ChatGPT has been an enormous leap forward in AI capabilities. The software has written poetry and songs, revolutionized the high school essay, held deep conversations with users, and more. It’s also spurred discussion about the future of work and what role AI could have in the years ahead.
One of the most incredible things about ChatGPT is how widely it’s been used since it was released. In this guide, we’ll take a look at some key ChatGPT statistics around how this AI model is being used and how it works.
ChatGPT User Stats
The ChatGPT website had 1.6 billion users in March 2023.
This counts all web and mobile users visiting OpenAI’s ChatGPT portal. It does not include users who interact with ChatGPT through the Bing search engine or other software that has integrated ChatGPT.
ChatGPT is the fastest-growing online service ever.
ChatGPT had 1 million users in the first 5 days after it launched in December 2022 and grew to 1 billion users within 3 months. For comparison, Instagram, previously the fastest-growing software of all time, took 2.5 months to reach 1 million users.
|Monthly Active Users
12% of Americans said they had used ChatGPT themselves to generate text, according to a January 2023 YouGov poll.
Another 38% of respondents said they had seen someone else use ChatGPT, but had not used it themselves. 15% of Americans aged 18-29 and 17% aged 30-44 have used ChatGPT. Only 5% of respondents aged 65 or older have used ChatGPT.
|18 to 29
|30 to 44
|45 to 64
|I’ve used it myself to generate text
|I’ve seen text generated for someone else, but haven’t used it myself
|I’ve never used or never seen anyone else use it
20% of people have used ChatGPT to search for information.
- 19% have used it to create content like social media posts.
- 14% have just been playing around with ChatGPT.
- 11% have used it for academic purposes.
- 11% have used it to find recommendations.
- 10% have used it to write code.
ChatGPT is available in more than 185 countries.
The only countries where ChatGPT is not available are:
- North Korea
Traffic to ChatGPT’s website comes from around the world.
- United States: 14.84% of users
- India: 6.23% of users
- Japan: 3.63% of users
- Columbia: 3.19% of users
- Canada: 2.94% of users
- Other countries: 69.18% of users
ChatGPT can accept prompts and answer questions in more than 95 languages.
The model was primarily written in Python.
But ChatGPT understands most popular programming languages. Languages ChatGPT can code in include:
ChatGPT was created by OpenAI in 2015.
OpenAI was launched as a nonprofit in 2015 with seed funding from Sam Altman, Greg Brockman, Reid Hoffman, Jessica Livingston, Elon Musk, Ilya Sutskever, Wojciech Zaremba, and Peter Thiel.
OpenAi transformed from a nonprofit to a capped for-profit company in 2019.
OpenAI’s profits are capped so that early investors can receive at most a 100X return on investment. Any additional profits will be diverted to OpenAI’s nonprofit arm.
Microsoft invested $1 billion in OpenAI in 2019.
The deal made Microsoft the exclusive cloud services provider for all OpenAI services, including ChatGPT. Microsoft invested an additional $2 billion in OpenAI between 2020 and 2022.
Microsoft invested an additional $10 billion in January 2023.
Microsoft now has a 49% ownership stake in OpenAI.
ChatGPT Revenue Statistics
OpenAI expects ChatGPT to generate $200 million in revenue by 2023 and $1 billion in revenue by 2024.
Open AI made less than $10 million in 2022.
Microsoft valued OpenAI at $29 billion during its January 2023 funding round.
ChatGPT was valued at $14 billion in 2021.
ChatGPT Software, Algorithm & Parameters Stats
ChatGPT is a large language model (LLM).
It was created using supervised fine-tuning and Reinforcement Learning from Human Feedback (RLHF).
ChatGPT was trained on 45TB of textual data.
This includes data from books, websites, Wikipedia, news articles, and more. In total, the language model was fed 300 billion words.
The training process required 1,000 NVIDIA V100 GPUs.
It ran on a Microsoft Azure supercomputer. The training model has 24 layers and 96 attention heads. The model was fine-tuned using several gigabytes of data.
GPT-4, the newest version of ChatGPT, has more than 100 trillion parameters.
GPT-3 has an estimated 175 billion parameters.
ChatGPT has passed the bar exam for lawyers.
It also scored 60% on the US Medical Licensing Exam.
ChatGPT Keyword & Web Traffic Statistics
ChatGPT received 53 million unique visitors per day in March 2023.
That’s up from 13 million unique visitors per day in January 2023 and 35 million unique visitors per day in February 2023. The site’s traffic has grown by 56% from February to March.
