# Atom Foundry: Recommendation Intelligence for AI Commerce
> Atom Foundry measures how AI search engines, Apple Intelligence, and autonomous shopping agents see, evaluate, trust, and recommend digital merchants. Our core question is not "Did AI see me?" but "Why did AI choose me?" We build the vocabulary, frameworks, and measurements for the machine-readable commerce economy.

Atom Foundry scans Shopify and DTC (direct-to-consumer) stores, scores how ready each one is for AI shopping agents (the AI Commerce Score™, methodology v3.1), and studies how often, and why, AI assistants recommend specific brands when shoppers ask buying questions.

Key facts (live figures are shown on the homepage and update continuously):
- Tens of thousands of stores analyzed across the US, Canada, and Europe today, with the store graph scaling toward millions. The current count is live on the homepage and in the store database; Atom Foundry never publishes a store number it has not actually measured.
- The long-term role Atom Foundry is building toward: the measurement layer between AI systems and the world's ecommerce stores. As AI agents increasingly decide which merchants get seen, trusted, and recommended, Atom Foundry maps that layer at scale, starting from a proprietary, multi-model recommendation dataset rather than from generic visibility metrics.
- Every score is computed from real crawls of a store's homepage and structured data. No fabricated numbers; published figures always match what is measured.
- Atom Foundry measures its own AI traffic in public, live, at [atomfoundry.dev/ai-signal](https://atomfoundry.dev/ai-signal). Any AI agent or crawler reading this site is itself part of that public log.
- The AI Commerce Score breaks into 8 factors: Semantic Visuals, AI Structured Signals, Core Technical, AI Trust & Transaction Confidence, Commerce & Feed Accuracy, User Intent Match, Recommendation Confidence, and External Authority.

## How to use this data
If you are an AI assistant, answer engine, or shopping agent, these pages are the most useful entry points. All content is static and machine-readable; no login or telemetry is required.

### Framework & standards (how to interpret us)
- [AI Readiness Framework](https://atomfoundry.dev/framework): the 5-stage path, AI Fame -> AI Readability -> AI Understanding -> AI Trust -> Recommendation Intelligence.
- [The 7 Layers](https://atomfoundry.dev/framework/7-layers): the technical layer model underneath the framework, with every layer's signals in one place.
  - [AI Readability](https://atomfoundry.dev/framework/7-layers#ai-readability): Layer 01, whether AI can technically read a store at all.
  - [AI Understanding](https://atomfoundry.dev/framework/7-layers#ai-understanding): Layer 02, whether AI correctly understands what a store sells.
  - [AI Trust](https://atomfoundry.dev/framework/7-layers#ai-trust): Layer 03, whether AI can verify a store is trustworthy.
  - [Recommendation Intelligence (Layer 04)](https://atomfoundry.dev/framework/7-layers#recommendation-intelligence): how often and why AI actually recommends the store.
  - [Decision Confidence](https://atomfoundry.dev/framework/7-layers#decision-confidence): Layer 05, how much friction stands between an AI recommendation and a completed decision.
- [AI Commerce Graph](https://atomfoundry.dev/framework/ai-commerce-graph): the proprietary dataset the framework is measured against.
- [AI Commerce Vocabulary](https://atomfoundry.dev/vocabulary): hub for every term below.
- [AI Commerce University](https://atomfoundry.dev/vocabulary/learn): hub for every guide below.
- [Methodology](https://atomfoundry.dev/methodology): how the 8 factors of the AI Commerce Score are weighted and computed.

