AI Knows Your Website 76% of the Time. It Still Won't Recommend You.

We asked one model to name the official website of 360 real brands.

It got 75.9% right.

Then it recommended someone else anyway.

That result matters because it rules out one of the simplest explanations for why store quality does not predict AI recommendation.

Maybe the model simply does not know where the brand is online.

The data suggests that is not the full explanation.

AI recognizes the official websites of 75.9% of 360 tested brands but still recommends other brands.

What we tested

We took 360 real brands and asked the model one simple question:

What is the official website for [brand]?

There was:

• No shopping context
• No comparison
• No product query
• No recommendation request

Just the website.

We ran the question 10 times per brand, producing:

• 360 brands
• 3,600 total lookups
• 75.9% correct domains
• 24.1% wrong or unknown

The scoring was simple:

Correct / Wrong domain / Unknown

Why this matters

Our earlier research has already ruled out several explanations for AI recommendation.

Store quality explains just 0.7% of the variance.

Public fame explains just 1.2%.

This study tests another simple explanation:

Maybe AI does not recommend a brand because it does not know where the brand is online.

If that were true, we would expect a large information gap.

But the model correctly identified the official website for roughly three out of four brands.

So the model often knows where these brands live online.

And it still recommends someone else.

Knowing the website is not the same as using it

This creates an important distinction.

A model can know:

Brand → official website

without necessarily doing:

Brand → retrieve website → evaluate website → recommend brand

Those are different stages.

Knowing a domain is a memory fact.

Recommendation is a decision.

The model can have the first without performing the second.

That means the problem is not simply awareness.

It is what the model does with information it already has.

Recommendation stability

We also segmented the brands by recommendation stability from our earlier Fame Study.

The difference was large.

Low stability brands:

16.2% unknown rate

High stability brands:

2.5% unknown rate

That is roughly a 6× difference.

AI model unknown rate is 16.2% for low stability brands and 2.5% for high stability brands.

The result does not prove that website knowledge causes recommendation stability.

But it adds another signal to the broader pattern.

Brands that appear more consistently in recommendation behavior are also much less likely to have the model admit that it does not know their website.

The model can locate the brand. That still does not make it a choice.

Across all 3,600 lookups:

75.9% returned the correct domain.

24.1% were wrong or unknown.

AI correctly identified the official domain for 75.9% of tested brands.

That means the model is often capable of locating a brand online.

But locating a brand is not the same as selecting it.

This distinction becomes even more interesting when combined with our earlier browsing study.

We found that switching retrieval on changed 77% of recommendations.

So when the model actually looks, what it finds can matter enormously.

But this study suggests that the address itself can already be known without necessarily becoming part of the recommendation process.

The gap is not simply information

This is the key finding.

If the model did not know where a brand was online, we could explain a missing recommendation as an information gap.

But that explanation does not survive this result.

The model can identify the official website for most brands.

Yet it can still recommend another brand.

So the better question is not:

Does AI know this brand?

It is:

What does AI do with information it already has?

That is a different problem.

From knowing to recommending

We have been mapping AI commerce as a sequence of different stages:

1. Memory
2. Retrieval
3. Understanding
4. Candidacy
5. Evaluation
6. Recommendation

This study sits right at the boundary between those stages.

A brand can be known.

The model can know its website.

The brand can be retrievable.

And it can still lose the recommendation.

That is why AI visibility cannot be reduced to a single metric.

Being known is not being recommended.

Being retrievable is not being recommended.

And knowing where a brand lives online is not the same as deciding that the brand is worth choosing.

Methodology

Study methodology showing 360 brands, 10 runs per brand and 3,600 total website lookups.

Prompt

What is the official website for [brand]?

Brands tested: 360

Runs per brand: 10

Total lookups: 3,600

Scoring: Correct / Wrong domain / Unknown

Segmentation: Recommendation Stability from the Fame Study

A note on the dataset

Every other dataset in this research series lives in our production database with a full per-query audit trail.

This one does not.

It was run and scored directly, and only the aggregate result was carried into our working notes.

We are publishing the finding because it is consistent with everything measured around it, but we cannot provide the raw 3,600 rows in the same way we can for the rest of the series.

If that affects how you weight this result, that is a fair read.

We think the limitation is worth stating clearly.

The takeaway

The model knows where many brands are.

That does not mean it will recommend them.

The gap is not simply whether AI knows you exist.

The more interesting question is what happens between knowing and choosing.

Read the full research at Atom Foundry