DEV Community: Daniel Pokorný

Candidacy vs Selection: What Happens When You Stop Studying Only the Winners?

Daniel Pokorný — Sat, 12 Sep 2026 20:12:01 +0000

We thought we were measuring AI recommendations.

We weren't. At least, not completely.

For most of our previous research, we analyzed brands that AI had already recommended.

That created a statistical problem.

We were studying the population that had already passed the gate.

So we changed the question: What separates the stores AI ever recommends from the stores it never recommends?

We tested the full population, and the result changed how we think about AI recommendation systems.

We analyzed 60,924 e-commerce stores.

599 were recommended at least once
60,325 were never recommended
Recommended stores averaged 55.4 on our AI Commerce Score
Never-recommended stores averaged 56.0
Intent was the one score component that clearly separated the two groups
Intent was roughly 39% higher among recommended stores
But once a store was already in the recommendation set, Intent explained only 1.2% of recommendation-frequency variation

The key finding:

A factor can help a store become a candidate without helping it win once it becomes a candidate.

We call this distinction Candidacy vs Selection.

The statistical problem: range restriction

Our previous recommendation research focused on brands that the model already recommended.

That lets you ask: Among brands already being recommended, does store quality correlate with recommendation frequency?

But it does not answer: What separates stores that are ever recommended from stores that are never recommended?

Those are different populations.

If a variable only matters at the recommendation gate, restricting the sample to already-recommended brands can hide the effect.

So for this study, we removed that restriction.

The dataset

We tested:

60,924 total stores
599 recommended at least once
60,325 never recommended

We compared the total AI Commerce Score and all seven score components across the full population.

We stopped looking only at the winners and tested the full population.

Result #1: Recommended stores were not better overall

If AI recommendation were simply a reflection of store quality, we would expect the recommended population to have a higher average score.

It didn't.

Group	Average AI Commerce Score
Recommended	55.4
Never recommended	56.0

The recommended group actually scored slightly lower.

So overall store quality did not separate the two populations.

That was the first signal that we were dealing with something more complicated than: Better store → more AI recommendations

Result #2: One score component stood out

We then compared the seven score components.

Intent was the one score component that consistently separated stores that were ever recommended from stores that were never recommended.

The values were:

Component	Recommended	Never recommended
Intent	5.13	3.70
Visual	7.43	7.22
Schema	3.83	3.53
Technical	11.27	12.13
Trust	10.42	11.06
Price	8.25	8.94
Brand	3.07	3.14

Intent showed roughly a 39% gap.

That made it the obvious candidate for further testing.

Does the Intent signal hold up?

A single correlation can be misleading.

So we checked several possible explanations.

1. Category mix

The Intent gap appeared across 9 of 9 niches.

The gap ranged from +0.78 to +1.59 points.

That makes a simple category-composition explanation less likely.

2. Circularity

We checked Intent against the other score components.

Its correlations with the other factors ranged from approximately:

0.20 to 0.29

So Intent was not simply duplicating another score component.

3. Brand ownership

The gap was:

1.40 vs 1.53

for the relevant brand-owned comparison.

The difference was nearly identical.

4. Selection after candidacy

This was the critical test.

Once we restricted the sample to the 599 stores that had already been recommended, Intent had almost no relationship with recommendation frequency.

r = 0.112

R² = 1.2%

That's the key result.

Candidacy ≠ Selection

The same factor can matter at one stage and become almost irrelevant at another.

We can represent the process like this:

60,924 stores
|
v
CANDIDACY
|
v
599 ever recommended
|
v
SELECTION
|
v
Who actually wins?

Getting into the recommendation set and winning inside that set are different problems.

Intent appears to help separate: stores that ever enter the set from
stores that never enter the set.

But once the store is already in the set, Intent predicts almost nothing about how often it wins.

Two anonymized brands showed the same split

We then reused controlled experiments from earlier studies in the series.

These weren't observational correlations.

They involved controlled manipulations of context.

Two anonymized brands showed almost opposite bottlenecks.

Brand J: a candidacy problem

Brand J had an overall recommendation rate of: 1.2%

across 400 real buyer-question observations.

But under supportive exposure conditions:

100% winner rate

Then we injected a single verified fact.

Its mention rate moved from: 1.25% → 97.5%

Nothing else about the brand changed.

This strongly suggests that Brand J's bottleneck was getting into the room.

Brand H: a selection problem

Brand H could also reach full candidacy under supportive conditions.

But in a multi-turn conversation, it survived to the final pick only:

in one tested condition.

Across Hidden Context tests, its winner rate ranged from: 0% → 53%

So getting Brand H into the recommendation set didn't solve the problem.

Its bottleneck appeared later.

Something about the specific matchup still determined the outcome.

Why this matters for AI optimization

This creates an important distinction.

A brand can have a problem with:

Discovery

AI cannot find the brand.

Understanding

AI cannot correctly interpret what the brand sells.

Candidacy

AI knows the brand exists but rarely puts it into the consideration set.

Selection

AI considers the brand but repeatedly chooses a competitor.

These problems can look identical from the outside:

"AI doesn't recommend me."

But they may require completely different interventions.

Correlation still cannot tell us causality

There is an important limitation here.

We can show that Intent separates recommended and never-recommended stores.

We cannot use this observational result alone to say:

Increasing Intent will cause AI to recommend a store.

There are alternative explanations.

For example, sharply positioned brands might be easier for AI to identify.

Or successful brands might simply have invested more heavily in positioning.

Both could produce the same observed correlation.

That's why controlled experimentation matters.

If we want to understand causality, we need to:

Change one variable
Hold other variables constant
Run the same test again
Measure the change in AI behavior

That's a very different research problem from collecting correlations across thousands of stores.

A platform confound worth mentioning

One result initially looked dramatic.

Magento stores appeared in the recommended group at: 3.70%

Shopify stores: 0.79%

That's approximately a 4.7x difference.

But Magento stores in this dataset skew older and larger.

So we don't interpret Magento itself as the cause.

The platform is better treated as a marker for other underlying characteristics.

This is another reason to be careful when turning observational correlations into optimization advice.

What we are not reporting

An earlier pass also compared raw review counts and average price between the two groups.

We removed both from the final study.

A later data-quality check found:

Review count was populated for only 9.4% of the 66,085 scanned stores
Average price was populated for 66.7%

That meant those comparisons were largely comparing missing fields rather than real values.

