AI Commerce Research™ · Research Map

How AI Decides

AI does not simply see a brand and recommend it. Somewhere between discovery, retrieval, understanding, evaluation and selection, a decision gets made. Atom Foundry is researching what happens at each stage, one study at a time.

What we know. What we have measured. What we still don't understand.

This page is not a claim that Atom Foundry has solved AI recommendation. It is the opposite. Here is the system we are trying to understand. Here is the evidence we have collected so far. Here are the questions we have not solved yet. Every new study we publish becomes a new piece of this map, not just a new article.

The Decision Path

One possible path from memory to purchase

This is a working research model, not a claim about a fixed AI pipeline. Real systems may skip stages, run several at once, or loop back before answering.

1
Memory
What the model already knows before the conversation starts.
2
Retrieval
What live information the model finds and pulls in.
3
Understanding
What the model concludes your brand actually is.
4
Candidacy
Whether your brand enters the consideration set at all.
5
Evaluation
How the model compares the candidates against each other.
6
Recommendation
Which brand the model actually chooses to say out loud.
7
Stability
Whether the same answer holds up on the next run.
8
Confidence
How certain the model actually is behind that answer.
9
Purchase
Whether the recommendation ever turns into a transaction.
1 · Memory
Partially Measured

What does AI already know?

Before a shopping question is ever asked, a language model already carries some notion of which brands exist and how they tend to be positioned. That prior knowledge comes from training, not from the live conversation, and it can shape an answer before retrieval or evaluation even begin.

What we've found

Brand prominence appears to matter, a little. Wikipedia fame alone explains 1.2% of recommendation frequency. Widening to four public signals, cross-lingual reach, domain age, and recent media volume, raises that to just 11.2% combined, still far under the 61.4% the model's own past behavior explains about itself. Measured store quality explains almost none of it.

What we don't know

What exactly enters a model's memory, how strongly a brand is represented there, and how that internal representation shapes a recommendation later. Four easily-measurable public-footprint signals have now been ruled out as the main answer; the actual mechanism remains genuinely open.

2 · Retrieval
Measured

What information does AI find?

When a model has live web access, it can supplement whatever it already knows with real time retrieval. A single switch, search on or off, turns out to matter more than almost anything else we have tested.

What we've found

77% of product recommendations changed when web search was switched on, same model, same questions, the only change was retrieval.

What this tells us

Retrieval is not a minor adjustment to AI recommendations. In many categories, it can fundamentally change the answer, and how much it changes the answer differs by category.

3 · Understanding
Measured

What does AI think your brand actually is?

There is a difference between AI being able to read your information and AI understanding what that information means. A store may clearly describe itself as a performance running brand, while a model interprets it primarily as a fashion brand.

What we measure

Category understanding, product understanding, attribute understanding, brand positioning, intent alignment, and the semantic relationships between them.

4 · Candidacy
Measured

Does your brand enter the consideration set?

A brand cannot be recommended if it never becomes a candidate.

Getting considered and getting selected are different problems, governed by different things. Our full population study identified one store factor that consistently separates brands that are ever recommended from brands that never are, and it holds up under every check we ran.

What we've found

We went further and tested it causally. A brand with zero history, zero reviews, and zero press was invented from scratch and pitted against an entrenched category leader across 36 shopping intents. It won 0 of 360 runs. Giving it a review score changed that immediately, it started winning 53.1% of comparisons. Press mentions and sales-volume claims barely moved the needle, 1.9% and 0.3%.

5 · Evaluation
Measured

How does AI compare the candidates?

Once a handful of candidates are inside the consideration set, the model has to compare them somehow. By price? By quality signals? By reviews? By authority? By familiarity? By product fit? By evidence? By category relevance?

What we've found

We handed the model one comparison fact at a time between two genuinely close brands. A better star rating flipped the verdict 160 of 160 runs, 100%, zero exceptions across 16 contested pairs. A longer, more specific spec list won 81.9% of runs. A lower price and faster availability won only 60.6%, barely above a coin flip.

What we don't know

How these signals interact when they conflict, for example a cheaper product with a worse rating, and whether the same hierarchy holds across categories.

6 · Recommendation
Measured

Which brand does AI actually choose?

This is where Recommendation Intelligence™ begins. We measure Recommendation Frequency™, Recommendation Share™, Recommendation Position™, the reasons a model gives for its pick, and the differences between the brand that wins and the brands that don't.

The open part

We can measure who gets recommended and how often. The mechanism behind the selection itself is not fully explained.

We can also measure what reasons a model gives for its recommendation. We do not yet know whether those stated reasons are the actual causal drivers of the selection, or a justification constructed after the fact.

7 · The Winner vs The Loser
Active Research

We know who won. Now we want to know why.

Every recommendation has a winner and, quietly, a brand that lost. We can already observe both sides of that outcome. What separates them is one of the questions we consider most important to answer next.

Shopping intent: "Best waterproof running shoe for wide feet under $200"
Winner
Brand A
Last place
Brand B
Compare
PriceProduct attributesReviewsContentAuthorityBrand familiaritySemantic positioningEvidenceStructured informationAI understandingOther observable signals

What separates the brand AI selects from the brand it leaves behind?

8 · Stability
Emerging

Does the decision persist?

A model might recommend a brand today, but will it recommend the same brand again tomorrow, or even in the next run of the same question? We measure recommendation stability, position stability, whether the same winner keeps winning, and how much changes across repeated runs.

What we've found so far

Early results show recommendation behavior can be more stable than expected, even in places where drift would seem likely. That stability is itself a variable we now factor into other studies.

9 · Confidence
Open Research

How certain is AI about the recommendation?

A model can say it recommends Brand A. But how strong are the reasons behind that pick? How much would the answer change with slightly different phrasing? How close are the alternatives, and what is the actual margin between the first choice and the second?

This is one of the biggest open questions in the whole decision path. Recommendation Confidence™ is the metric we expect to grow out of answering it.

10 · Purchase
Open Research

Does recommendation actually influence the transaction?

Recommendation is not purchase.

A model may recommend a brand, and the person on the other end might still click somewhere else, change their mind, compare more options, abandon the purchase entirely, or buy from a competitor anyway. How an AI recommendation translates into actual commercial behavior is the final, and least understood, link in this chain.

11 · What We Know, What We Don't

The map so far

Memory
Prominence and fame appear to matter more than measured store quality.
Partially Measured
Retrieval
Search access can dramatically change which brand gets recommended.
Measured
Understanding
Semantic understanding of a brand can be measured directly.
Measured
Candidacy
Zero-history brands lose every run until reviews appear, then they win over half.
Measured
Evaluation
A better rating decides the comparison almost every time; price barely moves it.
Measured
Recommendation
We can measure who wins, and describe why they are described as winning.
Measured
Stability
Recommendation persistence across repeated runs can be measured.
Emerging
Confidence
Decision confidence remains largely unexplored.
Open Research
Purchase
The relationship between recommendation and transaction remains open.
Open Research
12 · The Open Questions

What we still don't know

Why does AI select one candidate over another?
What determines entry into the consideration set?
How much of recommendation behavior comes from training data provenance?
How does AI weigh evidence against familiarity?
What creates recommendation confidence?
When does a recommendation actually change purchasing behavior?

Every unanswered question is a research opportunity.

13 · Research Library

Research behind the map

Grouped by which stage of the decision path each study speaks to, not by publish date.

The map is incomplete.

We're still discovering what happens between being seen and being chosen.