Key Findings at a Glance

1. Remarkably Consistent Agreement – Both models agreed 75-76% of the time, whether judging well-known or niche brands.
2. GPT Relies Heavily on Earned Media – Over 90% of ChatGPT’s citations came from independent, third-party sources.

Gemini Leans on Brand & Social Content – Google’s model cited brand-owned sources 2-4x more than GPT, along with a notable reliance on social media.

In a large-scale experiment, thousands of brand comparisons were posed to two leading AI models—OpenAI’s ChatGPT and Google’s Gemini. The goal was simple: when asked "Which brand is better, A or B?", how do these models justify their answers?

The brands were divided into two categories:

• Well-known brands (easily recognizable, mainstream)
• Niche brands (lesser-known, specialized)

Each model was instructed to support its choice with citations, which were then classified into three source types:

1. Earned media (news articles, reviews, third-party coverage)
2. Brand-owned (company websites, official blogs)
3. Social (user-generated content, forums, social media)

The results reveal fascinating patterns in how AI models form opinions—and where they place their trust.

1. AI Models Agree—But Their Trusted Sources Differ

Despite different architectures and training data, GPT and Gemini reached the same conclusion three out of four times. This suggests a strong consensus in how AI evaluates brand strength, regardless of popularity.

However, their reasoning diverges sharply:

• ChatGPT overwhelmingly prefers earned media (93-95% of citations), indicating a bias toward journalistic or expert opinions.
• Gemini uses more brand-owned content (21-27%), implying a possible influence from Google’s search rankings, where SEO-optimized brand pages dominate.

2. Niche vs. Well-Known: Surprisingly Little Difference

One might expect AI models to struggle more with obscure brands, but the agreement rate barely changed (75.93% for well-known vs. 75% for niche).

Yet, the citation patterns held:

• GPT remained rigidly dependent on earned media, even for niche brands.
• Gemini cited slightly more social sources for niche brands (12.7% vs. 10.6%), possibly due to fewer authoritative links in search results.

3. The Google Factor: Is SEO Shaping Gemini’s Bias?

The most striking finding is Gemini’s higher reliance on brand-owned content—a potential side effect of Google’s search ecosystem, where companies aggressively optimize for visibility.

• Brands invest heavily in SEO, making their pages rank high in Google searches—which may explain why Gemini cites them more.
• GPT, trained on broader data, seems less influenced by commercial SEO tactics, favoring independent journalism instead.

This raises an important question: Are AI models simply mirroring the biases of their parent companies’ ecosystems?

This experiment reveals that while AI models often agree on brand superiority, their reasoning reflects their origins:

• ChatGPT acts like a journalist, prioritizing third-party credibility.
• Gemini behaves like a search engine, leaning on brand and social signals.

For marketers and researchers, this highlights a critical insight: AI doesn’t just generate answers—it replicates the biases of its training data. Whether that’s a strength or a weakness depends on who’s asking—and why.

So next time an AI picks a "better" brand, ask yourself: Is it fact… or just algorithmic preference?