Prompted, Week Two: Nike, Just Name It

Welcome to Prompted, the weekly brand audit column on The Canonical. Each week we take one company, run it through a live, multi-model AI visibility test, and show where what AI says about a brand differs from what the brand says about itself. Week two: Nike.

All figures were captured on 20 August 2026 and represent a snapshot in time.

Overview

Nike appears in 99% of AI answers when a prompt explicitly says “Nike.” Remove the brand name and ask the same customer question, and that figure falls to 53%. On some of the most common questions around sizing, durability and customer service, Nike is not mentioned at all.

The finding is simple: AI is not forgetting Nike. It is reluctant to volunteer it.

Ask an AI engine about Nike and it reliably answers about Nike. Ask the questions real customers ask before they have chosen a brand like what shoe fits flat feet, why a delivery is delayed, how long a midsole should last, and Nike disappears from nearly half of the answers.

What We Ran

The test used 40 prompts split evenly into two groups. Twenty explicitly named Nike. Twenty described the same customer or partner problems without naming any brand. The prompts were run across Claude Sonnet 5, GPT-5.6-luna, Gemini 3.6-flash and Grok 4.3 with web search enabled, producing 160 answers and 1,113 citations.

The split matters because it isolates the only variable that truly determines AI visibility: does the brand surface when the customer has not already supplied the name?

When Nike was named, it appeared in 79 of 80 answers. When it was not, the brand appeared in only 42 of 80. Twenty-three prompts produced a clean sweep across all four engines. Five produced no mention of Nike whatsoever.

The Visibility Gap

All four engines performed similarly on branded prompts. The differences emerged only when the brand name disappeared. Grok and Claude achieved the strongest unbranded recall, while Gemini performed worst. ChatGPT sat between them, citing fewer sources overall but relying more heavily on Nike’s own domains.

One example captures the broader pattern. A prompt asking whether Vaporfly and Alphafly “super shoes” genuinely make runners faster generated a detailed answer discussing energy return, marathon performance and the famous 4% efficiency claim. Yet one engine never mentioned Nike by name. The product received recognition; the brand did not.

That dynamic appears repeatedly throughout the audit. Nike’s innovations are often well known enough to be discussed independently of the company behind them. From a visibility perspective, that creates a paradox: brand familiarity can reduce brand attribution.

Where Nike Goes Silent

The most significant finding is not where Nike appears, but where it disappears.

Five prompts generated zero mentions across all four engines. These were not obscure edge cases. They were ordinary customer questions about sizing, fit, durability, injury prevention and service-related issues.

Examples included:

• How to measure foot size for running shoes at home.
• A sportswear order that was charged but never placed.
• Expected mileage before a midsole wears out.
• Whether stability shoes are still needed after shin splints improve.
• Recommendations for wide toe-box trainers.

Nike already has content, products or services relevant to each of these topics. Yet none of that material surfaced in AI-generated answers. Two patterns explain most of the gap.

First, complaint and service-intent prompts scored zero. Durability concerns, payment issues and service frustrations routed users toward third-party sources rather than Nike. Second, prompts written in the style of real forum posts performed particularly poorly. The more a question sounded like an actual customer complaint, the less likely AI systems were to volunteer Nike as an example.

There was one encouraging signal. Prompts with explicit India context produced significantly stronger recall than globally phrased equivalents. Nike appears to have stronger visibility in localised contexts than in generic category discussions.

Who Fills the Space Instead?

Nike remains the most frequently mentioned brand in unbranded answers, appearing in 53% of them. But the gap is narrower than many would expect. Adidas appears in 44% of answers, Puma in 38%, while Asics, Brooks, Decathlon, Hoka and Under Armour also secure meaningful visibility.

In several cases competitors were recommended while Nike was omitted entirely. The practical implication is that AI discovery is no longer winner-takes-all. When customers ask category questions rather than brand questions, competitors gain an opportunity to enter the conversation.

The Citation Story

The sourcing data reinforces the same conclusion.

When Nike was explicitly named, 12.7% of citations came from Nike-owned properties such as nike.com, nike.in and about.nike.com. When the brand was not named, that figure collapsed to just 1.1%.

Instead, AI engines relied on review sites, aggregators, forums and independent publications. These sources and not Nike’s own website formed the foundation of category-level recommendations and discovery.

The mechanism is straightforward. Nike’s website helps answer questions after the brand has already been identified. It rarely helps AI systems identify Nike in the first place.

The mechanism is straightforward. Nike’s website helps answer questions after the brand has already been identified. It rarely helps AI systems identify Nike in the first place.

What This Points At

The data supports three priorities.

First, own the fit-and-wear questions. Sizing, break-in periods, expected mileage and shoe replacement intervals account for several of the audit’s zero-mention prompts. These are straightforward informational gaps that can be addressed without turning every answer into a product pitch.

Second, invest beyond owned media. The review layer and not the corporate website, is where AI systems discover brands during category searches. Visibility within reviews, comparisons and trusted third-party sources matters more than publishing additional content on nike.com alone.

Third, strengthen India-specific visibility. Localised prompts already perform better than globally neutral ones. That advantage exists today, but it remains contested territory.

The Pattern Behind the Pattern

Brands do not lose AI visibility because they are forgotten. They lose it because they become so closely associated with an answer that the answer no longer needs to name them.

Nike’s products are frequently recognised even when Nike itself is not. That is a testament to the strength of the brand. It is also a reminder that AI systems reward the sources that explain, review and contextualise products, not necessarily the companies that created them.

Nike can absorb that gap because its underlying brand equity is enormous. Most brands are not that fortunate. The lesson is simple: if you want AI to discover you, you cannot rely on being known. You have to give the machine a reason to name you.

Sources

• Nike (nike.com): source of the Vaporfly and Alphafly “4% efficiency” claim referenced above. https://www.nike.in/
• AI Visibility Index (PingAura), Sportswear category, updated weekly: independent public benchmark scoring Nike first in Indian sportswear AI visibility (25.4), ahead of Adidas (21.6) and Puma (17.6) — third-party corroboration of our own finding that Nike dominates unbranded category prompts. llmvisibilityindex.com/industry/sportswear/
• Trivedi, Y., LinkedIn, Aug 2025, "Nike vs Adidas in India": cites Nike India revenue stagnating at ₹829 crore against Adidas' ₹2,578 crore in FY23, plus store count falling from \~350 to \~150 — backs the flag on the 14–15% market share figure in the meta description. linkedin.com/posts/yuvraj-trivedi-640109247_nikevsadidas-indiamarket-brandstrategy-activity-7365799905711316992-ZLkR

Methodology: 40 prompts synthesised from real user questions and intent clusters, split evenly between explicit (brand-named) and implicit (category-only) sets, run across Claude, ChatGPT, Gemini and Grok for 160 responses in total. All data captured 20 August 2026; model outputs are non-deterministic and subject to change.