What Black Friday reveals about how LLMs understand ecommerce

Every Black Friday reveals how consumers search, compare, and decide. This year added something new: a real-world test of how AI models interpret commerce under true demand.

So we ran a structured test across major LLMs and analyzed 10,000 responses. The goal was simple: to see how these systems form their internal view of the retail landscape and which signals shape the answers they generate.

As we reviewed the dataset, a clear pattern emerged: Black Friday acts as a natural stress test for AI-driven discovery.

The sheer volume of queries, the range of categories, and the speed of shifting consumer attention expose the sources, structures, and behavioral tendencies that shape how LLMs reason about products, retailers, and intent.

The results offer a preview of how AI search is evolving – and how the broader commerce ecosystem will feel the impact.

TLDR; 

1. LLMs overwhelmingly rely on a small cluster of external domains with YouTube, big-box retailers, and U.S. review media dominating the landscape.
2. Generalist retailers win decisively, capturing nearly half of all retail mentions and becoming the “default funnel” LLMs use to answer shopping questions.
3. Social and UGC sources surge during Black Friday, growing +8.1%, while classic retail and media sites lose share.
4. Off-page signals matter as much as on-page signals: Reddit, YouTube, Amazon, and Consumer Reports collectively shape the “External Data Sources” LLMs use to compare and recommend products.
5. Structured comparison content is disproportionately influential, far more than brand-owned assets.
6. LLMs behave differently not only from Google, but from each other, with each Gemini, OpenAI, and Perplexity producing different formats, lengths, and reasoning patterns.

LLMs don’t look at the commerce ecosystem like search

In traditional search, the funnel starts with a query and ends with a ranked list of results, often dressed up with shopping carousels, popular products, and other curated touches. In AI search, the funnel flips.

The model begins with its internal map of the world – a compressed web of relationships, sources, and signals – and then builds an answer. In shopping, an LLM’s goal is to deliver a purposeful response, not a shopping experience.

When we reviewed the top 50 most-cited domains across 10,000 LLM responses – spanning deals, reviews, comparisons, and product recommendations – the distribution was far from neutral:

• YouTube: 1,509 citations
• Best Buy: 950
• Walmart: 885
• Target: 477
• TechRadar: 355
• RTings: 342
• Consumer Reports: 325

This cluster shapes much of the commercial “knowledge” LLMs draw from. It leans toward large retailers, widely cited media outlets, and platforms built around comparisons or reviews. Together, these sources create a collection of resources that lets models deliver direct answers across any vertical, product type, or consumer need.

How LLM behavior shifts before and during Black Friday

In our analysis of 10,000 responses, we compared the week leading up to Black Friday with the event itself. Before Black Friday, responses were anchored in planning behavior:

• Retail and brand domains: 59.6%
• Media: 23.4%
• Social and UGC: 17%

Users prepare by comparing, researching, and setting baselines – and LLMs mirror that behavior. Even prompts that included “Black Friday” tended to produce expectation-setting responses:

• “Isnt it too soon to start searching for black friday?”
• “Althought it is before black friday…”

When the event began, the mix shifted fast. Social and UGC content jumped to 25.1%, gaining more than eight points of share, while retail and media both edged down.

This shows a shift inside the models: as uncertainty rises and pricing and inventory move around, LLMs lean harder on human discussion and experiential content.

This pattern mirrors consumer behavior but also shows how heavily models rely on conversation-driven sources for real-time decision cues.

The weight of off-page content

One of the clearest insights from the dataset is the weight third-party domains have on AI reasoning. Today’s LLMs win by absorbing as much human interest in products as possible. The players that supply huge volumes of consumer insight, reviews, product demos, sentiment, and structured data end up shaping how models reason and decide.

In an Athena analysis of external influence in retail and ecommerce (October 2025), five domains appeared consistently as the dominant off-page signals LLMs rely on:

1. Reddit: 34%
2. YouTube: 19.5%
3. Amazon: 15.5%
4. Business Insider: 9.2%
5. Walmart: 8.9%

Each one shapes a different part of the model’s decision-making process. Across all of them, we see the same pattern: LLMs depend on content that captures real human interest, organizes consumer-driven options, and reduces uncertainty with verifiable data.

Today, LLMs are building a fortress of product data that will unlock the most powerful shopping-discovery tool consumers have ever used.

The role of brand-owned content

Although third-party domains dominated, brand websites still played a measurable role in the dataset. They create a crucial path forward for any consumer brand that wants to win in AI discovery.

A site’s internal structure plays a major role in how a model interprets a brand.

According to the Athena retail & ecommerce dataset:

• The homepage accounted for 40%
• Blog content accounted for 10.6%
• Product pages accounted for 10.5%

The homepage serves as the brand’s primary identity layer. It sets the tone, defines the positioning, and gives the model the simplest semantic signals to read.

Blogs and product pages play a different role. They provide definitional clarity, long-tail context, and the factual detail the model needs.

Brands that rely on promotional copy, unclear hierarchy, or thin product content leave major visibility on the table.

Today, LLMs use brand content to validate and deliver direct responses—but only when off-page content and data justify the brand’s place in the conversation.

Which retailers rise to the top

Across the entire dataset, a few categories dominated model responses.