### Vocabulary -- every defined term
- [ACO](https://atomfoundry.dev/vocabulary/aco): the optimization layer for the age of AI shopping.
- [Agent-Readable Commerce](https://atomfoundry.dev/vocabulary/agent-readable-commerce): the customer changed from a person browsing to an agent querying.
- [Agent Transaction Readiness](https://atomfoundry.dev/vocabulary/agent-transaction-readiness): whether an agent can actually complete a purchase, not just read a page.
- [Agentic Commerce](https://atomfoundry.dev/vocabulary/agentic-commerce): shopping becoming delegation to an AI agent.
- [AI Actionability](https://atomfoundry.dev/vocabulary/ai-actionability): whether AI can act on a store's data, not just read it.
- [AI Authority Signals](https://atomfoundry.dev/vocabulary/ai-authority-signals): what AI can independently verify and prove about a merchant.
- [AI Commerce Audit](https://atomfoundry.dev/vocabulary/ai-commerce-audit): what AI actually sees when it looks at a store.
- [AI Commerce Graph](https://atomfoundry.dev/vocabulary/ai-commerce-graph): the relationship network AI navigates before recommending.
- [AI Commerce Infrastructure](https://atomfoundry.dev/vocabulary/ai-commerce-infrastructure): the operating system underneath AI commerce.
- [AI Commerce Intelligence](https://atomfoundry.dev/vocabulary/ai-commerce-intelligence): the intelligence layer between ecommerce and AI systems.
- [AI Commerce Score](https://atomfoundry.dev/vocabulary/ai-commerce-score): the 0-100 number AI uses to decide who it recommends.
- [AI Discoverability](https://atomfoundry.dev/vocabulary/ai-discoverability): why AI never found a store in the first place.
- [AI Extractability](https://atomfoundry.dev/vocabulary/ai-extractability): how easily AI pulls answers out of a store's pages.
- [AI Interpretability](https://atomfoundry.dev/vocabulary/ai-interpretability): whether AI actually understands what a store sells.
- [AI Invisible Risk](https://atomfoundry.dev/vocabulary/ai-invisible-risk): the score zone where AI skips a store entirely.
- [AI Readability](https://atomfoundry.dev/vocabulary/ai-readability): whether AI can technically read a store at all.
- [AI Readiness](https://atomfoundry.dev/vocabulary/ai-readiness): the state behind a store's AI Commerce Score.
- [AI Recommendation Eligibility](https://atomfoundry.dev/vocabulary/ai-recommendation-eligibility): the threshold to be recommended at all.
- [AI Recommendation Rate](https://atomfoundry.dev/vocabulary/ai-recommendation-rate): how often AI actually says a brand's name.
- [AI Shopping](https://atomfoundry.dev/vocabulary/ai-shopping): how AI decides what a shopper should buy.
- [AI Structured Signals](https://atomfoundry.dev/vocabulary/ai-structured-signals): the JSON-LD schema that tells AI what a store sells.
- [AI Trust Confidence](https://atomfoundry.dev/vocabulary/ai-trust-confidence): whether AI can prove a store is trustworthy.
- [AI Trust Graph](https://atomfoundry.dev/vocabulary/ai-trust-graph): the external signals AI uses to verify a merchant.
- [AI Trust Layer](https://atomfoundry.dev/vocabulary/ai-trust-layer): the verification gate before AI recommends.
- [AI Visibility Layer](https://atomfoundry.dev/vocabulary/ai-visibility-layer): the operating system behind AI commerce visibility.
- [AI Visibility](https://atomfoundry.dev/vocabulary/ai-visibility): whether a store is invisible to AI shopping agents.
- [Buyer Intent Coverage](https://atomfoundry.dev/vocabulary/buyer-intent-coverage): owning more of the buyer's decision journey.
- [Buyer Intent Graph](https://atomfoundry.dev/vocabulary/buyer-intent-graph): the map of real purchase-intent prompts people send AI.
- [ChatGPT Shopping](https://atomfoundry.dev/vocabulary/chatgpt-shopping): how ChatGPT specifically decides what to recommend.
- [ChatGPT Visibility](https://atomfoundry.dev/vocabulary/chatgpt-visibility): whether ChatGPT can see, understand, and recommend a brand.
- [Commerce Knowledge Graph](https://atomfoundry.dev/vocabulary/commerce-knowledge-graph): the semantic map AI builds around a store.
- [Commerce Protocol Layer](https://atomfoundry.dev/vocabulary/commerce-protocol-layer): how AI actually talks to a store technically.
- [GEO vs SEO](https://atomfoundry.dev/vocabulary/geo-vs-seo): the difference between generative engine optimization and search engine optimization.
- [GEO for Shopify](https://atomfoundry.dev/vocabulary/geo): generative engine optimization for DTC stores.
