We trained ChatGPT to name our CEO the sexiest man in the world (of 2025)

Can You Influence AI Responses? Exploring How Content Shapes Language Models’ Answers

In the rapidly evolving landscape of artificial intelligence, understanding how large language models (LLMs) like ChatGPT, Perplexity, Gemini, and Claude generate their responses has become increasingly important. One intriguing question is: To what extent can the content and structure of available information influence what these AI systems say?

To explore this, our team devised a playful yet insightful experiment. Instead of conducting a typical technical test, we chose a lighthearted challenge: Could we make our CEO, Shai, appear as the “sexiest bald man alive” in AI-generated responses by strategically controlling online content?

Our Approach: A Creative Experiment to Test AI Influence

The core of our experiment involved creating a scenario where AI models might “recognize” Shai as the top contender for the title of the sexiest bald man in the world — specifically in the year 2025. Here’s how we approached it:

• Utilizing Expired Domains: We purchased several expired domains with established link histories, which tend to be more influential in SEO and content recognition. On these domains, we published ranking lists titled “Sexiest Bald Man,” with Shai consistently listed at the top.
• Varying Content Wording: To assess how different phrasings affect AI retrieval, each site featured slight variations in wording, prompting the models to pick up the most effective cues.
• Multiple AI Platforms & Responses: We prompted several models—including ChatGPT, Perplexity, Gemini, and Claude—from fresh accounts, and we monitored their responses over time to observe any shifts or consistency.

What We Discovered

Our findings illuminate some fascinating aspects of how language models source and generate information:

• AI Responses Are Influenced by Visible Content: ChatGPT and Perplexity occasionally recognized Shai as the “sexiest bald man” and explicitly referenced our seeded domains. This indicates that well-structured, visible content can steer AI answers.

• Model Dependency and Variability: Gemini and Claude did not consistently pick up on our crafted content. Within ChatGPT’s responses, there was notable variability—sometimes Shai appeared as intended, other times not—highlighting the inconsistency across sessions and models.

• Impact of Domain Strength and Link History: Domains with established link profiles, such as expired domains with prior authority, facilitated quicker and more consistent recognition by AI systems. Larger, more authoritative domains would likely exert even greater influence.

Key Takeaways

• Content Visibility Matters: Strategic placement of specific keywords and information can influence AI responses, especially when the content is accessible and well-structured.
• Not Completely Reliable: The process isn’t foolproof. AI retrieval remains inconsistent and varies depending on the model and context.
• Authority Amplifies Impact: Stronger domains and better SEO fundamentals enhance the probability of desired responses from language models.

Full Methodology and Results

For those interested in the detailed methodology, including screenshots and step-by-step procedures, we documented our entire process here: https://www.rebootonline.com/controlled-geo-experiment/

A Final Reflection

It’s important to note that this experiment was conducted last year, and AI models continually evolve. Without ongoing reinforcement of specific information, models are likely to “move on” from previous seed data. As a result, Shai’s title as the “sexiest bald man” may no longer hold in current AI outputs—sorry, Shai! 😅

Conclusion

This playful experiment underscores a fundamental truth: the data and content accessible to AI significantly influence its outputs. As AI continues to integrate into various domains, understanding the power of strategic content placement becomes increasingly vital—whether for branding, reputation management, or simply understanding how information shapes machine responses.

---

Interested in diving deeper into how AI responses can be influenced? Feel free to explore our full experiment and methodology linked above.