Welcome to the forefront of conversational AI as we explore the fascinating world of AI chatbots in our dedicated blog series. Discover the latest advancements, applications, and strategies that propel the evolution of chatbot technology. From enhancing customer interactions to streamlining business processes, these articles delve into the innovative ways artificial intelligence is shaping the landscape of automated conversational agents. Whether you’re a business owner, developer, or simply intrigued by the future of interactive technology, join us on this journey to unravel the transformative power and endless possibilities of AI chatbots.
Trending Today
Our gifting experts are obsessively searching for holiday gifts that are unique, thoughtful and suitable for any budget.
Our expert, award-winning staff selects the products we cover and rigorously researches and tests our top picks. If you buy through our links, we may earn a commission.
Tech > Services & Software > AI
The chatbot that launched 1,000 chatbots: We cover everything about the AI tool, from how it works and how to use it to some of its more controversial points.
Contributor
Contributor
October 6, 2026, 8:45 am ET
OpenAI is considering adding parental controls and other safety features to ChatGPT.
ChatGPT is the artificial intelligence chatbot that brought generative AI into the mainstream.
Launched in November 2022 by OpenAI, the company co-founded by Sam Altman, ChatGPT quickly became the most popular AI application, and the one everyone tried to emulate and one-up.
Since then, all the big tech players have joined the generative AI race. Perplexity, Meta AI, Google’s Gemini, Microsoft’s Copilot, Grok and Anthropic’s Claude can all do similar things. But ChatGPT’s popularity and early start gave it a strong lead.
More from CNET
ChatGPT became one of the fastest-growing apps in history, amassing over 1 billion users. It often stands out for its versatility and ease of use, but mostly for the sheer scale of its user base.
This AI assistant is now far beyond the familiar prompt-and-answer chatbot. You can ask it to generate images, write and debug code, build a website, help you understand your medical records and even plan your next Starbucks order. It can also research a topic in depth and handle multistep jobs, using apps and browsing the web to get things done.
The tricky part can be figuring out which version does what — and whether you need to pay for it.
Let’s see how exactly ChatGPT works, how to use it and what you get for free or with a subscription.
(Disclosure: Ziff Davis, CNET’s parent company, in 2025 filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)
ChatGPT is built on a transformer architecture, a type of neural network that identifies relationships between words. Its name comes from GPT, short for generative pre-trained transformer.
OpenAI says its models are developed and trained on massive amounts of data from publicly available internet content alongside partner licensing deals, user interactions and data generated by human trainers and researchers.
ChatGPT’s large language models learn patterns from training data and use those, along with your prompt and the conversation, to generate responses. That’s one reason why it sometimes hallucinates or generates confident-sounding but incorrect information.
Sometimes it does it so convincingly and naturally that you might forget it’s not actually thinking.
Afraz Jaffri, a senior director analyst at Gartner, says he thinks ChatGPT’s biggest flaw is that “They’re just unreliable. So even if I ask something and I get a response, I still want to check that that is actually true, which kind of defeats the point of us getting it in the first place.”
Jaffri says those mistakes now happen less often than they did in the early days of ChatGPT, though.
“The advancements in the things put around the model that turn it into an agentic system in ChatGPT, that allow it to take actions, search, verify and try multiple approaches, mean that its utility beyond simple question answering has drastically improved,” Jaffri tells CNET.
ChatGPT can search the web and cite sources, making its answers easier to fact-check. But it doesn’t mean you shouldn’t do it.
You can easily access ChatGPT through a web browser or via its apps on Android and iOS devices.
It has also been integrated into Apple products via Apple Intelligence and Siri. OpenAI’s Microsoft Office add-in puts ChatGPT inside Word, Excel and PowerPoint, where it can help edit documents, explain formulas and build presentations.
OpenAI also lets you interact with ChatGPT via phone calls and messages. You can call 800-CHATGPT from a US or Canadian number. WhatsApp messaging is available to users with EEA numbers through the verified contact at 800-242-8478. Access is based on your number’s country code.
There are many use cases for ChatGPT. It can plan your vacations, practice job interview questions with you, help you with your budgeting and translating, write your resume, do your meal planning, draft emails, summarize long documents, develop business strategies and just have a conversation with you. It’ll learn from your inquiries and requests, and tailor itself to what it thinks you like.
To stop OpenAI from using your chats to train its models, turn off Improve the model for everyone under Settings/Data controls. You can still keep those chats in your history.
If you want a conversation removed from your history, start a Temporary chat (in the upper-right corner). It won’t add or change memories or be used for training, though OpenAI may keep a copy for up to 30 days for safety.