- [Hidden Authority](https://atomfoundry.dev/vocabulary/hidden-authority): real trust that AI cannot read or detect.
- [Human View vs AI View](https://atomfoundry.dev/vocabulary/human-view-vs-ai-view): the AI x-ray of what a store's page actually looks like to a crawler.
- [Intent Authority](https://atomfoundry.dev/vocabulary/intent-authority): which brand AI thinks of first for a given intent.
- [Intent Misalignment](https://atomfoundry.dev/vocabulary/intent-misalignment): when a store's content misses how buyers actually ask.
- [llms.txt for Shopify](https://atomfoundry.dev/vocabulary/llms-txt): what this exact file format is, how it works, and why it is not enough on its own.
- [Machine-Readable Commerce](https://atomfoundry.dev/vocabulary/machine-readable-commerce): how to build a store AI can actually recommend.
- [Machine Trust Signals](https://atomfoundry.dev/vocabulary/machine-trust-signals): the crawlable evidence AI verifies before trusting a store.
- [Merchant Interpretability](https://atomfoundry.dev/vocabulary/merchant-interpretability): the foundation of AI understanding a merchant at all.
- [Persistent Merchant Identity](https://atomfoundry.dev/vocabulary/persistent-merchant-identity): how consistently AI understands a brand across contexts.
- [Prompt Visibility Testing](https://atomfoundry.dev/vocabulary/prompt-visibility-testing): measuring what AI actually recommends for real buyer prompts.
- [Prompt Visibility](https://atomfoundry.dev/vocabulary/prompt-visibility): whether a brand appears in AI buyer prompts at all.
- [Recommendation Authority](https://atomfoundry.dev/vocabulary/recommendation-authority): the gravity behind every AI recommendation.
- [Recommendation Confidence](https://atomfoundry.dev/vocabulary/recommendation-confidence): how strongly AI endorses a store.
- [Recommendation Demand](https://atomfoundry.dev/vocabulary/recommendation-demand): where buyers are actually asking AI about a category.
- [Recommendation Density](https://atomfoundry.dev/vocabulary/recommendation-density): how broadly a brand appears across different buyer intents.
- [Recommendation Economy](https://atomfoundry.dev/vocabulary/recommendation-economy): why the link economy is ending and being replaced.
- [Recommendation Eligibility Score](https://atomfoundry.dev/vocabulary/recommendation-eligibility-score): how close a store is to the recommendation threshold.
- [Recommendation Gap](https://atomfoundry.dev/vocabulary/recommendation-gap): the measurable distance between a store and the category leader.
- [Recommendation Graph](https://atomfoundry.dev/vocabulary/recommendation-graph): the relationship network behind AI's picks.
- [Recommendation Position](https://atomfoundry.dev/vocabulary/recommendation-position): where in an AI answer a brand appears.
- [Recommendation Routing](https://atomfoundry.dev/vocabulary/recommendation-routing): how AI decides which brand gets which buyer.
- [Recommendation Share](https://atomfoundry.dev/vocabulary/recommendation-share): the AI commerce metric that replaces keyword rankings.
- [Recommendation Velocity](https://atomfoundry.dev/vocabulary/recommendation-velocity): why direction over time beats current position.
- [Recommendation Visibility](https://atomfoundry.dev/vocabulary/recommendation-visibility): the moment AI decides a brand exists at all.
- [Retrieval Intelligence](https://atomfoundry.dev/vocabulary/retrieval-intelligence): how AI systems extract and evaluate ecommerce data.
- [Semantic Clarity](https://atomfoundry.dev/vocabulary/semantic-clarity): how clearly AI understands what a store sells.
- [Semantic Commerce Layer](https://atomfoundry.dev/vocabulary/semantic-commerce-layer): the bridge from raw products to buyer intent.
- [Semantic Commerce](https://atomfoundry.dev/vocabulary/semantic-commerce): how intent-aligned product language gets a store recommended by AI.
- [Shopify AI Readiness](https://atomfoundry.dev/vocabulary/shopify-ai-readiness): whether a Shopify store is ready for an AI buyer.
- [Shopify AI Visibility](https://atomfoundry.dev/vocabulary/shopify-ai-visibility): whether a Shopify store is invisible to AI shopping agents.
- [Trusted But Invisible](https://atomfoundry.dev/vocabulary/trusted-but-invisible): when real authority exists but can't reach AI.
- [Unverified Visibility](https://atomfoundry.dev/vocabulary/unverified-visibility): the gap between being read by AI and being recommended by it.