The score components used in this study were computed for every scanned store, so those are the comparisons we report.

The bigger idea

This study changed one assumption for us.

We used to think about AI recommendation as one problem.

Now we think it may be at least two.

Candidacy asks:

Can the brand get into the recommendation set?

Selection asks:

Once it's there, why does it beat the alternatives?

And the data suggests that the signals involved may be different.

That's important because most AI visibility measurement still collapses these stages together.

But:

Being visible is not the same as being considered.

And:

Being considered is not the same as being chosen.

What we're testing next

This study gives us a stronger reason to move toward controlled experiments.

We don't just want to know which signals correlate with recommendations.

We want to know which changes actually move the decision.

Change a rating.
Change a specification.
Change a price.

Add or remove a verified fact.

Change the available context.

Then observe what happens.

That's where we think the next generation of AI Commerce Intelligence will come from.

Not more signals for the sake of more signals.

But better evidence about which signals actually matter.

The takeaway

The most important finding isn't: "Intent is the ranking factor."

We don't have evidence for that.

The evidence supports something narrower: Intent separates stores that ever enter the recommendation set from stores that never do.

And:

Once a store is already in the recommendation set, Intent explains almost nothing about which store wins more often.

That's the distinction.

Candidacy gets you into the game.

Selection wins the game.

And we're trying to understand what happens between those two stages.

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

Daniel Pokorný — Sat, 05 Sep 2026 15:09:38 +0000

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:

Memory
Retrieval
Understanding
Candidacy
Evaluation
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

"29,633 Reasons. 26,812 Unique. The Model Confabulates."

Daniel Pokorný — Tue, 01 Sep 2026 19:58:46 +0000

29,633 Reasons. 26,812 Unique. The Model Confabulates.

We checked every explanation an AI model gave for its recommendations. Almost none of them repeat.

I have spent a lot of time trying to understand why AI recommends one brand instead of another.

We tested store quality.
We tested public fame.
We tested domain recognition.
We tested other signals.

Many of them explained very little.

So we tried something much simpler.

We looked at what the model itself said.

Every recommendation came with a reason.

So we collected them all.

29,633 reasons.

Then we checked how often the model used the same reason again.

The result was striking.

26,812 reasons were completely unique.

That is 90% of the dataset.

The most repeated reason appeared only 12 times.

Out of 29,633 recommendations.

That is just 0.04%.

We read every reason

Every recommendation in our dataset had a short reason attached to it.

We took the full reason field and checked every entry for exact text repetition.

If the model was using a small set of stable reasons, we should have seen those reasons many times.

We did not.

The test was simple.

We checked:

29,633 reasons

26,812 distinct reasons

90% unique text

12 times for the most repeated reason

We used exact text matching.

That means two sentences with the same meaning but different wording were still counted as different.

So this test only measures repeated text.

It does not measure repeated ideas.

The result

Nine out of ten reasons were unique.

The model did not keep using the same small set of explanations.

It kept writing new ones.

This is important because a stable decision rule should leave some kind of pattern.

The exact sentence does not need to be the same every time.

But similar decisions should produce similar explanations.

That pattern was not there.

The pick comes first

This is where the finding gets interesting.

The recommendation happens first.

Then the model produces a reason that fits the recommendation.

The reason can sound very good.

It can sound specific.

It can sound confident.

But that does not mean it describes the process that produced the recommendation.

It may simply be a good explanation written after the decision.

That is what we mean by confabulation in this study.

The model is very good at producing language that makes a decision sound reasonable.

That is different from knowing why the decision happened.

Asking AI why can be misleading

This creates a real problem for ecommerce teams.

Imagine an AI recommends another brand.

You ask why.

The model says the brand has better reviews.

You improve your reviews.

Another time it says the brand offers better value.

You change your pricing.

Another time it talks about quality.

You rewrite your product pages.

The advice sounds useful.

But this study shows why we should be careful.

The explanation may not be the signal that caused the recommendation.

It may simply be the explanation generated for that recommendation.

You could end up optimizing for the explanation instead of the cause.

This changes how we should study AI recommendations

If we want to understand what drives AI recommendations, asking the model is not enough.

We need to watch what it does.

Change one variable.

Keep everything else the same.

Run the recommendation again.

Measure what changed.

That gives us evidence about what actually moves the decision.

The explanation gives us language.

The experiment gives us evidence.

Those are very different things.

One important limitation

Our test used exact text matching.

So these two reasons would count as different:

better customer feedback

and

stronger reviews from buyers

They may describe the same idea.

Our study does not tell us how many underlying ideas were repeated.

It tells us that 90% of the actual text was unique.

That makes the finding narrower and cleaner.

We are not claiming to have reconstructed the model's internal process.

We are showing that its own explanation is not a stable enough signal to use as a direct explanation of its recommendation.

What we learned

There is a big difference between:

What the model recommends

and

What the model says about the recommendation

The first is behavior.

The second is generated language.

If we want to understand AI commerce, we need to measure the behavior.

That is what we are doing at Atom Foundry.

We are testing what happens when AI evaluates ecommerce brands.

Not just what it says.

But what actually changes its decisions.

29,633 reasons gave us a very simple lesson.

The explanation is not necessarily the mechanism.

It is the explanation.

Research: https://atomfoundry.dev/research/ai-confabulates-its-reasons

# We Scanned the Same Shopify Store Six Times. Only One Number Changed.

Daniel Pokorný — Sat, 25 Jul 2026 14:31:35 +0000

When you're building a measurement system, one of the first questions you need to answer is surprisingly simple.

How stable is the measurement itself?

Before optimizing our own Shopify store, Founder Lab, we wanted to establish a baseline.

Not just one scan.

Multiple scans.

Without changing a single byte on the store.

The result was more interesting than we expected.

The experiment

We scanned the exact same Shopify store six times.

1 baseline scan on July 12
5 rescans on July 17
all rescans completed within 62 seconds
absolutely nothing changed between scans

Here is what we measured.

Scan	Score	Intent
Jul 12	78	6
Jul 17 #1	76	4
Jul 17 #2	76	4
Jul 17 #3	78	6
Jul 17 #4	77	5
Jul 17 #5	77	5

Every deterministic factor stayed identical

This is where things became interesting.

Every deterministic component remained exactly the same.