Visitors to ChatGPT spent an average of 8 minutes and 44 seconds on the site.
They visited an average of 6.67 pages. The bounce rate was 16.07% in March 2023.
Traffic to ChatGPT came primarily from direct connections.
Referrals from other sites were the second-most common source of traffic.
|Traffic Source to ChatGPT
|Percentage of Total Traffic
Search traffic indicates that users are searching for ChatGPT by name.
Although only 4.37% of users are finding ChatGPT through a search engine, the search terms used to find it suggest that they aren’t just looking for any AI model.
|Monthly Traffic to ChatGPT
|Chat GPT Login
ChatGPT on Social Media
Nearly 4% of referrals to ChatGPT came from social media.
- 51.47% came from YouTube
- 16.27% came from Twitter
- 11.75% came from Facebook
- 8.80% came from WhatsApp
- 3.93% came from Reddit
- 11.71% came from other social media channels
Cost of Using ChatGPT & Its Pricing
ChatGPT costs around $100,000 per day to run.
That’s nearly $3 million per month. Most of the costs stem from computing in the Microsoft Azure cloud. When Microsoft invested $1 billion in OpenAI in 2019, most of this investment came in the form of Azure computing credits.
The details of Microsoft’s contract aren’t public, but Azure typically charges $3 per hour per GPU. At that rate, each word generated by ChatGPT would cost $0.0003.
ChatGPT-3 and 3.5 remain free for anyone to use.
However, free users may not be able to access the algorithm when demand is spiking.
OpenAI introduced ChatGPT Plus, a premium subscription that costs $20 per month.
This provides access to ChatGPT even during peak times as well as faster responses from the AI model. It’s also currently the only way to access GPT-4, the latest model from OpenAI.
Stats on What People Think of ChatGPT
70% of people believe that ChatGPT will improve people’s lives.
However, 87% of people also report fearing that ChatGPT will be used by companies or governments to control the population.
In the US, only 13% of people think ChatGPT will be good for society.
36% think it will be bad for society, 23% think it will be neutral, and 28% aren’t sure.
63% of users think that businesses should be able to advertise on ChatGPT.
OpenAI has not yet said whether it will enable advertising in ChatGPT responses.
85% of people would use ChatGPT as a therapist.
60% said they would use ChatGPT to get medical advice. 13% said they would use ChatGPT for flirting or dirty talk.
83% of people are scared that ChatGPT will be biased towards minorities.
Implicit bias against minorities is a well-documented problem with existing AI models, such as facial recognition algorithms.
50% of people are afraid ChatGPT will replace their jobs.
The top jobs that people think could be replaced are:
- Software development
- Data analysis
- Customer service
- Content creation
14% of Americans believe that ChatGPT will end up creating more jobs.
59% of teachers say that ChatGPT will be useful for education.
38% of teachers have already allowed their students to use ChatGPT. 25% believe ChatGPT will only be used by students to cheat.
What Elon Musk & Other Celebrities Think of ChatGPT
Elon Musk says that AI technology like ChatGPT is “one of the biggest risks to the future of civilization.”
Musk has pointed out that AI can do a lot of good. But he says that if it is not handled properly, AI could be more dangerous than nuclear weapons.
Microsoft founder Bill Gates calls AI like ChatGPT “as fundamental as the creation of the microprocessor, the personal computer, the internet, and the mobile phone.”
He says that although ChatGPT is still imperfect, AI will be the biggest technological evolution of the decade.
Google CEO Sundar Pichai says generative AI “is the most profound technology” Google is developing.
Google recently released its own competitor to ChatGPT, called Bard. After ChatGPT was released to the public, Pichai issued a “code red” within Google because of the threat ChatGPT poses to its search business.
Microsoft CEO Satya Nadella cautions that humans must be “unambiguously, unquestionably” in control of AI models.
Nadella has not commented on the fact that OpenAI – backed by Microsoft – launched a race among tech companies to make their AI models public.
Apple CEO Tim Cook sees great potential in ChatGPT and related AI models: “It will affect every product and every service that we have.”
Apple already integrates AI assistants into many of its devices. ChatGPT and other AI models have the potential to supercharge Apple’s product line.
Warren Buffett says ChatGPT could reduce jobs: “I would certainly think it would result in significantly less employment in certain areas.”
Buffett believes that’s a net good thing for the economy and could allow people to have more leisure time. However, he also cautions that the disruptive effects of AI could cause problems for society in the short-term.
