EXECUTIVE SUMMARY
The ways in which the global workforce and everyday consumers interact with artificial intelligence are undergoing a fundamental transformation. Newly released data from OpenAI’s latest transparency initiative, Signals V3.0, reveals that information-seeking queries on consumer ChatGPT plans nearly tied with writing tasks in late 2024 and mid-2025, capturing roughly 20.7% of all user messages in August. This milestone marks a monumental shift in user behavior away from purely generative writing tasks—such as drafting emails, essays, and code—toward transactional inquiries, factual lookups, product research, and instructional guidance.
Published on October 26, the updated Signals dataset—alongside a complementary companion study examining how professionals integrate AI into their daily occupations—paints a nuanced picture of an increasingly mature user base. While writing once dominated consumer interactions by commanding over one-third of all conversational volume in mid-2024, its share has steadily eroded, ceding ground to practical guidance, informational lookups, and specialized work functions.
However, parsing these trends requires a rigorous analytical approach. Discrepancies between OpenAI’s public Signals datasets and independent academic literature—such as a collaborative 2025 paper authored by researchers from OpenAI and Harvard University—highlight the complexities of tracking millions of decentralized user interactions across global markets. As enterprises and individual practitioners increasingly lean on large language models (LLMs) for complex, multi-step problem-solving, understanding these underlying shifts is essential for product developers, market researchers, and business strategists alike.
1. DETAILED CHRONOLOGY: THE SHIFT FROM GENERATIVE WRITING TO INFORMATIONAL INQUIRIES
To comprehend the magnitude of OpenAI’s recent data releases, one must trace the longitudinal trajectory of user behavior captured across the platform’s consumer tiers. The timeline provided by Signals V3.0—which tracks a monthly sample of one million anonymized messages from account holders aged 18 and older across Free, Go, Plus, and Pro plans (excluding Enterprise and Codex users)—documents a steady erosion of text-generation tasks in favor of research-oriented behaviors.
The Rise and Fall of Writing and Information-Seeking
In July 2024, generative writing stood as the undisputed heavyweight king of ChatGPT use cases, accounting for a staggering 34.1% of all consumer messages. Users primarily leveraged the model to draft professional correspondence, construct long-form written content, and refine prose. By stark contrast, information-seeking—encompassing factual queries, product research, and cooking or recipe assistance—hovered at a modest 16.7%.
Over the ensuing months, a quiet revolution occurred within consumer interaction patterns:
- The Trajectory of Writing: The share of writing-related prompts began a prolonged, consistent descent. By the close of the period analyzed in September 2026, writing had dropped to 21.4% of total message volume—a precipitous decline of nearly 13 percentage points.
- The Asccent of Information Seeking: Conversely, user queries demanding factual data, service exploration, and guidance trended upward. By August, information-seeking reached 20.7%, nearly dead-locking with writing at 20.6%.
- The September Rebound: Although writing temporarily recaptured the lead in September at 21.4% versus 20.2% for information-seeking, the narrow margin signals a permanent structural shift in how consumers utilize generative tools. ChatGPT is no longer viewed merely as an automated typewriter or creative writing assistant; it is increasingly consulted as a primary research engine.
Practical Guidance Dominates the Landscape
Beyond raw information retrieval, practical guidance—a broad category that captures how-to advice, troubleshooting, and tutoring—has established absolute dominance over the ecosystem. According to the Signals V3.0 dataset, practical guidance has remained the single largest topic category every single month since March 2025, claiming an impressive 28.1% of total messages by September 2026.
This sustained preference for guidance over pure generation or open-ended chat underscores a utility-driven consumer base. Users are turning to AI not just to create out of whole cloth, but to solve immediate, real-world problems, optimize technical workflows, and acquire new skills on demand.
2. SUPPORTING CONTEXT & METRICS: WORKPLACE INTEGRATION AND DIVERSIFICATION
While consumer metrics provide a wide-angle lens on general adoption, OpenAI’s companion blog post on the expanding world of work dives deep into professional utilization. By mapping workplace-related messages against the standardized O*NET framework—which categorizes tasks across hundreds of distinct occupations into 165 discrete activities—the data offers unprecedented clarity on how artificial intelligence is being operationalized on a global scale.
Work-Related Message Breakdown
Work-related inquiries comprised 32.2% of all analyzed consumer messages in September 2026. Within this professional subset, the proportion dedicated specifically to seeking information remained remarkably stable at between 10.7% and 12.1% each month from July 2024 through July 2026. However, as workplace adoption matured, this metric broke out of its historical band, climbing to 13.9% in August and hitting 15.2% in September.
Decentralization of Professional Tasks
One of the most profound insights revealed in the O*NET-mapped work activity post is the growing decentralization of AI use cases. In the first quarter of 2025 (Q1 2025), professional utilization across the 74 countries with consistent, multi-quarter data was heavily concentrated in a small handful of routine tasks. Specifically, the 10 most common work activities accounted for 56.0% of all classified professional messages in those countries.
Fast forward to the third quarter of 2026 (Q3 2026): that concentration dropped to 51.1%.
This 4.9 percentage-point decline demonstrates that professional use of ChatGPT is maturing and diversifying. Rather than relying on the tool for a narrow suite of repetitive text-generation exercises, global workers are incorporating AI into a wider, more specialized array of occupational tasks.
