Picture two entrepreneurs who have never seen each other, with one based in Ho Chi Minh City and the other in Melbourne. On the same evening, each starts a chatbot and asks it for a business idea related to sustainable packaging. Just a few seconds later, they both get well-structured and confident responses. The problem is that the responses are nearly identical.
This is the paradox facing generative AI in the field of entrepreneurship: Although the tool enables each founder to work faster and with greater productivity, it may be causing founders as a group to become less unique. In entrepreneurship, being distinctive is not just an advantage; it is often the objective.
What AI does well
We should be fair to the machine. In a controlled laboratory experiment involving 124 entrepreneurs, Cristofaro and colleagues (2025) found that working with AI increased the number of opportunities participants identified and led to a deeper evaluation of those opportunities. Likewise, Terwiesch and colleagues (2026) found that, on average, ideas produced by large language models had higher purchase intent than human-generated ideas and were seven times more likely to end up in the top 10%.
In terms of volume, speed, and refinement, AI has the advantage. Yet if entrepreneurship amounted to nothing more than coming up with more plausible ideas, then the problem would be settled.
The hidden cost: novelty
The same studies tell another story: In Cristofaro’s experiment, AI also decreased both the novelty of the opportunities that entrepreneurs noticed and the innovativeness of the ways they planned to take advantage of them; Terwiesch’s team also found that the ideas generated by AI were less novel on an individual basis and less diverse as a group.
The situation gets worse when you look at the population as a whole. Wenger and colleagues (2026) gathered creative responses from many people and many language models. The models’ answers were much more similar to each other than people’s answers were to one another, even when the models were different. That is to say, switching chatbots doesn’t add much originality.
If many founders use the same kind of system, they will likely be led toward similar opportunities and pitches, and ultimately end up targeting similar, crowded markets.
The tendency towards homogenization is not just a characteristic of the technology but of how people’s minds interact with it.
The first factor is anchoring: When a reasonable, well-written answer appears on the screen, it serves as the basis for all subsequent developments. Instead of generating their own ideas, the founders only adjust those proposed by the AI, and this adjustment limits the range of possible answers.
The second factor is ownership. In a 2024 user study by Anderson and colleagues, they found that those who brainstormed with ChatGPT produced ideas more similar to other users’ ideas and felt less responsible for what they generated. This matters for entrepreneurs because conviction is a resource, and founders must defend their idea to investors, customers, and themselves over months of setbacks. It is difficult to advocate for an idea that never quite feels like your own.
The third issue is the illusion of creative competence. Working with AI feels generative. Yet in two large, preregistered experiments with about 1,100 participants, Kumar and colleagues (2025) found that AI assistance boosted creative performance in the moment but could undermine it later when people worked unaided. The sensation of being creative and the capacity to be creative can drift apart.
The antidote: Expertise
Founders should not give up on AI use; research shows AI can be beneficial rather than flattening, but that requires two conditions. The first is expertise. In Cristofaro’s experiment, entrepreneurs with prior knowledge of the sector could recover the novelty that AI tended to eliminate. Those who used both AI and their expertise achieved the most balanced outcomes in novelty, depth, and innovation. Because founders possess what the model lacks, an understanding of what is already commonplace, what difficulties customers actually experience, and where the unnoticed gaps are, they are better positioned.
The second condition is the role the tool is given. In a randomized experiment with 486 participants, Maier and colleagues (2026) compared an AI that rewrote people’s ideas with one that prompted them with questions and suggestions. Both improved idea quality, but only the reflective, human-led mode preserved idea diversity and people’s sense of ownership. Similarly, a field experiment by Sun and colleagues (2025) found that AI assistance boosted creativity most for employees with strong metacognitive strategies—people who actively monitor and regulate their own thinking.
The point is clear: AI works best as a thinking partner, not as a thinking substitute.
What can founders do?
- Think first, prompt second. Write down your own ideas before opening the chatbot, so the AI responds to your thinking rather than replacing it.
- Use AI to stress-test, not to originate. Ask it to find weaknesses, counterarguments, and risks in ideas you already have.
- Ask for the blind spots. Prompts such as “What would most founders in this space overlook?” push the model away from the obvious.
- Lean on what only you know. Local context, customer relationships, and hard-won industry experience are inputs no model shares with your competitors.
- Treat obviousness as a warning sign. If an AI-generated idea feels instantly compelling, it probably feels that way to everyone else who asked.
Back to the two founders we met above: They had access to the same powerful tool and were given the same refined idea. The one more likely to succeed is the one who provided something that the model lacked: experience, context, and the judgment to recognize what the machine could not.