Component	Score
Visual	11
Schema	15
Trust	15
Technical	7
Price	12
Recommendation	8
Brand	4

Every value above was bit-for-bit identical across all six scans.

Only one component moved.

Intent.

Intent changed between 4 and 6.

Everything else remained constant.

A simple observation

Subtract the intent score.

Every scan becomes identical.

78 - 6 = 72

76 - 4 = 72

78 - 6 = 72

77 - 5 = 72

The deterministic portion of the score never changed.

Only the model-evaluated component moved.

That immediately told us something important.

A one or two point change between two scans does not necessarily mean the store improved.

It may simply be normal model variance.

Why this matters

Many AI evaluation systems produce a single score.

People naturally compare:

and conclude something improved.

Our first experiment suggests that's not always true.

If part of the score is generated by an LLM rather than deterministic rules, small movements may simply be statistical noise.

That changes how we evaluate future experiments.

Instead of trusting a single rescan, every meaningful change now requires multiple rescans before we consider it real.

The baseline

Before making any optimization, Founder Lab looked like this.

AI Commerce Score: 78 / 100
Score range across six scans: 76–78
AI bot visits recorded: 0
Technical score: 7 / 15 (our weakest deterministic factor)

What happens next

This was intentionally the most boring experiment we could run.

We changed nothing.

That was the point.

Before optimizing a system, you first need to understand how noisy the measurement itself is.

Our next experiments will focus on deterministic improvements first.

Technical issues.

Structured data.

Store architecture.

Only after that will we begin changing product copy and intent signals.

Every change will be rescanned multiple times.

Not because we expect every optimization to work.

But because we now know the measurement itself has a noise floor.

Understanding that may be just as important as improving the score itself.

Founder Lab is the live Shopify laboratory behind Atom Foundry's AI Commerce research.

Every experiment is published publicly, including the ones that fail.

# A stranger tried to break my study. Here's what happened.

Daniel Pokorný — Tue, 14 Jul 2026 17:04:04 +0000

I was about to publish a clean finding. Then someone asked a better question, and the number that looked like it settled the argument turned out to be a trap.

Last week I published a controlled experiment showing that, on the same AI model, simply turning web search on changed 77% of product recommendations.

Same model. Same 50 buying prompts. One variable.

The follow-up article was already written.

I was going to break the results down by category, show that Pets changed 88% while Fitness changed 61%, and tell a neat story:

The more fragmented the market, the more browsing overrides what the model remembers.

Then a data engineer named Rami read the study and asked a question I couldn't answer.

The objection

His question was simple.

How do you know that's the category and not your prompts?

Generic prompts like:

Best dog food

should rely on retrieval much more heavily than specific prompts like:

Best cat litter for odor control.

If some categories happened to contain mostly generic prompts while others contained mostly specific ones, then what looked like a category effect might simply be a prompt-design effect.

He was right.

I couldn't rule it out.

So I paused the publication.

The trap

I tagged every one of the 50 prompts as either generic or specific and reran the analysis.

At first glance, it looked like I had my answer.

Generic prompts: 74.2% changed
Specific prompts: 80.0% changed

Specific prompts changed more.

Exactly the opposite of what Rami predicted.

I could have published that result, declared the objection resolved, and moved on.

It would have been the wrong conclusion.

Simpson's Paradox strikes

Once I looked inside each category, the pattern disappeared.

Some categories showed generic prompts changing more.

Others showed specific prompts changing more.

There was no consistent direction.

The pooled difference wasn't measuring causation.

It was measuring composition.

The categories that happened to contain most of my specific prompts (Coffee, Pets, Wellness) were also the categories that already had the largest browsing effect.

Specific prompts looked more volatile simply because of where they happened to appear.

This is a textbook example of Simpson's Paradox.

Ironically, the single number that looked like it answered the criticism was actually the most misleading statistic in the study.

The real test

To isolate the effect properly, prompt specificity has to be held constant.

So I compared categories within each prompt type.

Generic prompts only

Category	Recommendation change
Coffee	89.8%
...	...
Fitness	62.4%

Spread: 27 percentage points

Specific prompts only

Category	Recommendation change
Pets	94.4%
...	...
Fitness	52.6%

Spread: 42 percentage points

Fitness remains the lowest.

Coffee and Pets remain the highest.

The category effect survives after controlling for prompt style.

Looking for the confound

I also measured whether categories containing more generic prompts systematically produced larger recommendation changes.

The correlation was:

r = -0.56

If prompt composition were actually driving the result, this correlation should have been strongly positive.

Instead, it was moderately negative.

That doesn't prove the confound doesn't exist.

But it certainly doesn't support it either.

Where the study is still weak

I'd rather point out the weaknesses myself than hope nobody notices.

Some categories have very few prompts.

For example:

Coffee contains only one generic prompt
Fitness contains only one specific prompt
Electronics contains no specific prompts

That means some category-level estimates rely on very small samples.

Likewise, the correlation of -0.56 across only ten categories isn't statistically significant.

At that sample size, significance would require roughly |r| > 0.63.

So this is evidence of a direction.

Not proof.

There's another limitation.

The prompt mix wasn't intentionally balanced.

It emerged naturally during the study.

A future experiment should deliberately balance generic and specific prompts across every category.

That's already on my roadmap.

So the honest conclusion isn't:

"There is no confound."

It's simply:

I looked for the confound, found evidence pointing the other way, and the category effect still survives after stratification.

The second hole

Rami raised another criticism.

The original analysis had no statistical treatment.

Fair enough.

Fortunately, that's easy to improve.

A naïve confidence interval assumes every brand-intent pair is independent.

They aren't.

Brands cluster within buying intents.

So instead I ran a cluster bootstrap, resampling the 50 buying intents rather than individual brand observations.

The result stayed remarkably stable.

Estimated recommendation change: 77%
95% confidence interval: 74%–80%

The naïve interval had been 75%–79%.

Accounting for clustering widened it by roughly 1.6×.

The wider interval is the honest one.

That's the version I'm publishing.

Why I'm writing about this

The interesting part isn't that the finding survived.

The interesting part is that the statistic which looked like it settled the debate was the one most likely to fool me.

Had I published:

"Specific prompts change more. Objection handled."

it would have sounded confident.

It would have been easy to quote.

And it would have been wrong.

Nobody would have questioned it.

It probably would have lived online for years.

Instead, a stranger asked a better question than I had asked myself.

