Daniel Pokorný

Posted on Jun 18

The State of AI Recommendations in Home & Living

#ai #ecommerce #machinelearning #marketing

We Captured 4,000 AI Recommendations in Home & Living. The Results Were Unexpected.

AI is supposed to reward better websites.

At least that's what many brands assume.

If a store is easier for AI systems to understand, trust, and interpret, it should receive more recommendations.

Our data suggests otherwise.

We analyzed 4,000 AI-generated recommendations in Home & Living and compared recommendation behavior against AI Commerce Score™.

The correlation came back at:

r = 0.108

Effectively zero.

This is the fifth study in our Recommendation Intelligence Research™ series.

And after more than 20,000 captured recommendations across five industries, the pattern remains unchanged.

AI does not appear to recommend the most AI-ready brands.

It recommends the brands it already knows.

The Research

This study follows the exact same methodology used in our previous research across Beauty, Supplements, Coffee, and Pets.

Everything below is based on captured responses.

Nothing is estimated.

Nothing is projected.

Category: Home & Living

Model: GPT-4o-mini

Shopping Intents: 20

Runs Per Intent: 20

Prompt Runs: 400

Recommendations Captured: 4,000

Distinct Brands: 271

We measured three core metrics.

Recommendation Share™

The percentage of all captured recommendations belonging to a brand.

Recommendation Frequency™

The percentage of prompt runs in which a brand appeared at least once.

Recommendation Position™

The average position a brand occupied when recommended.

Lower numbers indicate stronger placement.

As with every study in this series, this represents one model, one category, and one point in time.

The Brands AI Recommended Most

The leaderboard immediately revealed the same pattern we observed in previous categories.

IKEA appeared in 48.3% of all prompt runs.

West Elm appeared in 47.0%.

Pottery Barn appeared in 41.5%.

These brands dominated recommendations despite only moderate store readiness.

IKEA scored 64 on AI Commerce Score™.

West Elm scored 60.

Meanwhile, some of the strongest stores in the category received very little attention.

Article scored 80.

Burrow scored 78.

Uplift Desk scored 77.

GhostBed scored 76.

Yet each appeared in fewer than 7% of prompt runs.

The strongest stores were not winning recommendations.

The most familiar brands were.

The Central Finding

The correlation between Recommendation Frequency™ and AI Commerce Score™ came in at:

r = 0.108

Across 71 measured brands, recommendation frequency and store readiness were effectively unrelated.

In practical terms, knowing how often AI recommends a home brand tells you almost nothing about how AI-ready its store actually is.

The result mirrors what we observed in Beauty, Supplements, Coffee, and Pets.

Once again, recommendation behavior appears disconnected from store quality.

Perhaps the most surprising finding is the average readiness of the brands AI recommends most often.

The average AI Commerce Score™ across all on-index recommended brands was just 50.7.

That sits directly on the border between Low Readiness and AI Invisible.

In other words:

AI is frequently recommending stores that are barely readable to the very systems doing the recommending.

Five Categories. One Result.

Home & Living was the final test.

At this point, the pattern is difficult to dismiss as noise.

Across more than 20,000 captured recommendations, recommendation frequency has never demonstrated a meaningful positive relationship with store readiness.

Five categories.

More than 20,000 recommendations.

Not one meaningful positive relationship.

Three categories produced essentially zero correlation.

One produced a negative relationship.

Home & Living produced essentially zero again.

The conclusion is no longer a single finding.

It is a replicated result.

AI recommendation frequency is not positively related to store readiness.

The Most Marketplace Driven Category Yet

Home & Living also produced another notable finding.

Approximately 21% of all recommendations went to retailers and marketplaces.

That is the highest retailer share we have measured across the entire research program.

Amazon. Wayfair. Home Depot. Macy's. Lowe's.

Retail remains deeply embedded in how AI understands this category.

At the same time, Home & Living produced the highest on-index coverage of any category in our research.

Roughly 60% of recommendations mapped directly to a single-brand store already included in our database.

This gave us more visibility into the relationship between recommendation behavior and store quality than any previous study.

The result remained unchanged.

Recommendation frequency still failed to track with readiness.

What It Means

Visibility is not recommendation.

Recommendation is not readiness.

And readiness is not currently being rewarded.

Today's AI systems still recommend heavily from memory.

They reach for the brands they encountered repeatedly during training.

Brands with decades of awareness.

Brands with large retail footprints.

Brands that appear constantly across reviews, articles, forums, and public discussion.

That helps explain why IKEA, West Elm, and Pottery Barn dominate recommendations despite only modest levels of AI readiness.

But memory is unlikely to remain the dominant signal forever.

The future of AI commerce will not be powered entirely by what models remember.

It will increasingly depend on what agents can verify.

As shopping agents gain the ability to browse websites, compare products, evaluate trust signals, interpret structured information, and complete purchases autonomously, recommendation behavior may gradually shift from memory-based recommendation toward evidence-based recommendation.

When that happens, readable stores become more important than famous names.

The brands benefiting most from familiarity today may be the brands with the most to lose tomorrow.

The brands investing in AI Readability™, AI Understanding™, and AI Trust™ today may become the brands best positioned for the next generation of AI commerce.

That transition is exactly what the Recommendation Intelligence Framework™ was designed to measure.

Recommendation Intelligence™

After five studies and more than 20,000 captured recommendations, we believe the industry may be looking at the wrong layer.

Most AI search conversations focus on visibility.

Can AI find you? Can AI crawl you? Can AI cite you?

Those questions matter.

But they do not explain recommendation behavior.

Recommendation behavior appears to be a separate system.

A separate problem. A separate competitive advantage.

That layer is what we call: # Recommendation Intelligence™

Not whether AI knows a brand.

But whether AI chooses it.

What's Next

Home & Living closes the first phase of the Recommendation Intelligence Research™ program.

Across five industries and more than 20,000 captured recommendations, we found no meaningful relationship between store readiness and recommendation frequency.

The next phase asks a different question.

If store quality does not explain recommendations, what does?

The upcoming State of AI Recommendations Across Commerce 2026 report will combine findings from all five industries into a single cross-category analysis.

After that comes a new hypothesis.

The State of Fame in AI Recommendations

Perhaps recommendation behavior is not driven primarily by store quality.

Perhaps it is driven by fame.

If recommendation frequency consistently ignores store readiness, the next logical question becomes whether AI systems are simply recommending the brands they remember best.

After 20,000 recommendations, that possibility is becoming increasingly difficult to ignore.