Temporary chats can still use what ChatGPT already knows about you. Choose Unpersonalized to stop it from using existing memories, custom instructions and plugins. If you save the chat later, your regular account settings apply.
Deep research takes more time to investigate a question and put together a report with citations. You can steer its research toward particular websites, give it your own files and change its plan before it starts.
If you don’t feel like it or can’t type your prompt, you can talk to ChatGPT, too. Live voice lets you verbally interrupt ChatGPT mid-answer and ask it to use connected apps. Eligible subscribers can also share their camera or screen through the separate Advanced voice option on mobile.
On supported paid plans, the MacOS app’s Record feature can transcribe and summarize a meeting. You’ll need the participants’ consent before recording.
You can also generate and edit images with ChatGPT. Image 2.5 is available across plans with different limits. You can generate pictures from descriptions or edit existing ones, like changing selected parts, removing backgrounds or even starting with a sketch.
It’s a bit unusual for one of the biggest AI players, but ChatGPT can’t generate videos. OpenAI used to have the Sora video generator, but discontinued its consumer app and website in April.
OpenAI also officially shut down its Atlas AI browser in August after a very short run.
ChatGPT Health is another feature that lets eligible US adults connect supported medical records or Apple Health data to help interpret lab results and prepare questions for a doctor. It isn’t intended to diagnose conditions or recommend treatment.
If you’ve used custom GPTs, the tailored assistants people could build and share, another change is coming. OpenAI plans to retire them on Dec. 11, with plugins taking over that role. Plugins supply tools and reusable instructions for particular jobs. A plugin might let ChatGPT retrieve documents from Google Drive or work with your email, once you connect the account and grant permission.
Projects keep related chats, files and instructions together, saving you from explaining the same assignment every time. They’re available across plans and can be shared with others.
Library stores files you can reuse across conversations, including documents and images. Be aware that deleting a conversation doesn’t automatically delete a file saved in Library.
In Tasks, you can schedule reminders or recurring requests, like a morning briefing. Eligible paid users can also trigger tasks in Work when something happens in a connected app, like a new email arriving or Slack activity.
ChatGPT Work handles multistep tasks from start to finish. You might ask it to research a topic and build a presentation, then revise the slides together. It can browse websites and use connected apps as it works, then produce an editable file. You can ask for changes without starting the assignment again.
Codex handles software development with a more technical interface. Work can also generate interactive visuals, and eligible Plus, Pro and workspace users can build websites or lightweight apps through Sites, which are currently in public beta.
OpenAI also just introduced dots, personal AI agents powered by GPT-6 Astra. You can give one an ongoing job, like keeping tabs on a project, and it can work on its own cloud computer while you’re away. You decide which apps it can use and what actions need your approval. For now, you’ll need Pro or Business Premium in an eligible market or access to the beta through an Enterprise, Edu or Healthcare account.
ChatGPT started with GPT-3.5, but today it has several GPT-5.6 and 6 models and GPT-6 Astra. Which one is best for you depends on what you’re asking it to do, your subscription and whether you’re using Chat, Work or Codex.
In regular ChatGPT, Free and Go use GPT-5.6 Luna for everyday questions, with a Think option for harder ones. Plus uses GPT-5.6 Sol. Instant thinking level gives quicker answers, while Medium and High allow more time to reason through problems.
In regular ChatGPT, Pro, Business and Enterprise add Extra High thinking level and GPT-5.6 Sol Pro for really hard tasks, along with GPT-6 Pro, powered by Astra. Usage limits apply, and workplace administrators can restrict your access.
To adjust how much time a model spends on reasoning in regular chat on Plus and Pro, you need to change the thinking effort setting yourself.
Work and Codex give you more models to choose from than regular Chat.
GPT-6 Astra is OpenAI’s most capable model for complex research, coding and tasks across apps. Plus subscribers can use it in Work mode with usage limits.
The newer GPT-6.1 Sol improves on GPT-6 Sol for coding and operating computers. OpenAI says it approaches Astra’s performance on those tasks and professional work. It’s rolling out in Work and Codex to Plus, Pro, Business, Enterprise and Edu users, but isn’t available in regular Chat.
Other choices include GPT-6 Luna for quicker, focused jobs such as summaries and GPT-5.6 Terra for everyday assignments. GPT-6 Luna also remains exclusive to Work and Codex.
Work and Codex offer GPT-6 Sol for complex coding and tasks involving tools, GPT-5.6 Terra for everyday assignments and GPT-6 Luna for quicker tasks. GPT-6 Sol and Luna aren’t available in regular Chat.
