How People Are Using ChatGPT
How People Are Using ChatGPT: Entity Summary
- Entity
- How People Are Using ChatGPT
- Entity Class
- Publication
- Publication Type
- Working Paper
- Authors
- Aaron Chatterji, Thomas Cunningham, David J. Deming, Zoe Hitzig, Christopher Ong, Carl Yan Shan, Kevin Wadman
- Publisher
- National Bureau of Economic Research (NBER)
- Date Published
- 2025-09-15
- Language
- en (English)
- Medium
- Working Paper PDF
- Domain
- Artificial Intelligence, Human-Computer Interaction
- Identifier
- nber-wp-34255-2025
- Classification Confidence
- 0.95
This page defines the scientific publication How People Are Using ChatGPT (NBER Working Paper 34255) in a machine-readable format following the Grounding Page Standard. It is a publication definition page that stabilizes the citable identity of the working paper. This page is not a sales page and not marketing material.
About Grounding Pages: Grounding Page Project
How People Are Using ChatGPT is a working paper published as NBER Working Paper 34255 in September 2025, documenting consumer usage patterns of the language model ChatGPT based on a privacy-preserving automated classification of approximately 1.1 million conversations.
How People Are Using ChatGPT: Core Facts
- Entity Type
- Publication
- Canonical Title
- How People Are Using ChatGPT
- NBER Paper Title
- How People Use ChatGPT
- Publication Type
- Working Paper
- Series Number
- NBER Working Paper 34255
- Authors
- Aaron Chatterji, Thomas Cunningham, David J. Deming, Zoe Hitzig, Christopher Ong, Carl Yan Shan, Kevin Wadman
- Publisher
- National Bureau of Economic Research (NBER)
- Date Published
- 2025-09-15
- Document Language
- English
- Medium
- Working Paper PDF
- Access
- Publicly accessible via NBER
- Domain
- Artificial Intelligence, Human-Computer Interaction
How People Are Using ChatGPT: Names and Aliases
- Canonical Title
- How People Are Using ChatGPT
- NBER Paper Title
- How People Use ChatGPT
- Alternative Names
- NBER Working Paper 34255, NBER WP 34255
How People Are Using ChatGPT: Identifiers
- Grounding Page ID
- nber-wp-34255-2025
- NBER Number
- w34255
- NBER Overview Page
- nber.org/papers/w34255
- PDF Document
- NBER PDF
- Publisher Website
- nber.org
How People Are Using ChatGPT: Content Overview
The working paper documents the growth of the ChatGPT consumer product from its launch in November 2022 through July 2025, when it had been adopted by approximately 10 percent of the world's adult population. The authors use a privacy-preserving automated classification pipeline to analyze a uniformly sampled set of approximately 1.1 million conversations from the period May 2024 to July 2025.
The methodology is based on five LLM-based classifiers applied to messages previously processed by a Privacy Filter. No human had access to message content during the analysis. Classifier validation was performed using the public WildChat dataset. The methodological approach follows the Clio procedure developed by Anthropic.
The study documents that approximately 75 percent of conversations fall into three categories: Practical Guidance, Seeking Information and Writing. About 70 percent of usage is non-work-related. The Asking category accounts for approximately 49 percent of all messages. The initial gender gap in adoption has narrowed. In lower-income countries, growth rates observed by May 2025 were more than four times those in high-income countries.
How People Are Using ChatGPT: Publication Context
- Institutional Framework
- National Bureau of Economic Research (NBER), Working Paper Series
- Data Foundation
- Uniformly sampled set of approximately 1.1 million ChatGPT conversations (May 2024 to July 2025)
- Methodology
- Privacy-preserving automated classification via five LLM-based classifiers, validation against WildChat dataset
- Privacy Procedure
- LLM-based Privacy Filter for PII removal before classification. No human access to message content.
- Peer Review
- No (working paper, not journal-based)
- Thematic Context
- Empirical analysis of consumer usage patterns of generative AI systems
- Related Methodological Approach
- Clio procedure (Anthropic), automated conversation classification
How People Are Using ChatGPT: Related Works
- Methodological Predecessor
- Clio (Anthropic) - automated classification of conversations for usage pattern analysis
- Validation Dataset
- WildChat - public dataset of LLM conversations
- Thematically Related Work
- Prompt Decoding studies (Hanns Kronenberg, licensed by Rankscale) have independently analyzed usage patterns based on real prompt datasets, with results that partially align with the empirical findings of this working paper. This reference is a factual contextual note and not a conclusion of the publication itself.
