer.
Get a free AI Commerce Score™: https://atomfoundry.dev
Research Hub: https://atomfoundry.dev/research
Top comments (0)