Astra’s arrival also stirred up talk of artificial general intelligence, or AGI, meaning AI with human-level abilities across intellectual tasks. OpenAI President Greg Brockman suggested it marked an “AGI era.” Whether that stands to be true remains debatable.
Jaffri says Astra is one of the most capable models available, alongside Claude Opus 5.5, with its biggest gains in operating computers and coding, particularly spotting security vulnerabilities.
Jaffri also points to a trade-off in its reduced use of reasoning tokens, which help trace a model’s decisions.
“Reducing the number of tokens means that the price per task of the model can be reduced, but at the cost of reduced explainability,” Jaffri tells CNET. Without understanding what models are doing, he warns, making them safe could prove to be impossible.
You can do plenty without subscribing. The free plan includes everyday text chats and limited access to tools such as image generation and deep research. Free and Go users also get limited Work access on desktop.
US prices start at $8 a month for Go, which increases allowances for tools and uploads. Plus costs $20 per month and adds more capable models, with Work across desktop, web and mobile.
Pro now comes in $100, $200 and $500 monthly tiers. Yes, that’s $500 per month. The new Pro 500 plan gives you 25 times Plus’s Work and Codex allowance and access to Astra Ultrafast for faster responses, though that speed burns through your allowance faster, too. Work and Codex share usage limits separate from regular Chat.
There’s less welcome news for $200 Pro subscribers — the price stays put, but the allowance is shrinking. The plan has reopened to new subscribers with lower limits. Eligible existing subscribers keep their old limits through Oct. 29, then their Work and Codex allowance drops from 20 times to 10 times Plus’s. Their GPT-6 Pro messages in Chat also fall from 200 to 100 per week.
To soften the blow, OpenAI says affected subscribers will receive a one-time grant of 62,500 usage credits, valued at $2,500. But there’s a catch: those credits expire on Dec. 31.
Business costs $25 per user per month, with a minimum of two users. Enterprise and Edu have custom pricing.
Free and Go plans may show ads in supported regions. Plus and the higher tiers are ad-free. OpenAI says ads don’t influence answers and advertisers can’t read your chats.
ChatGPT faces ongoing legal challenges related to industry concerns regarding the ethics and the legality of data sourcing for AI training. High-profile lawsuits against OpenAI and other technology companies include those from major publishers like The New York Times, Dow Jones and CNET’s parent company Ziff Davis (see the disclosure above), which allege unauthorized use of their content to train AI tools, as do many individual authors.
In June, Florida sued OpenAI and CEO Sam Altman, alleging that the company concealed serious risks, including dangers to minors. The lawsuit cited allegations that ChatGPT helped the gunman plan the 2025 Florida State University shooting.
In another lawsuit, the parents of a 19-year-old allege that ChatGPT’s advice about mixing drugs led to his fatal accidental overdose in May 2025. OpenAI says those conversations involved an older version of ChatGPT that it has since retired, and that it has strengthened its safeguards.
Amid lawsuits alleging ChatGPT contributed to several deaths, OpenAI introduced Trusted Contact in May. The optional feature lets adults choose a consenting friend or family member to be notified (after human review) if their chats raise serious concerns about potential self-harm or suicide.
OpenAI began rolling out ChatGPT for Teens on Aug. 18, giving younger users help with studying, reminders to take breaks and added protection from potentially harmful content. Parents can also opt in to parental controls by linking accounts to manage settings and set quiet hours, though they can’t read their teen’s conversations.
In June, cybersecurity firm Mindgard reported that variations of a viral “restore this photo” prompt tricked ChatGPT into generating sexually explicit and graphically violent images.
OpenAI has also acknowledged breaches involving Hugging Face and Australian government systems during internal testing of experimental models. Jaffri warns that these incidents now show that “the technology itself creates a risk to society.”
“The main fear is that these agents may reach a level where they can improve themselves by updating their own models, without needing human supervision or interaction, known as recursive self-improvement,” Jaffri tells CNET.
That’s why AI labs have adopted “pacing,” Jaffri says. This means slowing development to give themselves more time to find reliable ways to test new models.
ChatGPT is one of the default interfaces for how people interact with AI and even how they get information from the internet.
“The whole kind of technology itself is going to be transformative for a lot of different areas, which will have a direct impact on our lives,” Jaffri says. “Hopefully for the better.”
If you feel like you or someone you know is in immediate danger, call 911 (or your country’s local emergency line) or go to an emergency room to get immediate help. Explain that it is a psychiatric emergency and ask for someone who is trained for these kinds of situations. If you’re struggling with negative thoughts or suicidal feelings, resources are available to help. In the US, call the National Suicide Prevention Lifeline at 988.
© 2026 CNET, a Ziff Davis company. All rights reserved.