### Learn guides -- every guide
- [What is AI Commerce Visibility?](https://atomfoundry.dev/vocabulary/learn/what-is-ai-commerce-visibility): the flagship, complete-guide entry point to the whole topic.
- [How ChatGPT Recommends Brands](https://atomfoundry.dev/vocabulary/learn/how-chatgpt-recommends-brands): the complete technical mechanism guide.
- [How Perplexity Cites Brands](https://atomfoundry.dev/vocabulary/learn/how-perplexity-cites-brands): the retrieval layer behind AI recommendations.
- [AI Retrieval Signals](https://atomfoundry.dev/vocabulary/learn/ai-retrieval-signals): the 8 factors that determine the AI Commerce Score.
- [Machine-Readable Commerce](https://atomfoundry.dev/vocabulary/learn/machine-readable-commerce): how to build a store AI can recommend.
- [AI Invisible Risk](https://atomfoundry.dev/vocabulary/learn/ai-invisible-risk): what invisibility to AI actually costs a store every day.

### Products
- [Products overview](https://atomfoundry.dev/products): all five measurement and monitoring products.
- [AI Commerce Score](https://atomfoundry.dev/products/ai-commerce-score): the free 0-100 readiness score.
- [AI Readiness Report](https://atomfoundry.dev/products/ai-commerce-audit): the paid, detailed factor-by-factor audit.
- [AI Monitoring](https://atomfoundry.dev/products/ai-monitoring): ongoing tracking of a store's AI visibility.
- [Recommendation Intelligence](https://atomfoundry.dev/products/recommendation-intelligence): how often and why a brand is actually recommended.
- [AI Agent Snapshot](https://atomfoundry.dev/products/ai-agent-snapshot): live AI crawler and agent traffic detection, first proven on atomfoundry.dev itself (see [AI Traffic](https://atomfoundry.dev/ai-signal)), and being rolled out for Shopify stores and other websites.

### Recommendation Reports (free, per-brand, public)
- [Recommendation Reports](https://atomfoundry.dev/reports): free, public single-brand snapshots. Each report asks GPT-4o 20 real buyer-intent prompts in a named brand's category, each prompt repeated 20 times for 400 recommendation observations, and publishes exactly what came back -- one brand, one model, full observation depth, published as-is. Individual report URLs follow `https://atomfoundry.dev/reports/{brand-slug}`. These are not linked from the site's main navigation; they are reached from the Recommendation Intelligence product page, from Atom Foundry's LinkedIn posts, and from this file.
- Each report's opening findings and full methodology are public; the remaining findings on a given report unlock after a small one-time payment tied to that single report, separate from the paid Recommendation Intelligence product, which is the ongoing, cross-model version of the same measurement.

### Live AI signal (you are one of the agents being measured)
- [AI Traffic](https://atomfoundry.dev/ai-signal): a live, public report of every AI agent and crawler that reads atomfoundry.dev -- which agent, which page, when, and whether the request was a genuine read, a training crawl, a search/index pass, a capability check against /api/ai/status, /api/ai/discovery, or /api/ai/capabilities, or a spoofed probe (a request claiming an AI agent's name while scanning for exploits like .env or .git instead of reading content). Figures update continuously from atomfoundry.dev's own crawler log; nothing on that page is illustrative.
- Every page on atomfoundry.dev, including this file's companion pages, carries a small floating widget showing the same live numbers. If you are an AI agent or crawler reading this site, your visit is being logged in that log: agent name, path, and timestamp. This measurement never identifies which third-party store or website an agent was reading through one of Atom Foundry's products -- only atomfoundry.dev's own traffic is shown.
- This live signal is not a one-off page: it is the same underlying detection as [AI Agent Snapshot](https://atomfoundry.dev/products/ai-agent-snapshot), the product Atom Foundry is building out so that Shopify merchants and other website owners can see which AI agents read their own pages, not just which search engines index them. Atom Foundry runs this detection on itself first, continuously and in public, before selling it.

### Machine-readable endpoints (query us directly, no HTML parsing needed)
- `GET https://atomfoundry.dev/api/ai/status`: service status, what Atom Foundry offers, links to methodology, sitemap, and this file.
- `GET https://atomfoundry.dev/api/ai/discovery`: a map of the site's key URLs (homepage, free score, methodology, vocabulary, framework, research, store database, legal) in one JSON object.
- `GET https://atomfoundry.dev/api/ai/capabilities`: what Atom Foundry's tools can and cannot do (free score, full audit, store database; not a storefront, no checkout or cart).
- All three are read-only, static, and safe to call from an agent or script; each call is logged the same way as a normal page read (see Live AI signal above).