Generalist retailers own the conversation with 48% share

Walmart, Target, and Best Buy capture nearly half of all retail citations. Their breadth, familiarity, and content depth put them at the center of LLM commerce reasoning.

Electronics specialists own 23% of the share

Best Buy leads by a wide margin, followed by Newegg and Micro Center. Tech-focused queries consistently push models toward these sources – though the surge in electronics during Black Friday likely amplifies this effect.

Other verticals remain far behind

Fashion, beauty, pharmacy, home, DIY, and pets each take smaller slices, even with strong category leaders in play. The imbalance reflects the sheer volume of content generalist retailers produce compared with niche verticals.

Different platforms, different behaviors

As we reviewed the platforms, another pattern stood out: major LLMs don’t just answer differently – they think differently. Each one has its own rhythm, preferred structures, and style of presenting commercial information.

Gemini produces the most expansive outputs. Its responses averaged 606 words, with 97.6% using lists and 92.3% using headings.

The model often delivers essay-length explanations, averaging nearly 28 list items per response. It treats Black Friday as if every query deserves a full article.

OpenAI sits in the middle. It averaged 401 words per response, with 99% including lists and nearly two-thirds using headings. Its lists were even denser, averaging 32 items.

Perplexity moves in a different direction. Its typical response was 288 words, with far fewer list items – about 9.7 on average – and fewer headings overall. It favors short, direct summaries. Even with complex topics, it compresses the information into something that reads like an executive brief.

These differences reveal distinct retrieval and reasoning strategies that shape how each model interprets brands, categories, and commercial intent.

As AI-driven discovery takes a larger role in search, teams will need to think about visibility in terms that respect each platform’s internal logic – not in broad strokes.

What are the implications for retailers and brands?

The data points to a clear direction: AI search is becoming its own ecosystem – shaped by familiar SEO inputs, source quality, content structure, and off-page signals, all interpreted by language models to deliver a clear response.

If your content isn’t clearly labeled, semantically structured, and reinforced across the web, it risks becoming invisible to AI systems surfacing answers or product suggestions.

In this new environment, retailers and brands must rethink how they communicate—not just on their own domains, but across the entire digital discovery surface.

On-page actions that matter

• Build semantically coherent homepages that reflect brand, product categories, and relevance to core queries. LLMs prefer clarity over cleverness.
• Strengthen product pages with structured, factual content, clear specifications, variant descriptors, and Q&A content that mirrors user research intent.
• Create educational content clusters tied to core product themes. These serve as reusable “content scaffolding” for AI models looking to contextualize a product.

Off-page actions that matter

• Foster review ecosystems and discussion forums (e.g., Reddit, Quora, third-party review sites). These validate trust signals LLMs associate with product quality.
• Ensure regular presence in comparison and recommendation-driven media (e.g., “best of” lists, product roundups, influencer explainers).
• Invest in rich media that features the value of products, especially YouTube and TikTok. Video content trains LLMs on product use cases, sentiment, and experiential value.
• If you participate in marketplaces, ensure product data is accurate and indexable. Structured product availability data from Amazon, Walmart, Etsy, and others is increasingly being ingested into AI discovery pipelines.

Why this matters now: The shopping research shift in ChatGPT

OpenAI’s recent Shopping Research announcement further raises the stakes. Through ChatGPT, OpenAI is now capturing real-time consumer research behavior – preferences for price, color, variants, availability, and more – to build what is essentially a user-trained targeting engine for commerce.

This isn’t just AI learning about your product. It’s AI learning how users shop.

For decades, retailers like Amazon, eBay, and Walmart have invested in complex taxonomies and refinement layers for discovery: variant mapping, filters, availability rules, and more. Now OpenAI is absorbing that logic not just by crawling, but by interacting with users and watching intent unfold.

For brands and retailers, this marks a shift from passive search optimization to active AI participation. If your content isn’t present, structured, or referenced in these systems, it won’t show up in the AI’s answers – or in the consumer’s journey.

The future of retail will be AI transactions

Black Friday gave us more than a look at which products sold best or which deals consumers chased. It revealed how LLMs behave under real-world demand—how they reason, reference, and prioritize across a fragmented content landscape.

The answers they generated were structured, confident, and increasingly influential, yet incomplete – shaped more by the sources they see most often than by the full depth of what brands offer.

What we’re witnessing isn’t just a new search interface. It’s the emergence of a new shopping architecture – one where agentic commerce replaces traditional browsing, and AI models, not consumers, drive product discovery, comparison, and even transaction.

OpenAI’s launch of Shopping Research makes this shift unmistakable. These models are no longer just language tools; they’re intent engines, trained not only on product data but on how people actually shop. Price sensitivity, variant preferences, real-time availability – all of it is now part of how AI interprets and responds to commercial intent.

For brands, the implications are significant. Visibility will no longer hinge on SEO rankings or ad placements alone. It will come from structured, semantically rich content, surfaced across the right off-page ecosystems, and aligned with the reasoning patterns of each major model.

We call this AI-native visibility – a discipline built to ensure brands aren’t just discoverable, but understood by the systems shaping modern commerce.

Black Friday was only the stress test. The real transformation is still ahead. And it won’t be won by who ranks, but by who is represented – accurately, contextually, and everywhere AI shows up.