History of ChatGPT
OpenAI, the company behind ChatGPT, was founded in 2015.
The company launched as a nonprofit in San Francisco and was led by Elon Musk, Sam Altman, Greg Brockman, Wojciech Zaremba, Ilya Sutskever, and John Schulma. The goal was to create safe AI that humans could control, rather than allow AI to become dangerous to society.
OpenAI developed the first generative pre-trained transformer (GPT) in 2018.
The first model was trained on 7,000 unpublished books.
GPT-2 contained 1.5 billion parameters.
The second version of the GPT model was developed in 2018 and was trained on 8 million web pages. OpenAI decided not to release GPT-2 to the public out of fear that it could be used to write scam emails or provide misinformation.
Elon Musk left OpenAI in 2018.
Musk has said since before founding OpenAI that AI is one of the greatest threats to the future of civilization. He left around the same time that OpenAI decided not to release GPT-2 to the public.
OpenAI became a capped for-profit company in 2019.
OpenAI decided to switch from a nonprofit to a for-profit company. However, it made the unusual decision to cap its future profits. Any additional profits would go to OpenAI LP, a nonprofit that owned OpenAI the company.
OpenAI released DALL-E in 2021.
DALL-E uses a similar neural network architecture to ChatGPT, but creates images based on a prompt rather than text.
OpenAI created GPT-3 in 2022.
To build GPT-3, Microsoft built the world’s 5th-largest supercomputer, made with 10,000 GPUs. ChatGPT is built on top of GPT-3 and adds the ability to have a conversation that involves repeated prompts in a single session.
Microsoft integrated GPT-3 into Bing and Edge.
In February 2023, Microsoft took the first steps to incorporate ChatGPT into its existing products. Microsoft Edge users who searched the internet with Bing could see both standard search results and a result from ChatGPT.
OpenAI launched GPT-4 in March 2023.
GPT-4 is the latest iteration of OpenAI’s GPT models. It’s currently available only with a ChatGPT Plus subscription. OpenAI says that GPT-4 is 40% more likely to provide accurate answers and 80% less likely to break the rules that its creators set for it.
How Does ChatGPT Work?
ChatGPT is a transformer model.
This is a type of neural network built specifically for processing language. In a transformer, the model looks at a sequence of words to determine the meaning and context of the next word. This is similar to how most people read text.
ChatGPT has 24 layers.
A transformer model contains multiple layers, which are like networks of neurons in the human brain. As a sequence of words is passed through each layer, a slightly different weighting is applied to the importance of each word. By the time it passes through all of the layers, ChatGPT is relatively confident in the context, meaning, and importance of each word in the sequence.
ChatGPT was trained on 45TB of text data from across the internet.
The initial part of the model training process is called pre-training. ChatGPT “read” 300 billion words of text from a database called WebText2. ChatGPT’s training was largely unsupervised, meaning that the model looked at text and detected patterns on its own. It did not have a human prompter asking it to read something, then asking it questions and giving it feedback.
ChatGPT was fine-tuned using dialogue between two people.
To help ChatGPT understand conversations, the model was given 160,000 dialogues between two people. In each dialogue, the two people had different tones, motivations, and backgrounds.
After ChatGPT was pre-trained, humans stepped in to adjust the model.
AI researchers had many interactions with ChatGPT before the model was released. They could give the model feedback on its answers, enabling ChatGPT to adjust how it responded to questions in the future.
ChatGPT can have multi-turn conversations.
One of the most noteworthy aspects of ChatGPT is that it doesn’t just issue a response and then reset. The AI model can hold a back-and-forth conversation or ask clarifying questions around a prompt.
What is ChatGPT Used for?
ChatGPT is free for anyone to use.
It’s already common in many industries, even if it hasn’t yet been integrated into many existing software applications.
Common uses of ChatGPT include:
- Drafting emails to coworkers
- Writing a resume or cover letter for a job application
- Summarizing complex topics or long articles
- Getting answers to questions without a traditional web search
- Writing songs, poetry, and screenplays based on existing content
- Writing and debugging computer code
- Translating content into multiple languages
- Writing essays on any topic
- Solving math problems
- Finding new recipes based on a set of ingredients
People have also used ChatGPT for:
- Getting free medical advice
- Getting free legal advice
- Getting free online therapy
- Flirting with an AI
These uses aren’t recommended by OpenAI, but many ChatGPT users have found that they work reasonably well.