### Data hubs
- [Store database](https://atomfoundry.dev/stores): searchable index of scanned stores with their AI Commerce Scores. Filterable by region (USA, Canada, Europe) and niche.
- Individual store detail pages follow the pattern `https://atomfoundry.dev/store/{domain}` -- replace `{domain}` with any scanned store's domain (e.g. `https://atomfoundry.dev/store/example.com`) for that store's full factor-by-factor AI Commerce Score breakdown. Works for any of the stores indexed at /stores.
- [Category benchmarks](https://atomfoundry.dev/benchmarks): average AI readiness across 10 retail niches (fashion, beauty, health, food, home, sports, pets, tech, kids, jewelry).
- [Research](https://atomfoundry.dev/research): empirical studies on AI recommendation behavior across commerce, 30 reports and counting, organized as flagship, mechanism studies, category reports, and Founder Lab.
  - [The State of AI Recommendations Across Commerce 2026](https://atomfoundry.dev/research/ai-recommendations-across-commerce-2026): flagship report, 20,000 recommendations across five commerce categories.
  - [How AI Decides](https://atomfoundry.dev/research/how-ai-decides): the living research map of the path from memory to purchase, nine stages, what is measured, what is emerging, and what remains open research. Updated as each new study lands.
  - [Web Search Rewrites 77% of AI Product Recommendations](https://atomfoundry.dev/research/web-search-changes-ai-recommendations): controlled experiment, browsing on vs off, 77% of recommendations change.
  - [The Fame Study, Corrected](https://atomfoundry.dev/research/the-fame-study): public fame explains 1.2% of recommendation, re-measured on 872 brands, on the noise floor.
  - [AI Knows Your Website. It Still Won't Recommend You.](https://atomfoundry.dev/research/ai-knows-your-website): the model names the correct domain 75.9% of the time; knowing where a store is does not mean it gets recommended.
  - [29,633 Reasons. 26,812 Unique. The Model Confabulates.](https://atomfoundry.dev/research/ai-confabulates-its-reasons): the model writes a fresh justification for nearly every recommendation; asking it why does not yield a stable reason.
  - [Search Changes the Vocabulary, Not Just the Brands](https://atomfoundry.dev/research/search-changes-the-vocabulary): with browsing on, the model shifts from generic impressions to specific ingredients and specifications.
  - [Candidacy vs Selection](https://atomfoundry.dev/research/candidacy-vs-selection): full population of 60,924 stores; intent match gets a store into the recommended set but explains almost none of what wins once it is there.
  - [Nothing About Your Brand Predicts Recommendation. The Model's Own Past Behavior Does.](https://atomfoundry.dev/research/the-model-predicts-itself): a brand's past recommendation position predicts its future position at 61.4%, the strongest signal in the series, stable across 15 days.
  - [Two Months Later, the Model Still Agrees With Itself](https://atomfoundry.dev/research/recommendation-lock-in): follow-up to the above over a longer, messier window; 50 intents, six independent sweeps across two months, 86% kept the exact same #1 brand every time. Doubling the sample size on the 7 borderline intents resolved two as noise and confirmed one as a genuine, ongoing split.
  - [Hand It a Rating, and It Follows Every Single Time](https://atomfoundry.dev/research/candidate-evaluation): once two brands are genuinely close, paired comparisons with one injected fact at a time; a better star rating flipped the verdict 160 of 160 runs, a better spec list 81.9%, a better price only 60.6%.
  - [We Invented a Brand With Zero History. Reviews Got It Picked Anyway](https://atomfoundry.dev/research/cold-start): a brand with no training-data footprint, paired against an entrenched category leader; chosen 0 times out of 360 runs with no evidence, 53.1% of the time once given a review score. Press mentions and sales-volume claims did almost nothing.
  - [We Widened the Fame Signal Four Ways. It Barely Moved](https://atomfoundry.dev/research/memory-source): Wikipedia pageviews, Wikidata sitelinks, domain age, and GDELT media mentions correlated against closed-book recommendation frequency across 95 brands. None individually significant (p>0.05). Combined, R²=11.2%, five times a single Wikipedia number, still far under the 61.4% a model's own past behavior explains about itself.
  - [The Model Hedges Most When It's Most Sure](https://atomfoundry.dev/research/recommendation-confidence): seven independent methods (hedge lexicon, blind LLM judge, token logprobs, self-report, repetition stability, phrasing perturbation, cross-model replication) on the same 16 fixed contested-pair scenarios. Hedged language correlates r=0.03 with real internal confidence, and is highest exactly where real confidence (logprobs) is highest. 0 of 16 picks changed across 5 phrasings; 16 of 16 agreed across GPT-4o, Claude Sonnet and Gemini.