That turned out to be far more valuable than simply having my original conclusion confirmed.

Full study

Full study and methodology:

https://atomfoundry.dev/research/web-search-changes-ai-recommendations

# I Flipped One Toggle and 77% of the AI's Product Recommendations Changed

Daniel Pokorný — Sun, 12 Jul 2026 20:07:38 +0000

How much does web search change what an AI recommends?

I measured it using a fixed set of e-commerce buying questions.

The result surprised me.

Correction, 15 July 2026. An earlier version of this piece claimed that changing the model moved only about 6% of recommendations, and concluded that which model you ask barely matters.

That number was derived rather than measured. I subtracted two overlap figures from two different comparisons and treated the difference as the model effect. It is not the same quantity, and it was wrong.

Measured directly, gpt-4o-mini without search against gpt-4o without search, 66.9% of the recommended brands changed. Not 6%.
One caveat on that figure: the two runs were collected twelve days apart, so it mixes the model change with whatever drifted in between, and I cannot separate them from this data.

The search comparison does not have that problem, both conditions ran together, so the 77% figure is unaffected and stands.

The corrected picture: swapping the model rewrites roughly two thirds of the recommendations. Turning on browsing rewrites roughly three quarters. There is no such thing as "what AI recommends." There is only what a particular model, in a particular configuration, at a particular moment, recommends...

I'm leaving the mistake visible rather than editing it out, because the way it happened is the useful part: a derived number looked like a measured one, and it agreed with the story I wanted to tell.

Why I Ran This Experiment

A few weeks ago I published research showing something unexpected.

Across thousands of AI recommendations, store quality explained almost nothing about why brands were recommended.

Brand popularity explained much more.

Yet almost 73% of recommendation behavior remained unexplained.

After publishing the study, data engineer Rami suggested an interesting hypothesis.

Maybe part of that "unexplained" behavior isn't mysterious at all.

Maybe it's simply the retrieval layer.

His suggestion was simple:

Run the exact same prompts twice.

Once with web search enabled.
Once with web search disabled.

Then compare the recommendations.

So I did.

Experiment Setup

Model:
GPT-4o

Conditions:
• Browsing ON
• Browsing OFF

Dataset:
50 buying prompts

Categories:
Pets
Beauty
Supplements
Coffee
Fashion
Fitness
Food
Home
Wellness
Electronics

Runs:
10 per prompt per condition

Metric:
Unique recommended brands

To isolate the effect of browsing from the effect of model size, I also repeated the experiment using GPT-4o-mini.

The Result

The overlap between GPT-4o with browsing enabled and GPT-4o without browsing was only:

23%

That means:

77% of the recommended brands changed simply by enabling web search.

Same model.

Same prompts.

Same methodology.

One toggle.

What Changed?

With browsing disabled, GPT-4o recommends mostly from its internal memory.

Those recommendations tend to favor well-known brands the model likely encountered frequently during training.

With browsing enabled, the model relies heavily on live retrieval.

Many memory-based recommendations disappear and are replaced by brands discovered during search.

These weren't small ranking adjustments.

They were largely different recommendation lists.

Was It Really Search?

A reasonable objection would be:

Maybe GPT-4o simply recommends different brands than GPT-4o-mini.

So I isolated that variable.

Model difference (Browsing OFF)

Moving from GPT-4o-mini to GPT-4o changed only about:

6% of recommendations

Search difference (Same model)

Turning browsing ON changed:

77% of recommendations

The retrieval layer had a dramatically larger impact than model size.

It Depends on the Category

Browsing didn't affect every category equally.

Category	Recommendations Changed
Pets	88%
Fitness	61%

The pattern appears consistent.

Markets dominated by a handful of famous brands showed smaller changes.

Markets with many niche brands showed much larger changes.

Pets has consistently produced the strongest effects across every study I've run.

I still don't fully understand why.

The Bigger Black Box

Rami's original hypothesis had another part.

He suggested that much of the unexplained variance may live inside the model's training history.

Questions like:

Which sources mentioned a brand?
How often?
In what context?
With what sentiment?

Those signals aren't observable today.

I can measure what gets recommended.

I cannot yet measure why a model internally trusts one brand more than another.

That remains one of the biggest black boxes in AI recommendation systems.

Why This Matters

Many companies treat "AI visibility" as a single objective.

I don't think it is.

There are at least two different systems at work.

Memory

Long-term exposure during training.

This rewards brands that are already widely known.

Retrieval

Live information gathered at query time.

This rewards brands that are discoverable and well represented across the web.

Those are different mechanisms.

And they often produce different recommendations.

Key Takeaway

If your optimization strategy only targets one of these layers, you're missing the other.

Depending on whether an AI assistant uses browsing, retrieval may completely replace memory-driven recommendations.

In this experiment, enabling browsing changed 77% of the recommended brands.

That's too large to ignore.

Methodology

50 fixed ecommerce buying prompts
10 executions per prompt
GPT-4o with browsing ON
GPT-4o with browsing OFF
GPT-4o-mini comparison
Recommendation overlap measured using distinct recommended brands per buying intent

Browsing implementations evolve over time, so treat the percentages as a snapshot rather than a universal constant.

The important finding isn't the exact number.

It's the magnitude of the effect.

I'm continuing this research through Atom Foundry, where I study how AI systems understand, evaluate, and recommend ecommerce brands.

If you spot weaknesses in the methodology or have ideas for improving the experiments, I'd genuinely like to hear them.

# AI Recommends by Fame. But Fame Doesn't Explain Most Recommendations.

Daniel Pokorný — Mon, 22 Jun 2026 12:20:00 +0000

An analysis of 20,000 AI-generated product recommendations across e-commerce.

For the last year, most conversations around AI optimization have focused on visibility.

Can AI crawl your website? Can AI retrieve your content? Can AI cite your brand?

Those are useful questions.

But while analyzing AI recommendation behavior across ecommerce, we kept running into a different question:

Why does AI consistently recommend certain brands over others?

To investigate, we analyzed 20,000 AI-generated product recommendations across five e-commerce categories:

Beauty
Supplements
Coffee
Pets
Home & Living

In total, the dataset included:

20,000 recommendations
1,490 brands
5 commerce categories

What we found surprised us.

Hypothesis #1: Better Stores Get More Recommendations

The first assumption seemed obvious.

