July 1, 2026
Your AI Chatbot Is Giving Everyone the Same Marketing Ideas — Here's What to Do About It
Ask ChatGPT, Claude, or Gemini for a random number between 1 and 10. You are almost certainly going to get 7. That is not a coincidence. It is a window into one of the most underappreciated problems i
Your AI Chatbot Is Giving Everyone the Same Marketing Ideas — Here's What to Do About It
Ask ChatGPT, Claude, or Gemini for a random number between 1 and 10. You are almost certainly going to get 7. That is not a coincidence. It is a window into one of the most underappreciated problems in AI-powered marketing today: the chatbots your competitors are using are serving up the same ideas, the same taglines, and the same strategies to everyone who asks.
According to a July 2026 MIT Technology Review article by Will Douglas Heaven, large language models are far more predictable and far less creative than most people realize. A November 2025 research paper titled "Artificial Hivemind: The Open-Ended Homogeneity of Language Models (and Beyond)" — which won the best paper award at NeurIPS, one of the most prestigious AI conferences in the world — studied 25 different LLMs and had each one respond 50 times to the prompt asking for a metaphor about time. Out of 1,250 total responses, the overwhelming majority were variations of either "Time is a river" or "Time is a weaver." The researchers concluded that different LLMs converge on very similar answers when given open-ended questions, likely because most models today are trained in similar ways on similar data to accomplish similar tasks.
This convergence shows up in marketing in concrete, painful ways. When MIT Technology Review's Will Douglas Heaven asked multiple leading chatbots to generate a tagline for a New Balance running shoe campaign, both Claude and ChatGPT independently produced the exact same tagline: "Run your way." When researchers asked models to suggest band names, most outputs contained the words "glass," "neon," "velvet," or "static." Ask any of the major chatbots to recommend a car type and the answer is almost always a Toyota or a Honda. Australian startup Springboards, which built an LLM called Flint specifically to address this problem, demonstrated these patterns side by side. Springboards built Flint on top of Qwen 3, an open-source model from Alibaba, and trained it to identify the specific points in its output where more variety is possible, then insert less predictable options at those precise moments rather than simply turning up overall randomness, which can make models incoherent or cause them to switch languages mid-sentence.
For small and mid-size business owners using AI tools to generate marketing copy, social media content, product names, or campaign concepts, this research carries a direct implication: if your AI output looks polished and professional, there is a strong chance your competitors' AI output looks nearly identical. Zoe Scaman, founder of business strategy startup Bodacious and chief strategy officer at 77X, tested Flint against Claude, Gemini, and ChatGPT using a classic business case study. The three mainstream models all suggested essentially the same approach to the problem. Flint went in a different direction entirely. Scaman described the mainstream models' outputs as "nothing new" and called Flint's divergent framing "really interesting." That gap between expected and unexpected is exactly where differentiated marketing lives.
Maximilian Weigl, cofounder and chief strategy officer at the marketing firm Uncommon, put it plainly: "You can't really create something boundary-breaking with tools that pull you back to the average." His team uses Flint alongside ChatGPT, Claude, and Gemini as a way of breaking out of the gravitational pull of consensus AI output. At the same time, Weigl is honest about the limits. Nine times out of 10, average is fine. Not every piece of content needs to be boundary-breaking. But when you are developing a brand voice, naming a product, writing a headline meant to stop the scroll, or positioning your business against competitors, average is not just fine — average is invisible.
The practical upshot for your business this week is this: stop treating any single AI tool as your only creative source. The homogeneity problem is real, documented, and built into how these models are trained. Weigl's own team runs multiple models in parallel and uses the divergent outputs as creative raw material rather than finished copy. You do not need access to Flint to do this. Run your next marketing prompt through ChatGPT, Claude, and Gemini at the same time. Force yourself to look at all three responses before choosing. Then edit from your own voice and real knowledge of your customers, not from whichever AI answer felt most fluent. As Weigl said directly to his own team: "Think, talk to other people, use your own voice."
The businesses that win with AI marketing are not the ones using it most — they are the ones using it most strategically, treating AI output as a starting point rather than a destination.
Originally inspired by: LLMs are stuck in a groupthink groove. This startup is trying to get them out. (https://www.technologyreview.com/2026/07/01/1140003/llms-are-stuck-in-a-groupthink-rut-this-startup-is-trying-to-get-them-out/) See how Leads to Conversion can help you break out of the AI sameness trap and build marketing that actually stands out. Get your free AI audit