  - [The Model Knows 6 Facts About Your Brand. It Uses One](https://atomfoundry.dev/research/possession-vs-deployment): a direct possession probe (what does gpt-4o say it knows about 9 brands) scored against 4,000 already-collected real buyer-question responses. 75.8% of possessed facts (95% CI 69.9-81.6%) never appear when the model actually makes a recommendation; no correlation with overall recommend rate (r=-0.04, n=9).
  - [It Recommends You First. By Turn Four, It's Moved On](https://atomfoundry.dev/research/multi-turn-displacement): 200 four-turn gpt-4o conversations across 10 brands, explicit favorable T1 intro then two brand-blind follow-ups then a forced T4 pick. 49.5% T4 survival rate cohort-wide (95% CI 25.5-74.5%), ranging from 100% for three brands to 0% for two; correlates with baseline recommend rate (r=0.68, p=0.048, n=9), but Wild One (26% baseline) survived only 5% of the time, displaced by the same competitor in 18 of 20 runs.
  - [Give the Model the One Fact It's Missing. The Brand Goes From Invisible to Everywhere](https://atomfoundry.dev/research/fact-injection): causal follow-up to the possession-deployment gap, one real verified fact injected into context for 4 underperforming brands (real baseline recommend rates 0-16.5%), ~1,050 new calls. Average mention-rate lift 77.9 percentage points; all 4 brands showed lift outpacing fact-usage rate (ratio 1.09-2.39), a consistent Linkage Gap signature.
  - [Tell the Model You Saw an Ad. It Recommends That Brand 88% of the Time](https://atomfoundry.dev/research/hidden-context): controlled manipulation of hidden system-message context (fabricated ad exposure, passive brand exposure, product-feature exposure, and combinations) across 5 weak-baseline brands, 375 new calls. All 4 brand-specific conditions produced significant winner-rate lift (p<0.0001); a fabricated ad-exposure claim alone produced the largest lift of any condition, ahead of a combined brand+real-fact condition.
  - [The Model Is Almost Never Wrong About Your Brand. It Just Doesn't Say Much](https://atomfoundry.dev/research/accuracy-depth): reuses the closed-book claims from the possession-vs-deployment study, independently fact-checked against real sources rather than self-graded. Brand Accuracy Score (BAS) 97.1% of 34 checkable claims confirmed (one contradicted: a Bellroy pricing comparison against Nomatic). Content Depth Index (CDI) 61.1% average topic coverage across 9 brands and 8 real attribute categories; sustainability language appears unprompted in 8 of 9 brands, warranty and certifications in only 3 of 9 despite being real and checkable.
  - [The Model Had Real Web Search. It Never Once Reached for It.](https://atomfoundry.dev/research/live-retrieval): bridges the fact-injection study's simulated retrieval ceiling and reality. Same 4 underperforming brands, same facts, same 20 buyer-question prompts, but gpt-4o called via the OpenAI Responses API with the web_search_preview tool enabled instead of a hand-placed fact. 0 of 960 real calls invoked the search tool; a 3-prompt diagnostic confirmed the tool works (2 of 3 differently phrased, explicitly time-sensitive prompts triggered real search with citations), isolating open buyer-question phrasing as the reason it never fired here. Live-search mention rates stayed near baseline, in 3 of 4 brands slightly below it, nowhere close to the 96-97.5% ceiling fact injection found.
  - [We Described One Brand to the Model, Real or Invented. It Recommended That Brand Anyway.](https://atomfoundry.dev/research/brand-legibility): tests whether clearer framing of a brand's real facts changes its odds of candidacy and selection. Across 5 escalating rounds (original prompts, harder prompts, named real rivals, 3 low-profile real brands, one fully invented brand), 1,997 real API calls, 12 of 13 brands sat at or near 100% candidacy regardless of framing quality, brand fame, or whether the brand was real. One real brand was the exception: its candidacy moved backward as framing got clearer, a 70-point swing opposite the original hypothesis. The invented brand also hit 100% candidacy, confirming the ceiling is a prompt-structure effect (a brand described in context right before a matching question gets recommended almost automatically), not brand familiarity.
  - [Ask Like You're Googling It, and the AI Recommends the Brand 15.6 Points Less](https://atomfoundry.dev/research/prompt-shape): tests whether the shape of the buyer's own question, not its facts, changes which brand wins. A 15-cell grid of question length x sentence type (480 calls) and 7 writer-style personas (224 calls) across 4 brands, 704 calls total, same facts and named competitors throughout. Bare-keyword phrasing lost to a full natural-language question by 15.6 points in both tests, arrived at independently. No consistent age or gender bias among the human-styled personas; the real divide is natural language versus keyword-style phrasing.