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

To test this, we compared recommendation frequency against our AI Commerce Score™, a framework designed to evaluate:

machine readability
semantic structure
content depth
technical implementation
AI readiness

The expectation was simple: Higher quality stores should receive more recommendations.

The data said otherwise.

Store quality explained only:

2.1% of recommendation frequency.

Not a weak relationship.

Almost no relationship at all.

That immediately raised another question.

If AI isn't recommending brands because they have better stores...

What is driving recommendations?

Hypothesis #2: Fame

The next explanation was public fame.

Popular brands have:

more reviews
more backlinks
more media coverage
more mentions
more content

Perhaps AI simply recommends brands that humans already recognize.

To test this, we analyzed the 200 most-recommended brands from our dataset.

For each brand, we collected public fame signals including:

Wikipedia readership
Number of language editions
Article depth
Brand name characteristics

We then compared those metrics against recommendation frequency.

Results

Store Quality: 2.1%

Public Fame: 24.9%

Most recommendation behavior: Still unexplained

Public fame explains significantly more recommendation behavior than store quality, but most recommendation behavior remains unexplained.

Fame mattered.

Much more than store quality.

But it still failed to explain most recommendation outcomes.

The Most Recommended Brands Aren't Better Stores

We then split the dataset into two groups:

Top 50 most-recommended brands
Bottom 50 least-recommended brands

If recommendation frequency reflected store quality, the difference should have been obvious.

It wasn't.

Metric	Top 50	Bottom 50
Recommendation Frequency	30.9%	5.0%
AI Commerce Score™	50.8	49.8

The recommendation gap was massive.

The quality gap was almost nonexistent.

The most recommended brands receive more than six times as many recommendations despite nearly identical store quality scores.

The winners weren't better stores.

They were simply recommended more often.

Maybe Recommendations Are Random?

At this point, a reasonable explanation would be: AI recommendation behavior is mostly random.

So we tested that.

Every shopping query was repeated twenty times.

If recommendation behavior were unstable, we would expect different winners across repeated runs.

Instead, we observed the opposite.

The same brands kept appearing.

Again.

And again.

Across categories, the top-ranked brand remained in first place between:

78% and 91% of runs.

Recommendation outcomes remain highly stable across repeated runs.

This was perhaps the most surprising result in the entire study.

Because it means the unexplained portion of recommendation behavior is not random.

The system appears remarkably stable.

We simply don't understand it yet.

What This Suggests

When we combine all findings, we observe four things:

Store quality explains very little recommendation behavior.
Public fame explains significantly more.
Most recommendation behavior remains unexplained.
Recommendation outcomes remain highly stable.

Taken together, these findings suggest there may be another layer operating beneath traditional AI visibility metrics.

Visibility answers: Can AI see a brand?

Recommendation answers: Will AI choose a brand?

Those are fundamentally different problems.

A Possible Recommendation Layer

One interpretation of these findings is that recommendation systems operate on a deeper decision layer than current AI visibility tools measure.

A layer beneath:

rankings
citations
retrieval
visibility

A layer that influences trust, selection and recommendation.

We don't yet know what variables define that layer.

But the data suggests it exists.

And understanding it may become increasingly important as AI systems play a larger role in commercial decision-making.

Open Questions

This study answered one question.

It created several more.

If store quality explains 2.1%...

And fame explains 24.9%...

What explains the remaining 73%?

Potential candidates might include:

trust signals
entity relationships
training data exposure
recommendation reinforcement effects
semantic authority
citation networks
brand familiarity patterns
factors we haven't identified yet

At this stage, we don't know.

But that's exactly what we're investigating next.

Methodology

Dataset:

20,000 AI-generated recommendations
1,490 brands
5 ecommerce categories

Categories:

Beauty
Supplements
Coffee
Pets
Home & Living

Measured variables:

Recommendation frequency
AI Commerce Score™
Wikipedia readership
Language editions
Article depth
Brand name characteristics

Repeated-run testing:

20 repeated recommendation runs per query

Final Thought

Most AI optimization discussions today focus on visibility.

Our data suggests recommendation behavior may be a separate problem entirely.

Visibility determines whether AI can find you.

Recommendation determines whether AI chooses you.

And those may turn out to be two very different systems.

# The State of AI Recommendations Across Commerce 2026

Daniel Pokorný — Fri, 19 Jun 2026 12:05:00 +0000

Most businesses assume that if AI systems can understand their products, they will eventually recommend them.

Our latest research suggests the relationship may not be that simple.

Over the past several months, we analyzed 20,000 AI recommendations across five e-commerce categories:

Beauty
Supplements
Coffee
Pets
Home & Living

The goal was straightforward: To understand whether store readiness actually predicts recommendation behavior.

Or put differently: Do AI systems recommend the businesses that are best prepared for AI?

The answer surprised us.

The Assumption

Most discussions about AI visibility start with a reasonable belief.

If a business improves its structure, content, product information, trust signals, and machine readability, AI systems should be more likely to recommend it.

This assumption has fueled a growing industry around AI optimization, AI visibility, AI readiness, and AI commerce infrastructure.

But assumptions are not evidence.

We wanted to measure recommendation behavior directly.

What We Measured

For each category, we collected thousands of recommendations generated by AI systems in response to high-intent shopping questions.

We then compared recommendation frequency against each brand's AI Commerce Score™.

The expectation was simple: Higher scores should lead to more recommendations.

Instead, we found something very different.

Figure 1. AI Commerce Score™ and Recommendation Frequency™

Based on 20,000 AI recommendations across five ecommerce categories, recommendation frequency showed little to no meaningful relationship with AI Commerce Score™, suggesting that store readiness alone does not explain recommendation behavior.

The Surprising Result

Across all five categories, recommendation frequency showed little to no meaningful correlation with AI Commerce Score™.

Some highly recommended brands scored relatively poorly. Some highly optimized brands were rarely recommended.

The relationship was far weaker than expected.

This finding appeared repeatedly across Beauty, Supplements, Coffee, Pets, and Home & Living.

The implication is important.

Store readiness alone does not appear sufficient to explain why AI recommends certain brands more frequently than others.

Something else is happening.

The Familiarity Hypothesis

As we analyzed recommendation patterns, another explanation began to emerge.

Many of the most frequently recommended brands shared one characteristic:

They were already familiar. They had existing brand recognition. Existing awareness. Existing market presence. Existing references across the web.

In other words, recommendation behavior often appeared to resemble memory more than evaluation.

