Student Summer Break Halves ChatGPT Usage
Educators across the spectrum, from K-12 to university professors, have increasingly voiced concerns over students’ growing reliance on large language model (LLM) chatbots like ChatGPT for academic tasks. This trend, many argue, contributes to a noticeable decline in student literacy and reading comprehension, with some reports indicating reading abilities among US children have hit historic lows. Yet, new data suggests that this dependency isn’t one-sided; it appears AI companies themselves rely heavily on students to constitute a significant portion of their user base.
Recent data released by AI platform OpenRouter, which facilitates interactions with various AI models, reveals a pronounced drop-off in ChatGPT queries from late May, when the school year is typically in full swing, to early June, coinciding with the start of summer break. While OpenRouter’s data, drawn from its 2.5 million users, doesn’t encompass every ChatGPT user, it offers a compelling snapshot of broader trends across AI models by collecting anonymous usage statistics from approved LLMs.
The daily statistics paint a clear picture: ChatGPT usage peaked on May 27, during what is often finals season, with users generating an astonishing 97.4 billion tokens in a single day. A token is a unit of AI data roughly equivalent to four English characters. Throughout May, ChatGPT users averaged 79.6 billion tokens per day. This figure sharply contrasts with the average daily usage in June, when schools typically let out, which plummeted to 36.7 billion tokens. OpenRouter’s graphical representation of this data strikingly illustrates the correlation. Interestingly, even during the school year, usage dips were observed to align precisely with weekends.
These findings are not isolated. The observed summer decline in ChatGPT usage echoes previous observations. In 2023, Business Insider first theorized that summer vacation was leading to a decrease in ChatGPT queries, a theory seemingly confirmed months later when Bloomberg reported a resurgence in traffic as schools resumed in mid-September. More recently, an in-depth study by Rutgers scholars analyzing 10,000 ChatGPT prompts found a strong correlation between student interactions with the bot and their academic calendar. Spring break and summer consistently emerged as “dead zones,” leading the research team to conclude that “most usage was academic.”
In a surprising turn for the AI industry, this seasonal dip in users may actually benefit OpenAI, at least in the short term. An analysis of OpenAI’s 2024 finances revealed a significant burn rate: the company spent $9 billion to generate $4 billion in revenue, with the vast majority of these costs attributed to processing tokens. This translates to OpenAI spending $2.25 for every dollar earned, explaining why the company does not anticipate achieving positive cash flow until 2029. The timing of its recent decision to launch GPT-5, therefore, might be strategic, as the inevitable surge in usage from a new model release won’t coincide with the extraordinary demands of a full school year.
The deep integration of AI into academics is no mere coincidence. Reports from July indicated that leading LLM developers, including Microsoft, OpenAI, and Anthropic, had contributed approximately $23 million to the American Federation of Teachers as part of a “partnership” aimed at introducing AI into classrooms. This financial commitment stands in stark contrast to the Trump administration’s decision to withhold over $6 billion in federal education grants earmarked for struggling schools across the US, even as it previously awarded OpenAI a $500 billion contract.
As students prepare to return to classrooms in September, the coming months will offer a crucial period to observe how usage data evolves, further solidifying the chatbot’s role as a seemingly permanent fixture in modern education.