Limitations, Challenges & the Future of ChatGPT
ChatGPT can generate false information.
While ChatGPT often gives correct answers, it can also give incorrect answers with the same level of confidence. This makes it difficult to trust ChatGPT, especially on complex or important topics.
ChatGPT is not aware of events after September 2021.
ChatGPT’s training database stops in September 2021. The AI model doesn’t have any world events that happened after that time. In the early days of ChatGPT, the AI argued with some users about events that happened, but which ChatGPT was not aware of.
ChatGPT has a word limit of around 500 words.
If a user’s prompt is longer than 500 words, ChatGPT may fail to respond. The model may also cut off responses at around 500 words. In that case, users have to ask ChatGPT to continue its response from where it left off. Reportedly, GPT-4 (available with a ChatGPT Plus subscription) has significantly raised this word limit.
ChatGPT is not always available.
During hours of peak demand, the ChatGPT website can fail to load. Users simply have to wait until demand dies down before they can make a request. ChatGPT Plus subscribers receive priority access during peak hours.
ChatGPT can be used for malicious purposes.
While ChatGPT is not supposed to answer certain types of prompts, the controls over the model are not perfect. ChatGPT has been used to generate misinformation, write malware code, or even write phishing emails.
The Future of ChatGPT
OpenAI released GPT-4 on March 13, 2023, just 3 months after ChatGPT launched.
The newest version of OpenAI’s model has more than 100 trillion parameters. The fact that it came out only 3 months after ChatGPT suggests that OpenAI could continue to release updates to ChatGPT throughout the year.
Microsoft announced plans to integrate ChatGPT into Office 365.
Office 365, which includes Word, Excel, Powerpoint, and Outlook, is used by 345 million people around the world. Integrating ChatGPT (specifically, Microsoft’s slightly modified Prometheus model) could make the AI model much more accessible to many businesses and industries.
Businesses could begin integrating ChatGPT into their workflows.
Right now, relatively few businesses are using ChatGPT on a daily basis. However, that could change in the near future. The result could be that workers’ job duties change or that companies need to hire fewer employees.
ChatGPT vs Competitors
Competitors to ChatGPT include Bing Chat, Chatsonic, and Google Bard.
Bing Chat and Chatsonic are closely related to ChatGPT, with a few slight differences to how the model is trained. Both are based on OpenAI’s GPT-3.5 model. Google Bard is Google’s own language model, which it just released to the public in March 2023.
|Type of Platform
|AI-powered chatbot with search engine features
|AI-powered chatbot with search engine features
|Prometheus (Microsoft’s version of GPT-3.5)
|45TB GB of textual data from websites, books, and other sources.
|Same as ChatGPT plus websites from Bing Search Index
|Same as ChatGPT plus real-time indexing
|Google Search Index
|No data after September 2021
|Aware of all events up to the present.
|Aware of all events up to the present.
|Aware of all present & past events
|Free (Plus subscription costs $20 per month)
|Free (only available in Microsoft Edge browser)
|Free (only available in the US and UK)
Fun Facts on ChatGPT
ChatGPT is creating a role for “prompt engineers.”
Getting the best answer out of ChatGPT – or pushing the algorithm to expose its flaws – requires carefully crafted prompts. As a result, OpenAI and other companies working on AI language models have begun hiring prompt engineers. These aren’t necessarily computer scientists, but rather creative writers who can figure out the best way to talk to the AI model.
ChatGPT can tell a great joke.
Although ChatGPT isn’t that funny by default, it can be with the right set of prompts. For example, users can ask ChatGPT to write a story in the style of their favorite comedian. Or they can tell ChatGPT to be sarcastic in its responses. Getting ChatGPT to be funny requires trial and error, but it’s very rewarding when it happens.
ChatGPT is constantly learning.
Every time users hold a conversation with ChatGPT, the AI model learns something new. It is constantly fine-tuning itself based on feedback. So, if a user prompts ChatGPT, gets an answer, and then asks for clarification, the model will take that follow-up question into account the next time it answers a similar question.
ChatGPT is already transforming some industries.
It’s more likely than not that people have run into ChatGPT-generated text without knowing it. Professionals across many industries have started using it to draft emails. Real estate agents are using it to write listings. Programmers are using ChatGPT to write and error-check their code.
- ChatGPT Plus
- Microsoft Azure & ChatGPT
- YouGov poll (January 2023)
- ChatGPT Website Traffic (Similarweb)
- How to Cite ChatGPT