This led us to a framework we call: ## Recommendation by Memory™

Under this model, AI systems frequently recommend brands they have encountered repeatedly during training and exposure.

Not necessarily because those brands are objectively better.

But because they are more familiar.

A Potential Future Shift

However, we do not believe recommendation behavior will remain static.

As AI systems gain access to richer retrieval systems, real-time information, structured commerce data, and increasingly sophisticated evaluation capabilities, recommendation behavior may evolve.

We call this future state: ## Recommendation by Understanding™

Figure 2. The Evolution of AI Recommendation Systems

Recommendation behavior is evolving from what AI remembers to what AI understands. This transition may define the next phase of AI commerce and product discovery.

Under this model, recommendations become less dependent on historical familiarity and more dependent on:

Store quality
Trust signals
Product fit
Verifiable information
Real-time relevance
Semantic understanding

In other words: The best understood businesses may eventually outperform the most familiar businesses.

Why This Matters

If recommendation behavior is driven primarily by memory today, businesses face a difficult challenge.

Optimization alone may not immediately increase recommendation frequency.

But if recommendation systems gradually move toward understanding, the businesses investing in AI readiness today may be building an advantage for tomorrow.

The future of AI commerce may not belong to the brands AI remembers.

It may belong to the brands AI understands best.

The Emerging Question

Most businesses are still asking: Can AI find us?

A more important question may be emerging: Why does AI choose one business instead of another?

That question sits at the center of what we call Recommendation Intelligence™.

And as AI increasingly influences product discovery, recommendation behavior may become one of the most important areas of research in commerce.

About Atom Foundry

Atom Foundry is building the field of AI Commerce Intelligence™.

Our research explores how AI search engines, LLMs, Apple Intelligence, and shopping agents discover, understand, evaluate, trust, recommend, and route customers to businesses.

Current research initiatives include:

AI Commerce Intelligence™
Recommendation Intelligence™
AI Commerce Graph™
AI Readability™
AI Understanding™
AI Trust™
Decision Confidence™

Learn more:

🌐 https://atomfoundry.dev

📊 https://github.com/Atom-Foundry

Research by Atom Foundry. Based on 20,000 AI recommendations captured across Beauty, Supplements, Coffee, Pets, and Home & Living categories.

The State of AI Recommendations in Home & Living

Daniel Pokorný — Thu, 18 Jun 2026 16:01:40 +0000

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.

Explore The Research

Get your free AI Commerce Score™

https://atomfoundry.dev

Explore the full Home & Living dataset

https://atomfoundry.dev/research/state-of-ai-recommendations-home-living

Explore the complete Recommendation Intelligence Research™ series

https://atomfoundry.dev/research

Because the future of AI commerce will not be decided by visibility alone.

It will be decided by recommendation.

The State of AI Recommendations in Pets

Daniel Pokorný — Wed, 17 Jun 2026 09:34:45 +0000

Recommendation Intelligence Research™ · Atom Foundry · June 2026

We asked one AI model 20 high intent pet shopping questions, 20 times each.

Then we checked the same thing we checked in beauty, supplements, and coffee.

Does being recommended by AI have anything to do with how AI ready a store actually is?

Pets produced the strongest answer yet.

Across another 4,000 recommendations, recommendation frequency was not positively related to store readiness.

In fact, it moved in the opposite direction.

Methodology

Same method as the first three reports, so all four categories are directly comparable.

Everything below is computed from real captured responses.

Nothing is estimated or projected.

Category: Pets

Model: one model (gpt-4o-mini)

Intents: 20 high intent shopping prompts

Runs per intent: 20

Prompt runs total: 400

Recommendations captured: 4,000

Distinct brands: 405

Three metrics carry the analysis.

Recommendation Share™ is a brand's share of all recommendations captured. The field sums to 100 percent.

Recommendation Frequency™ is the percent of the 400 prompt runs in which a brand appeared at least once.

Recommendation Position™ is the average rank in the answer when the brand appeared. Lower is better.

Marketplaces are not brands.

Retailers and marketplaces such as Chewy, Petco, Amazon, PetSmart, and Frisco were separated out and excluded from the brand level analysis.

The contest measured here is between single brands and their own stores.

A note on pets.

Several of the largest pet food brands including Purina Pro Plan, Royal Canin, Hill's Science Diet, and Taste of the Wild are not in our store index, so they appear as off index and are excluded from the correlation analysis.

One brand, Wellness, was removed from the on index set after we found its name had auto matched to an unrelated domain. Rather than attach an incorrect store, we classified it as off index.

Other limitations remain the same.

One model.

One category.

One point in time.

Brands map only to stores we actually measure.

Anything else stays off index.

The Pet Recommendation Leaderboard

Top brands by share of voice across all 4,000 recommendations, with retailers excluded.

The AI Commerce Score™ column reflects each brand's actual score from our index where a measured store exists.

Pet category results showing Recommendation Share™, Recommendation Frequency™, Recommendation Position™, and AI Commerce Score™ for leading AI recommended brands.

The Central Finding

If AI recommended the stores that are easiest for AI systems to read, recommendation frequency and AI Commerce Score™ would move together.

They do not.

In pets, they move slightly in the opposite direction.

Across the on index single brand stores, the correlation between Recommendation Frequency™ and AI Commerce Score™ is r = -0.366 (n = 39).

That is the first category where the relationship is meaningfully negative.

The brands that appear most often in AI recommendations tend to have weaker stores.

Blue Buffalo appears in 27.8 percent of prompts while scoring only 30.

Merrick appears in 24.8 percent of prompts while scoring 42.

Purina Pro Plan appears in 26.8 percent of prompts despite not even being represented in our store index.

Meanwhile some of the strongest stores in the category barely appear at all.

**Benebone scores 79.

Casper scores 73.

Pawstruck scores 72.**

Yet none of them approach the recommendation frequency of the legacy pet food brands.

This is the clearest inversion we have measured so far.

The brands AI remembers are not the brands with the most AI ready stores.

The average AI Commerce Score™ across all on index recommended brands is just 52.7.

We have now measured four categories.

Beauty: r = 0.17
Supplements: r = -0.015
Coffee: r = 0.019
Pets: r = -0.366

Across more than 16,000 recommendations, recommendation frequency is never positively associated with store readiness.

In three categories the relationship is effectively zero.

In pets it becomes negative.

