The esteemed startup accelerator Y Combinator (YC), a crucible for some of the tech world’s most transformative companies including Reddit, DoorDash, and Airbnb, has observed a significant and concerning trend: an increasing reliance on artificial intelligence (AI) by founders to craft their application essays. This shift, noted by YC leaders, suggests a potential departure from genuine founder-led narratives and raises questions about the authenticity and originality of early-stage startup ideas being presented to the prestigious program. Since its inception in 2005, YC has invested in over 5,000 startups, playing a pivotal role in shaping the modern entrepreneurial landscape. However, the recent influx of applications exhibiting tell-tale signs of AI generation is prompting introspection within the accelerator’s ranks.

The YC application process, designed to elicit a deep understanding of a founder’s vision, requires detailed explanations of how an idea was conceived, a thorough analysis of the competitive landscape, and a clear outline of revenue generation strategies. It is precisely within these critical sections that YC partners have identified patterns indicative of AI assistance. These observations are not mere anecdotal impressions; they are supported by data and linguistic analysis, pointing towards a widespread adoption of AI tools among aspiring entrepreneurs seeking YC’s coveted backing.

One of the most immediate indicators of AI involvement, as highlighted by YC partner Tyler Bosmeny, is a discernible increase in the sheer length of application responses. Bosmeny, in a recent post on the social media platform X (formerly Twitter), expressed his surprise, stating, "I wonder what could explain that…". He noted that over the past three years, YC applications have seen a remarkable 60% surge in length. This statistical jump, when considered against the backdrop of an evolving application process and the rapid advancement of AI language models, suggests that AI tools, capable of generating verbose and comprehensive answers, are being widely utilized. The ability of AI to produce lengthy, well-structured text, even if lacking genuine personal insight, can be a powerful lure for founders under pressure to present a compelling case.

Further linguistic analysis by YC partner Pete Koomen has provided even more granular evidence of AI’s influence. Koomen, who dedicated a week to meticulously reviewing applications, observed a distinct shift in the language and style employed by founders. In a series of posts on X, he remarked, "back in my day (one year ago) founders didn’t talk like this." His observations are corroborated by data illustrating a dramatic rise in the usage of specific words and grammatical structures commonly associated with AI-generated content.

A particularly striking example cited by Koomen is the word "wedge." His research, presented in a chart, revealed that the frequency of the word "wedge" in YC applications leaped from less than 1% in the Spring 2025 application cycle to over 20% by Summer 2026. While the word "wedge" itself is not inherently artificial, its sudden and widespread adoption within application narratives is a strong signal. In business contexts, "wedge" often refers to a strategy to capture a specific market segment or to differentiate a product. The AI’s ability to identify and deploy such business jargon in a seemingly strategic manner, but without necessarily reflecting a deep, founder-driven understanding of market dynamics, is a cause for concern.

Adding another layer to this linguistic forensic analysis, Koomen pointed to the increased prevalence of em dashes. He noted that their use began to climb significantly in 2023 and "really hit their stride in ’24." Data shared by Koomen indicated that by Summer 2025, more than 50% of YC applications incorporated em dashes. While em dashes are a standard punctuation mark, their overuse, particularly with spaces surrounding them, has become a recognized characteristic of certain AI writing styles. Koomen’s detailed analysis further connected the spike in spaced em dashes and the popularity of the word "wedge" to the release and widespread adoption of advanced AI models, such as Anthropic’s Claude Opus 4. This temporal correlation suggests a direct influence of these AI tools on the writing patterns observed in the applications.

The phenomenon of AI influencing writing styles is not new. YC co-founder Paul Graham has also weighed in on this evolving landscape. In an X post in April 2024, Graham shared an anecdote about receiving a cold email suggesting a "novel project" that employed the word "delve." He openly expressed his dislike for the word, but more importantly, identified it as a "sign that text was written by ChatGPT." Graham’s observation, which garnered millions of views, highlighted a broader trend he had been noticing.

To support his point, Graham shared a chart compiled by researcher Philip Shapira. This chart illustrated a dramatic increase in the usage of the word "delve" in published papers and articles, skyrocketing from near zero to approximately 18,000 instances between 1990 and 2024. Graham argued that "delve" is a word rarely used in spoken English and is more commonly found in written contexts where individuals aim to sound more sophisticated or academic. He contrasted this with the natural language typically employed by founders, suggesting that the inclusion of such words in application essays signals an artificial attempt to elevate the prose.

Interestingly, the very founders who might have been using "delve" to sound more impressive appear to have shifted their AI-assisted writing strategies. Koomen’s data indicated that the word "delve" saw a spike in only 1.25% of applications in the winter of 2024, prior to Graham’s widely publicized post. This suggests that as AI detection methods become more sophisticated and as founders become more aware of common AI tells, their AI prompts and outputs evolve, leading to new patterns of language usage.

More recently, in a May post on X, Graham noted a further evolution in AI-generated content, observing that many founder emails now adopt "a hard-hitting journalistic style." He expressed his clear recognition of these emails as AI-generated, stating, "I know they’re written by AI, because no founder ever wrote this way before." Graham’s sentiment of feeling "lied to" encapsulates a core concern for accelerators like YC: the potential erosion of authenticity and the genuine passion that drives entrepreneurial endeavors. The application process is not merely a hurdle; it’s an initial window into the founder’s mindset, their ability to articulate a vision, and their commitment to the problem they are trying to solve. When this window is obscured by AI-generated prose, it becomes difficult to assess the true potential of a startup and its leadership.

The implications of this trend extend beyond YC’s application review process. For accelerators and venture capitalists, the ability to discern genuine founder insight from AI-generated content is becoming increasingly crucial for effective due diligence. If AI can flawlessly mimic the language of compelling startup pitches, it could lead to misallocation of resources, backing ventures that lack the fundamental human drive and innovative spark required for long-term success.

Furthermore, this reliance on AI for application writing might inadvertently stifle genuine innovation. Founders who heavily depend on AI to articulate their ideas may not have fully internalized or deeply thought through the nuances of their market, their product, or their business model. The process of grappling with these challenges, articulating them in one’s own words, and defending them is an integral part of developing a robust startup. Outsourcing this cognitive heavy lifting to AI could mean that founders are not as prepared as they appear, potentially leading to future difficulties when faced with real-world business challenges.

The timeline of these observations is also noteworthy. The increased use of em dashes and the word "wedge" becoming more popular in applications coincided with significant advancements in large language models, such as the release of Anthropic’s Opus 4 model in late 2025 or early 2026. Paul Graham’s observation about the "hard-hitting journalistic style" in May 2024 suggests a continuous and rapid evolution in AI’s capabilities and its adoption by users. This rapid iteration means that detection methods must constantly adapt.

In response to these developments, YC and other similar organizations may need to explore new methods of evaluating applications. This could involve more in-depth interviews, live coding challenges, or even requiring founders to articulate their ideas in less conventional ways. The ultimate goal remains to identify founders with the vision, resilience, and genuine entrepreneurial spirit necessary to build groundbreaking companies.

The current situation presents a complex challenge for the startup ecosystem. While AI offers powerful tools for enhancing productivity and creativity, its misuse in core aspects of the entrepreneurial journey, such as articulating one’s foundational vision, could have unintended consequences. YC’s candid observations serve as a crucial signal, prompting a broader conversation about the ethical and practical implications of AI in the early stages of startup formation and the vital importance of preserving authenticity in the pursuit of innovation. The future of how startups present themselves and how they are evaluated will undoubtedly be shaped by this ongoing interplay between human ingenuity and artificial intelligence.