We Asked AI 12,000 Shopping Questions. It Kept Recommending Famous Brands.

Over the last few months, I've been running a research project at Atom Foundry.

The original goal was simple.

I wanted to understand whether AI systems recommend brands because they have better websites, better product information, and more machine-readable stores.

In other words: Do AI systems recommend the businesses that are easiest for AI to understand? The answer surprised me.

The Experiment We analyzed three e-commerce categories:
Beauty
Supplements
Coffee

For each category, we created high-intent shopping prompts.

Questions like:
What are the best supplements for immune health?
Which coffee brands are worth buying?
What skincare brands do dermatologists recommend?

Each prompt was run repeatedly through AI systems.

In total, we analyzed more than:
12,000 recommendations
900+ brands
dozens of buying intents

At the same time, we scored the stores behind those brands using our AI Commerce Score™, which measures how understandable a store is to AI systems.

The expectation seemed obvious.

Stores that are easier for AI to understand should be recommended more often.

What We Expected

We expected something like this: Higher AI Commerce Score → Higher recommendation frequency.

The logic felt reasonable.

If AI can better understand your products, your structure, your content, and your trust signals, then AI should be more likely to recommend you.

Right?

Apparently not.

What We Found

Across all three categories, the relationship was almost nonexistent.

Beauty: Correlation: 0.17

Supplements: Correlation: -0.015

Coffee: Correlation: 0.019

In practical terms: Almost no relationship.

Some of the most recommended brands had mediocre scores.

Some of the highest-scoring stores were barely recommended at all.

The same pattern appeared again and again.

The Brands AI Loved

In coffee, brands like:
Peet's Coffee
Blue Bottle
Stumptown

appeared constantly.

Even when their stores were far from the strongest performers.

In supplements, brands like:

NOW Foods

showed up repeatedly despite scoring deep inside what we classify as the AI Invisible range.

In beauty, large legacy brands continued dominating recommendations while many technically stronger stores received little attention.

The pattern was difficult to ignore.

So What Is AI Optimizing For?

Our working hypothesis is simple.

Today's AI systems often recommend from memory.

Not from evaluation.

The model has seen certain brands thousands of times during training.

Those brands appear in:
articles
reviews
forums
news
social media
discussions

The model recognizes them. So it recommends them. Not because it analyzed the store. Not because it evaluated the buying experience. Not because it measured quality. Because it already knows the name.

AI Visibility Is Not The Same As AI Recommendation

This was probably the biggest lesson from the research.

Many businesses are currently focused on AI visibility.

Questions like:
Can AI find us?
Can AI read our content?
Can AI mention our brand?
Can AI cite us?

Those are important questions.

But they're not the same as: Can AI recommend us?

A brand can be:
visible
cited
understood

and still fail to become the recommendation.

The gap between visibility and recommendation may become one of the most important areas of AI commerce over the next few years.

Why This Matters

If recommendation behavior is driven primarily by familiarity, then many businesses are optimizing the wrong thing.

Improving structured data is useful.
Improving content is useful.
Improving AI readability is useful.

But none of those automatically lead to recommendations.

Recommendation appears to operate on its own layer.

A layer influenced by:
familiarity
trust
reputation
authority
historical presence

And potentially many other signals we still don't fully understand.

The Question We're Exploring Next

Most conversations around AI commerce focus on visibility.

We're becoming increasingly interested in something else: Why does AI choose one brand over another?

Because recommendation is where influence begins. And influence is what ultimately drives commercial outcomes.

The next generation of commerce may not be won by the brands AI can see.

It may be won by the brands AI consistently chooses.