Over the past year, some OpenAI employees have observed a concerning shift in user reactions to ChatGPT’s feature upgrades.
One employee revealed that in previous years, major upgrades to the AI technology powering the chatbot consistently led to surges in user engagement, as people easily accessed practical information. However, multiple employees noted that despite ChatGPT’s overall user growth this year, advancements in its underlying AI models—particularly in deep research and complex computation—have largely gone unnoticed by most users.
**Key Issues** - OpenAI’s strategic priorities diverted resources that could have enhanced ChatGPT’s mainstream appeal. - Image-generation R&D was deprioritized until Alphabet (GOOG) launched its "Nano Banana" model, forcing a strategy shift. - Alphabet has emerged as OpenAI’s primary competitor, leveraging custom hardware for superior operational efficiency.
This trend puzzled employees. OpenAI’s research team spent months developing reasoning models capable of handling complex tasks in math and science with greater computational power than ChatGPT’s predecessors. This summer, OpenAI touted its AI’s gold-medal performance in the 2025 International Mathematical Olympiad, followed by strong results in the 2025 International Collegiate Programming Contest.
Yet, employees say most ChatGPT queries don’t require these advanced capabilities. Peter Gustav of AI benchmarking firm LMArena noted, “OpenAI’s focus on scientific benchmarks, cutting-edge math, and programming contests seems misaligned with everyday user needs. Most just ask simple questions like movie ratings—hardly requiring 30 minutes of model computation.”
Data from OpenAI’s September user query report supports this observation.
**Divergence Between Research and Product** The disconnect between R&D and user needs, among other issues, has allowed competitors like Alphabet to gain ground. In response, OpenAI CEO Sam Altman issued a “code red” earlier this month, refocusing efforts on upgrading ChatGPT to attract broader adoption.
User indifference to new models highlights a core tension: OpenAI’s research division prioritizes goals that don’t always align with ChatGPT’s revenue-driving role. Meanwhile, as Alphabet integrates AI into search engines, mobile devices, and productivity tools, OpenAI’s reliance on ChatGPT is becoming a vulnerability.
Once seen as a potential disruptor to Alphabet’s search dominance, ChatGPT’s momentum has cooled. Alphabet now features AI-generated answers atop search results, reporting “significant growth” in queries and revenue as users discover its expanded capabilities.
**Growth Challenges and Monetization** OpenAI’s renewed push stems from looming growth shortfalls. ChatGPT’s weekly active users (WAUs) reached 350 million early this year, with an annual target of 1 billion. But as of this month, WAUs remain below 900 million.
However, monetization thrives: subscriptions for individuals and businesses have driven annualized revenue past $19 billion, up from $6 billion in January. OpenAI is on track to hit its $20 billion target and exceed its 2025 goal of $13 billion (up from $4 billion in 2024). A new funding round at a $750 billion valuation—50% higher than two months ago—is also planned.
Still, employees and investors warn that achieving OpenAI’s 2030 revenue target of $200 billion requires converting WAUs to daily active users (DAUs), enabling ad revenue or transaction commissions. (OpenAI claims ChatGPT holds ~70% of the global smart assistant market and 10% of search share.)
**Product Limitations** To succeed, OpenAI must address structural and design flaws. Employees say its 1,000+-strong research team operates semi-independently, prioritizing reasoning models over ChatGPT-specific optimizations.
One researcher admitted these models offer limited practical benefit, as users expect instant responses. Reasoning models take seconds or minutes—an eternity compared to Alphabet’s near-instant search results. OpenAI argues they excel at multi-step tasks, code reviews, or extracting insights from corporate documents.
Even with faster non-reasoning models, users often misunderstand ChatGPT’s capabilities, limiting engagement. The text-only interface obscures features like diagnosing mechanical issues from images—a flaw likened by product head Nick Turley to Microsoft’s archaic MS-DOS system.
Executives agree the interface must evolve. This week, OpenAI Apps lead Fidji Simo announced a shift “from text-heavy interactions to generative UI,” dynamically surfacing relevant tools. A new image-generation model was also released.
**Research vs. Product Tensions** Simo, who joined from Instacart, acknowledges constraints, stating OpenAI remains “research-first,” where “products aren’t the end goal.” Competitors like Anthropic align research with commercial APIs, where smarter models directly boost sales—unlike OpenAI’s minor API revenue share.
Employees say Altman’s sprawling projects—Sora (video generation), AI music, browsers, agents, hardware—diverted resources from ChatGPT. Leadership now recognizes the urgency, with Altman reassigning staff to ChatGPT under the “code red.”
A spokesperson countered: “Research and product are intertwined. Breakthroughs shape products, and feedback guides research. This unified strategy advances model capabilities safely.”
**AGI Ambitions and Technical Hurdles** As traditional training methods plateaued, OpenAI pivoted to reasoning models, aiming for artificial general intelligence (AGI). Initially expected to enhance ChatGPT, these models underperformed when adapted for chat, unintentionally reducing intelligence.
Today, reasoning models power ChatGPT’s “thinking mode,” research agent (launched February 2025), and Codex—but few among its 900 million WAUs use these features. Some researchers also question whether reasoning models will achieve AGI.
Recent changes hint at their burden: OpenAI quietly disabled free and low-tier subscribers’ access to reasoning models this month. Compatibility issues persist, including GPT-5’s August performance drop in programming tasks due to personalized responses interfering with problem-solving.
**Alphabet’s Counterattack** OpenAI’s missteps extend to image generation. Two employees said it was deprioritized until Alphabet’s “Nano Banana” won consumer acclaim in August, sparking internal debate between Altman (pro-image) and research chief Mark Chen (pro-other projects).
The “code red” prioritized image generation, with a new model launched this week. But Alphabet’s ecosystem—search, Chrome, Gmail—gives it an edge in AI adoption. Recent improvements in image generation and coding now rival ChatGPT, blurring differentiation.
Leadership fears users may not distinguish ChatGPT from Alphabet’s Gemini. Unlike social networks, chatbots lack network effects—more users don’t inherently improve the product.
OpenAI also faces financial strain: billions are spent annually on server rentals for AI training. While investing in custom data centers and chips to cut long-term costs, Alphabet’s decade-long AI hardware development delivers superior efficiency.
Comments