Recommendation Intelligence Research™ · The Fame Study

AI recommends by fame, not by store quality

We took the 200 most-recommended brands across five ecommerce categories and asked what actually separates the brands AI names from the brands it ignores. The answer is not how good their store is. It is how famous they are. And even fame explains only a quarter of it.

Atom Foundry · June 2026 · Part two of the Recommendation Intelligence Research™ series

200

Brands analyzed

24.9%

Explained by fame

2.1%

Explained by store quality

~12×

Fame over quality

Where this comes from

First we proved store quality does not drive recommendation

In the first part of this series we captured 20,000 AI product recommendations across beauty, supplements, coffee, pets, and home and living, and matched every one to the measured AI-readiness of the real store behind the brand. Across all five categories the relationship between how AI-ready a store is and how often AI recommends it was statistically indistinguishable from zero.

That answered the first question and raised a harder one. If a clean, well-structured, machine-readable store is not what earns the recommendation, then what is? This study is the attempt to answer it with the same discipline: real numbers, no invented deltas, and an honest account of what we could and could not measure.

The headline finding

Fame outweighs store quality roughly twelve to one

We took the 200 most-recommended brands and, for each one, measured public fame signals that do not depend on the store at all: how many people read its Wikipedia article, in how many languages that article exists, how long the article is, and how short and nameable the brand is. We then ran a multiple regression against how often each brand is actually recommended, and compared it directly against the store's own AI Commerce Score on the same brands.

What explains how often a brand is recommended?

Share of recommendation frequency explained (R²), measured on the same on-index brands

Read it honestly: fame is by far the stronger factor, but notice how much track is empty. Even fame leaves about three quarters of the outcome unexplained. Store quality barely moves the needle at all.

The same finding, in plain brands

The most-recommended and least-recommended brands have nearly identical stores

Regression can feel abstract, so here is the same truth in plain terms. We sorted the 200 brands by how often they are recommended and compared the top 50 against the bottom 50. The top group is recommended about six times more often. You would expect their stores to be far better. They are not.

Top 50 vs bottom 50 most-recommended brands

Same store quality. Very different fame.

The winners are not better built. They are better known. Median Wikipedia readership tells the same story even more sharply: about 1,660 monthly readers for the top group, and zero for the median brand in the bottom group.

The part most people get wrong

These recommendations are not random. They are stable

It would be comforting to think the unexplained majority is just noise, that AI picks brands more or less at random and there is nothing to understand. We tested that directly. Every shopping question was run twenty times, and we measured how often the answer changed.

It almost never does. The same question puts the same brand in the top spot between 78 and 91 percent of the time, and across twenty runs only two or three brands ever reach first place. The brand-level variation across runs is effectively zero.

How consistent is the top recommendation?

Share of runs that return the same brand in first place, by category

This changes the whole picture. The unexplained majority is not chaos. It is a fixed hierarchy. AI reliably selects the same brands; what it selects on is simply not visible from the store or from public fame.

The conclusion

Three measurements, one answer

2.1%

Store quality barely matters. A cleaner, more machine-readable store does almost nothing for how often AI recommends you.

24.9%

Fame matters more, but not most. Public prominence explains about twelve times more than store quality, yet still leaves three quarters unexplained.

~85%

And it is stable. The same brands win again and again. The hierarchy is consistent, just invisible to ordinary metrics.

What this means for a brand: if you are not already inside the winning set, you are not invisible by accident or by bad luck on a given day. You are invisible consistently, for reasons that neither your store quality nor any public fame metric will reveal. The only way to know where you actually stand is to measure the recommendations themselves.

Method & honesty

How we measured it, and what we could not

Brands analyzed: 200 most-recommended across 5 categories

Recommendation data: 20 runs per intent, gpt-4o-mini

Fame signals: Wikipedia views, language editions, article length, name length

Entity matching: Wikidata, commercial entities only, humans excluded

Models used: multiple regression, Spearman correlation

Stability: top-1 consistency across 20 runs per intent

Limitations, stated plainly. Wikipedia is a rough proxy for fame, not a perfect one. It captures encyclopedic prominence, while the fame that actually drives recommendation is more likely commercial presence across reviews, listicles, and forums. Two signals we believe matter, advertising spend and reliable web or Reddit mention volume, are not freely or credibly obtainable, so they are excluded by design rather than estimated. The store-quality comparison rests on the brands for which we hold a measured store (30 in the top group, 18 in the bottom). The individual weights inside the fame regression should not be read on their own, because the fame signals overlap; the valid figure is the combined R². Every number here comes from a real query, never an assumed delta.

Find out where your brand stands

The market measures whether AI can see you. We measure why AI chooses you, or why it chooses someone else instead. Get a Recommendation Intelligence read on your category.

Get your Free Score