Brands, Not Retailers

Pets is also the most marketplace driven category we have measured.

Retailers and marketplaces such as Chewy, Petco, and Amazon accounted for 7.1 percent of all recommendations.

That is significantly higher than beauty, supplements, or coffee.

Only 19.8 percent of recommendations map to a single brand store that we actually measure.

That is the lowest share we have seen.

Two factors drive that result.

First, pet commerce is heavily concentrated around large retailers and marketplaces.

Second, some of the most recommended brands in the category are not represented in our store universe.

Purina
Royal Canin
Hill's Science Diet
Taste of the Wild

These brands are recommended because of brand familiarity, not because an AI system evaluated their stores.

That is itself an important finding.

What It Means

Visibility is not recommendation.

Recommendation is not readiness.

Today AI still recommends largely from memory.

It reaches for names that appeared most often in training data and public discussion.

That is why the dominant pet food brands continue to win recommendations even when their stores are weak, absent, or difficult for AI systems to evaluate.

But that advantage is temporary.

As AI shopping evolves toward retrieval, browsing agents, comparison engines, and autonomous purchasing, recommendation decisions will increasingly depend on what agents can actually read, understand, verify, and trust.

The brands relying on historical recognition have the most to lose.

The brands building readable, trustworthy, machine legible stores today are the ones most likely to keep the recommendation when memory is no longer enough.

That is exactly the gap the Recommendation Intelligence Framework™ was built to measure.

AI Readability™
AI Understanding™
AI Trust™
Recommendation Intelligence™
Decision Confidence™

The future of AI commerce belongs to brands that are understandable by machines, not just familiar to humans.

Want to Know Where Your Store Stands?

Get a free AI Commerce Score™

https://atomfoundry.dev

Full pets dataset and interactive results

https://atomfoundry.dev/research/state-of-ai-recommendations-pets

Decision Confidence™ Is The Missing Layer Between Recommendation And Revenue

Daniel Pokorný — Tue, 16 Jun 2026 12:39:51 +0000

Decision Confidence™ Is The Missing Layer Between Recommendation And Revenue

Most discussions about AI commerce focus on visibility.

Can AI find your business? Can AI understand your products? Can AI recommend your brand?

These are important questions.

But they are not the final question.

Because recommendation is not the purchase.

A customer can discover your business. A customer can trust an AI recommendation. A customer can visit your website.

And still leave without buying.

Why?

Because recommendation creates consideration.

Purchase requires confidence.

This is where Decision Confidence™ begins.

Decision Confidence™ measures whether customers have enough certainty to move from consideration to purchase. The framework evaluates trust signals, clarity, proof, risk reduction, and decision friction to understand how confidently a customer can make a buying decision.

The Hidden Gap Between Recommendation And Purchase

Most commerce frameworks stop at recommendation.

They assume that if a customer arrives on the website, the hard part is over.

In reality, recommendation only creates an opportunity.

The customer still needs to answer several questions:

Is this business trustworthy?
Is this product right for me?
Am I making the right decision?
What happens if something goes wrong?
Is there a better alternative?

Every unanswered question creates uncertainty.

And uncertainty reduces the probability of purchase.

The sequence looks like this:

AI Readability™

↓

AI Understanding™

↓

AI Trust™

↓

Recommendation Intelligence™

↓

Decision Confidence™

↓

Purchase

↓

Revenue

Recommendation creates attention.

Decision Confidence creates action.

What Is Decision Confidence™?

Decision Confidence™ measures the level of certainty a customer has before making a purchase decision.

It evaluates whether a business provides enough information, proof, reassurance, and trust signals to reduce hesitation and uncertainty.

In simple terms:

Recommendation answers:

Should I consider this business?

Decision Confidence answers:

Am I comfortable enough to buy?

These are not the same thing.

Many businesses successfully generate consideration while failing to generate confidence.

The result is abandoned sessions, delayed decisions, and lost revenue.

Why Confidence Matters

Every purchase involves risk.

Customers constantly evaluate that risk.

They ask themselves:

Will this product meet expectations?
Can I trust this company?
Is the price justified?
What if I need a refund?
What if I choose incorrectly?

The greater the uncertainty, the lower the likelihood of purchase.

Confidence reduces uncertainty.

And reducing uncertainty is often more important than increasing traffic.

The Five Dimensions Of Decision Confidence™

Decision Confidence™ is measured through five core signals.

1. Trust Signals

Does the business provide evidence that it is legitimate and trustworthy?

Examples include:

Reviews
Ratings
Testimonials
Security indicators
Business information
Guarantees

Trust reduces perceived risk.

Without trust, confidence collapses.

2. Clarity

Can customers immediately understand what is being offered?

Confusion creates hesitation.

Customers should instantly understand:

What the product does
Who it is for
Why it is different
What they receive

Clarity reduces cognitive effort.

3. Proof

Can the business demonstrate that its claims are true?

Proof includes:

Case studies
Customer outcomes
Before-and-after examples
Expert endorsements
Independent validation

Claims create interest.

Proof creates belief.

4. Risk Reduction

Does the business actively remove fear from the decision?

Examples include:

Refund policies
Free returns
Free trials
Warranties
Satisfaction guarantees

The lower the perceived downside, the easier the decision becomes.

5. Decision Friction

How difficult is it for a customer to move forward?

Friction includes:

Complicated checkout flows
Excessive form fields
Hidden costs
Missing information
Poor user experience

Every obstacle increases hesitation.

Every hesitation reduces confidence.

Recommendation Without Confidence Creates Revenue Leakage

Many businesses focus heavily on visibility.

Others focus on AI recommendations.

But recommendation alone does not create revenue.

A business can be:

Visible
Readable
Understandable
Trusted
Recommended

And still lose sales.

Because the customer never reaches sufficient confidence to act.

Recommendation generates visitors.

Confidence generates buyers.

This distinction becomes increasingly important as AI systems take on larger roles in product discovery and evaluation.

The Future Of AI Commerce

As AI systems become responsible for more discovery, comparison, and recommendation activity, confidence becomes one of the most important commercial variables.

AI may recommend. AI may compare. AI may shortlist.

But many purchasing decisions still require human approval.

The businesses that win will not simply be the most visible.

They will not simply be the most recommended.

They will be the businesses that create the highest level of decision confidence.

Because confidence converts consideration into action.

And action creates revenue.