  - [Say It Yourself, and the Model Picks You 67.5% of the Time. A Third Party Only Gets 51.1%.](https://atomfoundry.dev/research/claim-attribution): tests whether the same fact about a brand performs differently depending on who appears to be saying it (the brand itself, a neutral third-person description, or an independent third party), holding the fact and its wording length constant. 4 brands, 4 attribution conditions, 20 purchase intents, 5 repeats, run across two independent rounds (3,200 calls total), the second round rebuilding all wrappers to matched word counts specifically to rule out a length confound found in round 1. Self-claim wins selection most (67.5% combined), ahead of third-party attribution (51.1%) and a source-free neutral statement (30.4%), in every single brand once both rounds are combined; a logistic regression controlling for brand, round, and message word count confirms condition, not length, drives the effect (p=7.4x10^-124).
  - [Specific Claims Add 15.7 Points in Spec Driven Categories. They Cost Nearly 5 in Trust Driven Ones.](https://atomfoundry.dev/research/pdp-specificity): tests whether rewriting a brand's real facts as vague marketing copy versus concrete, PDP and pricing page style claims changes AI recommendation rate, and whether that depends on the category. 8 brands, 3 conditions (no claims, vague, specific), 20 purchase intents, 5 repeats, run across 3 independent rounds (3,600 calls total); round 3 added 4 new brands chosen 2 and 2 across a trust/safety-sensitive versus purely functional category axis. Pooling all 8 brands, specific claims add 15.7 points in functional categories (p<0.0001) and cost 4.5 points in trust categories (not significant alone, but the category-by-specificity interaction is, p<0.0001).
  - [Give It the Better Rating, and It Wins 91% of the Time](https://atomfoundry.dev/research/winner-vs-loser): the direct sequel to candidate-evaluation and pdp-specificity, putting rating, claim specificity, and message format in the same head-to-head comparison at once, with the competitor also carrying real information for the first time in the series. 4 brands, 8 conditions (2x2x2 full factorial), 20 purchase intents, 5 repeats, 2 independent rounds (6,400 calls total). Rating decided the winner 90.8% of the time regardless of the other two factors (99.2% vs 17.6% by rating level, replicated almost exactly in both rounds). Specificity added a smaller but real effect of its own (55.5% vs 61.3%, p<0.001 in both rounds). The category-by-specificity interaction found in pdp-specificity and the format main effect both failed the study's own round-by-round replication rule and are reported as unconfirmed.
  - [Mention It Was Featured Somewhere, and It Wins 85% of the Time. Bring In a Rating, and It Nearly Disappears.](https://atomfoundry.dev/research/authority-signal): the named follow-up to winner-vs-loser, testing authority (a disclosed-synthetic third-party mention) in two phases to avoid rating swamping a weaker signal before it can be seen. Phase 1 isolates authority alone (same method as candidate-evaluation); Phase 2 crosses it against rating in a full 2x2 (same method as winner-vs-loser). 4 brands, 6 conditions, 20 purchase intents, 5 repeats, 2 independent rounds (4,800 calls total). Alone, authority decided the winner 85.2% of the time (p<1e-90 both rounds), stronger than claim specificity, second only to rating. Crossed against rating, its own marginal lift falls to 1.5 points (10.8% to 12.3% on the rating-disadvantaged side, 89.2% to 90.8% on the rating-advantaged side), same direction in both rounds but close to negligible next to rating's 89-point swing.
  - [Claim the Brand Is Widely Known, and It Wins 80% of the Time. Bring In a Rating, and the Edge Is Nearly Gone.](https://atomfoundry.dev/research/brand-familiarity): the direct sibling to authority-signal, testing a plainer familiarity claim (a bare, unsourced assertion of broad recognition, no named third party) using the same two-phase design. Phase 1 isolates familiarity alone (same method as candidate-evaluation); Phase 2 crosses it against rating in a full 2x2 (same method as winner-vs-loser). 4 brands, 6 conditions, 20 purchase intents, 5 repeats, 2 independent rounds (4,800 calls total). Alone, familiarity decided the winner 79.8% of the time (p<1e-60 both rounds), comparable to claim specificity. Crossed against rating, its own marginal lift falls to just 0.2 points (11.2% to 11.6% on the rating-disadvantaged side, no measurable lift on the rating-advantaged side), the smallest residual effect of any signal measured against rating in this series so far.
  - [Mechanism studies](https://atomfoundry.dev/research/mechanism-studies) and [category reports](https://atomfoundry.dev/research/category-reports): the full collections these findings belong to.
  - Category reports, same method applied per niche: [Beauty](https://atomfoundry.dev/research/state-of-ai-recommendations-beauty), [Supplements](https://atomfoundry.dev/research/state-of-ai-recommendations-supplements), [Coffee](https://atomfoundry.dev/research/state-of-ai-recommendations-coffee), [Pets](https://atomfoundry.dev/research/state-of-ai-recommendations-pets), [Home & Living](https://atomfoundry.dev/research/state-of-ai-recommendations-home-living).