To illustrate this trend locally, the data tracking the specific ONET activity of "Getting information" in the United States showed it accounted for 11.6% of U.S. work messages in Q1 2025, before shifting to 9.9% in Q3 2026. (Note: Because this ONET metric relies on a specialized domestic classification system, it operates independently of the global consumer "seeking information" metrics, yet it corroborates the broader narrative of task diversification).
3. METHODOLOGICAL COMPARISONS: SIGNALS V3.0 VERSUS THE 2025 OPENAI-HARVARD STUDY
In empirical research, methodology dictates outcome. A critical aspect of analyzing OpenAI’s data transparency efforts involves reconciling Signals V3.0 with an influential academic paper published in 2025 by researchers from OpenAI and Harvard University.
Parallels and Divergences
The 2025 Harvard-OpenAI study tracked ChatGPT usage trajectories through mid-2025, famously reporting that information-seeking had surged from 14% to 24% of total usage by the midpoint of 2025. This figure notably outpaces the peak numbers recorded in the public Signals dataset over the exact same timeframe.
When evaluating why discrepancies exist between the two datasets, several methodological factors emerge:
- Classifier Consistency: According to the Signals data dictionary, the public reporting utilizes the exact same classifier prompts deployed in the academic study. Furthermore, both systems categorize the exact same core domains—facts, product/service inquiries, and cooking/recipe help—under the umbrella of "seeking information."
- Temporal Alignment: In the overlap period concluding in June 2025 (the final month covered by the research paper), the Signals graphs report information-seeking at 18.3%, whereas the academic paper pegs the category significantly higher at 24%.
- Sampling Scale and Adjustments: The Harvard-OpenAI study analyzed approximately 1.1 million distinct conversations using a sampling methodology that extracted a single message per conversation, averaging the results over 28-day rolling periods. By contrast, Signals processes a baseline sample of 1 million messages per month, periodically retrofitting and altering historical monthly baselines to maintain statistical consistency with newer data releases.
Despite these variations, the writing trendlines exhibit a much higher degree of harmony between the two datasets. The academic paper recorded a 12-point drop in writing tasks (plummeting from 36% down to 24%), while the Signals dataset captured an 8.4-point reduction (slipping from 34.1% to 25.7%) over the corresponding timeframe.
While OpenAI’s public documentation does not explicitly explain the variance in information-seeking percentages between the academic paper and Signals V3.0, data scientists emphasize that minor differences in post-processing normalization, prompt filtering, and conversation-level versus message-level aggregation can yield divergent macroeconomic percentages from similar micro-level inputs.
4. OFFICIAL INSIGHTS AND MARKET IMPLICATIONS
The publication of Signals V3.0 and its accompanying qualitative analyses offer profound takeaways for software developers, enterprise leaders, and digital marketers seeking to understand the shifting utility of generative AI.
The Rise of Transactional and Product Discovery Queries
With roughly one out of every five consumer messages dedicated to looking something up, ChatGPT has cemented its status as an influential discovery platform. The inclusion of product and service inquiries within the "seeking information" umbrella carries monumental implications for e-commerce, digital marketing, and Search Engine Optimization (SEO).
As consumers increasingly bypass traditional search engines in favor of conversational, direct answers provided by large language models, brand visibility within LLM outputs is becoming a primary battleground. However, a notable limitation persists within the current Signals data structure: while OpenAI tracks the overarching topic of conversation, the reporting does not disclose the precise frequency with which ChatGPT accesses external web sources or cites third-party websites in its responses. This opacity leaves digital marketers eager for more granular attribution metrics.
Limitations of Aggregated Classification
Observers must also account for the broad nature of OpenAI’s categorical grouping. By bundling factual inquiries, product/service questions, and recipe assistance into a single monolithic "seeking information" metric, the dataset obscures the granular breakdown between casual curiosity and high-intent commercial research. Analysts cannot yet definitively determine what proportion of these informational queries represent academic research versus consumer purchasing decisions.
5. FUTURE OUTLOOK: NAVIGATING ITERATIVE DATA REVISIONS
As the artificial intelligence landscape continues to evolve at a breakneck pace, initiatives like OpenAI Signals serve as an invaluable window into human-computer interaction. However, interpreting these datasets requires a commitment to methodological rigor and longitudinal tracking.
Version Control and Data Integrity
A crucial operational takeaway for industry analysts, journalists, and academic researchers is the mutable nature of the Signals repository. Each successive update to the Signals data dictionary and changelog not only introduces a new quarter of real-world data but frequently retroactively revises historical months to reflect improved classification models and larger sample baselines.
Consequently, a metric cited from Signals V3.0 may deviate subtly from the exact same month’s figure reported in a subsequent version. Professional reporting must therefore explicitly note the specific version number, data release date, and exact temporal scope when citing OpenAI platform statistics.
Looking Ahead
As work-related usage continues to decentralize across more than 160 specialized professional activities and practical guidance dominates consumer interactions, the traditional boundaries separating work, education, and entertainment within AI platforms are dissolving. OpenAI Signals will undoubtedly remain a primary barometer for tracking how humanity integrates artificial intelligence into the fabric of daily life and labor—provided stakeholders remain vigilant regarding its evolving metrics and methodological nuances.