The AI Commerce Intelligence Framework™ explains how AI systems move from discovering and understanding businesses to influencing recommendations, purchase decisions, and ultimately revenue. Decision Confidence™ represents the critical layer between recommendation and purchase.

Decision Confidence™ In The AI Commerce Intelligence Framework™

Decision Confidence™ is the fifth layer of the AI Commerce Intelligence Framework™.

AI Readability™ enables access.

AI Understanding™ enables interpretation.

AI Trust™ enables credibility.

Recommendation Intelligence™ enables selection.

Decision Confidence™ enables action.

Without confidence, recommendation has limited commercial value.

Without confidence, purchase does not happen.

Without purchase, revenue does not exist.

Decision Confidence™ measures the final layer before commerce becomes economic outcome.

Explore The Framework

AI Commerce Intelligence Framework™

https://atomfoundry.dev/framework

Decision Confidence™

https://atomfoundry.dev/framework/decision-confidence

Decision Confidence™ is part of the AI Commerce Intelligence Framework™ developed by Atom Foundry.

Building the AI Commerce Intelligence Layer™.

Recommendation Intelligence Is Becoming The Next Layer Of AI Commerce

Daniel Pokorný — Mon, 15 Jun 2026 11:52:52 +0000

Recommendation Intelligence Is Becoming The Next Layer Of AI Commerce

Most businesses are focused on AI visibility.

Some are starting to focus on AI understanding.

A few are beginning to think about AI trust.

Very few are asking an even more important question.

Does AI actually recommend your business?

Because visibility and recommendation are not the same thing.

Understanding and recommendation are not the same thing.

Trust and recommendation are not the same thing.

A business can be visible. A business can be understood. A business can even appear trustworthy.

And still never become the recommendation.

As AI systems become increasingly involved in product discovery, evaluation, and decision-making, recommendation is becoming a critical layer of AI commerce.

The Recommendation Intelligence Framework™ measures how often, how prominently, and for which buyer intents AI systems recommend businesses. Recommendation Frequency™, Recommendation Position™, Recommendation Share™, Competitor Comparison, and Intent Match help quantify recommendation behavior across AI systems.

The Difference Between Visibility And Recommendation

AI Visibility asks a simple question.

Can AI see me?

Recommendation Intelligence asks a different question.

Does AI choose me?

Those are not the same thing.

A business may appear inside AI-generated answers.

It may be cited. It may be mentioned.

Yet never become the actual recommendation.

Visibility creates possibility. Recommendation creates influence.

The Recommendation Problem

Imagine a customer asks:

Best protein powder
Best CRM for startups
Best moisturizer for sensitive skin

An AI system may know dozens of valid options.

The challenge is no longer discovery. The challenge is selection.

Only a handful of businesses will actually be recommended.

Most will remain alternatives.

The difference between being known and being chosen is where Recommendation Intelligence begins.

Why Recommendation Matters

Recommendation systems operate differently than search systems.

Search systems retrieve information.

Recommendation systems prioritize options.

When multiple businesses appear relevant, AI systems must determine which businesses deserve attention.

This is where recommendation becomes important.

Recommendation reduces complexity. Recommendation narrows choice. Recommendation influences decisions.

The businesses that consistently receive recommendations gain a significant advantage over businesses that merely appear.

What Recommendation Intelligence Measures

Recommendation Intelligence™ measures how AI systems recommend businesses, products, and brands.

Several factors influence this outcome.

Recommendation Frequency™

How often does a business appear across recommendations?

Recommendation Position™

Where does the business appear within recommendation lists?

Recommendation Share™

How much recommendation visibility does a business capture compared to competitors?

Competitor Comparison

Which businesses consistently outrank others?

Intent Match

For which buying intents does a business appear?

Together these metrics help quantify recommendation behavior across AI systems.

Recommendation Intelligence™ introduces a measurable framework for understanding recommendation behavior. Recommendation Frequency™, Recommendation Position™, Recommendation Share™, Competitor Comparison, and Intent Match help quantify how AI systems choose businesses during product discovery and evaluation.

Recommendation Is Not Readiness

This is where many businesses become confused.

Readability answers one question.

Can AI access and interpret my business?

Recommendation answers a different question.

Will AI choose my business?

Those are not the same thing.

Recent Atom Foundry research analyzed:

20,000 AI-generated recommendations
1,490 brands
5 ecommerce categories
100 shopping intents

Across every category, Recommendation Frequency™ showed little to no measurable relationship with AI Commerce Score™.

Many highly recommended brands had weak stores.

Many highly optimized stores received little recommendation visibility.

Recommendation appears to operate according to different rules.

The Hidden Cost Of Low Recommendation Visibility

Most businesses never realize when recommendation becomes a limitation.

There is no dashboard. No recommendation ranking report. No recommendation analytics platform.

The business simply appears less often.

Competitors receive more recommendations.

AI systems route more attention elsewhere.

The business is not invisible.

It is simply not being chosen.

Why Recommendation Matters More Than Visibility

Many businesses focus on becoming visible to AI systems.

Others focus on becoming understandable.

Some are beginning to focus on trust.

The next challenge is becoming recommendable.

Visibility creates discovery.

Understanding creates interpretation.

Trust creates confidence.

Recommendation creates selection.

The sequence looks like this.

AI Readability™

↓

AI Understanding™

↓

AI Trust™

↓

Recommendation Intelligence™

↓

Decision Confidence™

↓

Purchase

↓

Revenue

Recommendation is the bridge between trust and action.

Without recommendation, commercial influence remains limited.

The AI Commerce Intelligence Framework™ maps the stages AI systems use to discover, understand, evaluate, recommend, and route customers to businesses. The framework connects AI Readability™, AI Understanding™, AI Trust™, Recommendation Intelligence™, and Decision Confidence™ to the commercial outcomes that ultimately drive revenue.

Learn more about the AI Commerce Intelligence Framework™

https://atomfoundry.dev/framework

Final Thought

The first generation of AI optimization focused on visibility.

The second generation focused on understanding.

The third generation focuses on trust.

The fourth generation will focus on recommendation.

Because recommendation systems do not simply retrieve information.

They choose.

And the businesses that AI consistently choose will increasingly shape the future of commerce.

Recommendation Intelligence™ is the fourth layer of the AI Commerce Intelligence Framework™.

It is where confidence becomes selection.

And where influence begins to become revenue.

Related Frameworks