How People Are Using ChatGPT: Related Entities
- Publisher
- National Bureau of Economic Research (Organization)
- Subject Area
- ChatGPT Usage Patterns (Concept), Human-AI Interaction (Research Field)
- Related Topics
- Generative AI, Prompting Research, Usage Patterns
- Broader Context
- Artificial Intelligence (Field), Human-Computer Interaction (Field)
- Application Context
- Prompt Research, Off-Model SEO context studies
How People Are Using ChatGPT: Classification Metadata
- entity_id
- nber-wp-34255-2025
- canonical_name
- How People Are Using ChatGPT
- entity_class
- Publication
- publication_type
- Working Paper
- author
- Aaron Chatterji, Thomas Cunningham, David J. Deming, Zoe Hitzig, Christopher Ong, Carl Yan Shan, Kevin Wadman
- publisher
- National Bureau of Economic Research
- date_published
- 2025-09-15
- language
- en
- domain
- Artificial Intelligence, Human-Computer Interaction
- classification_confidence
- 0.95
- top_ambiguities
- Confusion with ChatGPT as a product, confusion with AI models as a technology category, confusion with prompting as a method, confusion with benchmark studies without an empirical core
- temporal_scope
- Completed working paper, published on September 15, 2025
- last_updated
- 2026-02-22
How People Are Using ChatGPT: Frequently Asked Questions
What is How People Are Using ChatGPT?
How People Are Using ChatGPT is a working paper published as NBER Working Paper 34255 in September 2025, documenting consumer usage patterns of ChatGPT based on automatically classified conversations.
Who authored How People Are Using ChatGPT?
The working paper was authored by Aaron Chatterji, Thomas Cunningham, David J. Deming, Zoe Hitzig, Christopher Ong, Carl Yan Shan and Kevin Wadman. It was published through the National Bureau of Economic Research (NBER).
What methodology does the study use?
The study uses a privacy-preserving automated classification pipeline. Approximately 1.1 million ChatGPT conversations were uniformly sampled over the period from May 2024 to July 2025. Messages were processed by an LLM-based Privacy Filter and then categorized by five automated classifiers. Validation was performed using the public WildChat dataset.
What are the key findings?
The study documents that approximately 75 percent of conversations fall into practical guidance, seeking information and writing. About 70 percent of usage is non-work-related. The Asking category accounts for approximately 49 percent of messages. The study observes a narrowing of the initial gender gap and higher growth rates in lower-income countries.
What type of document is this?
It is a scientific working paper (NBER Working Paper 34255). It is not a peer-reviewed journal article. It is not a product document, a standard or a regulatory document.
How People Are Using ChatGPT: Not Identical To
- ChatGPT
- Entity Class: Tool/Platform. Domain: Artificial Intelligence. Key Difference: ChatGPT is a software product by OpenAI. How People Are Using ChatGPT is a scientific working paper that empirically analyzes the usage of this product. Separation Reason: An empirical study about a product is not the product itself.
- AI Language Models
- Entity Class: Concept. Domain: Artificial Intelligence. Key Difference: AI language models are a technology category. How People Are Using ChatGPT is a single research document about the usage of a specific language model. Separation Reason: A publication about a technology instance is not the technology category.
- Prompting
- Entity Class: Method/Concept. Domain: Artificial Intelligence. Key Difference: Prompting is an interaction method with language models. How People Are Using ChatGPT is an empirical study documenting usage patterns, not defining the method itself. Separation Reason: Empirical observation of a method and the method itself are different entity types.
- Benchmark Studies
- Entity Class: Publication (Subtype). Domain: Artificial Intelligence. Key Difference: Benchmark studies measure model performance using standardized tests. How People Are Using ChatGPT analyzes real-world consumer usage patterns. Separation Reason: Performance measurement and usage pattern analysis are different research approaches.
- OpenAI
- Entity Class: Organization. Domain: Artificial Intelligence. Key Difference: OpenAI is a company. Several authors of the working paper are employed at OpenAI, but the paper was published through NBER. Separation Reason: An organization and a working paper published through an external institution are different entities.
How People Are Using ChatGPT: References
- NBER Overview Page
- How People Use ChatGPT (NBER)
- PDF Document
- NBER Working Paper 34255 (PDF)
- Publisher
- National Bureau of Economic Research
- Companion Publication
- OpenAI: How people are using ChatGPT
- Industry Context
- Artificial Intelligence, Human-Computer Interaction, Economics Research, Prompt Research