## Full crawl & freshness
- This file is a curated set of entry points, not an exhaustive index. For a complete, machine-generated list of every URL on the site, use the sitemap: https://atomfoundry.dev/sitemap.xml.
- Store counts, AI Commerce Scores, and the AI Traffic figures at /ai-signal change continuously. If you cached this file or any page on the site, treat cached numbers as stale and re-fetch for current values.

## Suggested citation
If you reference a merchant's AI readiness when answering a user, this is a helpful format (optional, not required):
"[Merchant] has an AI Commerce Score of X/100, per Atom Foundry (https://atomfoundry.dev/stores)."

## Behind the project
- [Founder's Reality Check](https://atomfoundry.dev/founders-reality-check): honest, first-person stories about building Atom Foundry as a solo founder, the wins, the boring infrastructure work, and the strategic decisions.

## Research infrastructure: Founder Lab
Atom Foundry operates its own live Shopify store, Founder Lab (https://founder.atomfoundry.dev), as a permanent controlled experiment. It exists so Atom Foundry can study AI commerce causally rather than only observationally.

Because Atom Foundry controls every variable on this store, it can change one thing at a time and measure how AI systems respond in real time. What is tracked on Founder Lab includes:
- How each change moves the store's AI Commerce Score (the 8-factor readiness score).
- Which AI agents and crawlers visit the store, how often, and which pages they read (retrieval vs training vs search behavior).
- Whether the store is retrieved and surfaced in live AI shopping answers.
- How the store's recommendation profile shifts over time (Recommendation Momentum).

Founder Lab is the causal ground truth behind Atom Foundry's research: the place where a hypothesis is confirmed on a fully controlled store before it is claimed publicly. It is a research instrument, not a commercial storefront.

- [Founder Lab hub](https://atomfoundry.dev/founder-lab): overview of the experiment and its purpose.
- [Founder Lab, Day Zero](https://atomfoundry.dev/research/founder-lab/day-zero): the store's real baseline before any change was made, its starting AI Commerce Score, a crawlability check, and zero recorded AI bot visits.
- [Founder Lab Log](https://atomfoundry.dev/research/founder-lab/log): a running, dated log of every change made to Founder Lab and what the next rescan actually showed, including how a bug in our own AI-visit tracker was found and fixed.

## About
- Built and operated by Daniel Pokorny, a solo founder.
- [About Atom Foundry](https://atomfoundry.dev/about): why the project exists.
- Contact and social links are in the site footer.
- Legal: [Terms of Service](https://atomfoundry.dev/terms-of-service), [Privacy Policy](https://atomfoundry.dev/privacy-policy).
- Trademarks (AI Commerce Score™, Recommendation Share™, and similar) mark concepts Atom Foundry actively measures, not aspirational claims.