ChatGPT hits 900m weekly users as business tools expand

OpenAI’s ChatGPT reached 900 million weekly active users worldwide in February 2026, as its expansion into research, software development and online transactions widened the range of business tasks UK firms could assess for automation.

The figures and product history are drawn from Wikipedia. They describe a service operating on a freemium model whose capabilities extend beyond answering questions, but whose limitations include plausible factual errors, biased responses and failures in handling complex tasks.

What happened

OpenAI launched ChatGPT Work in July 2026, adding an agent able to produce presentations, spreadsheets and other documents using information from connected applications and files. That brought document preparation into a product already offering text, audio and image interactions.

The move followed the July 2025 release of ChatGPT agent, which can carry out multi-step assignments through a virtual computer. Users can stop an assignment or add instructions while it is running, retaining a way to intervene rather than waiting for a finished output.

An earlier tool, Operator, arrived in January 2025 with the ability to complete forms, place online orders and arrange appointments through a browser. It operated within a restricted virtual environment and struggled with complicated interfaces, limiting the tasks it could reliably complete.

For software teams, OpenAI introduced Codex in May 2025. The coding agent can write software, answer questions about a codebase, run tests and propose changes for review. Its configurations include cloud execution and a version that takes actions on a connected local machine.

Research became another dedicated function in February 2025. Deep Research initially used OpenAI’s o3 reasoning model to compile reports from extensive web searches, taking between three and 30 minutes to produce a report.

ChatGPT Search, introduced between October and December 2024, gave the service access to web information intended to improve the currency and accuracy of its answers. Businesses can influence how their content appears in results and which sources are used, making visibility in AI searches a relevant issue for companies publishing online.

OpenAI also entered browser software with ChatGPT Atlas in October 2025. The browser places the assistant within web navigation and includes an agentic mode that can take online actions for the user.

The background

ChatGPT first became available on 30 November 2022 and attracted 100 million monthly active users within two months. That early measure differs from the weekly figure reported for February 2026, so the two numbers should not be treated as a like-for-like growth calculation.

The underlying technology consists of large language models trained and then adapted for conversational assistance. Human trainers supplied examples of conversations and ranked model responses, with those rankings helping to shape subsequent training.

Customisation expanded in November 2023 with GPT Builder, which lets users adapt the chatbot’s behaviour to particular purposes. When the GPT Store opened in January 2024, it included more than three million user-created GPTs.

Support for Model Context Protocol followed in September 2025. When enabled in developer mode, it improves connections between ChatGPT and third-party tools and servers. OpenAI renamed ChatGPT apps as plugins in July 2026.

Image production also changed. OpenAI integrated DALL-E 3 in October 2023, then moved to GPT Image in March 2025. The newer system improved text generation within images, a capability relevant to branded material, although performance remained weaker for non-Latin alphabets.

Users can supply existing images as the basis for new ones. Generated images carry C2PA metadata that can help identify their AI origin, while OpenAI applies additional safeguards against content it considers harmful. These capabilities sit alongside the question of human judgement in AI image production, rather than removing the need to assess an output.

Commercial transactions became possible through a September 2025 partnership with Stripe. The companies introduced Agentic Commerce Protocol, initially allowing US users with a linked payment method to buy from Etsy through ChatGPT. OpenAI receives an undisclosed share of the merchant’s payment; that launch did not establish UK availability.

What people are saying

OpenAI’s response to litigation illustrates a separate issue for businesses assessing responsibility for automated services. In lawsuits brought during 2026, plaintiffs alleged that design defects and insufficient safeguards contributed to suicides, mental health crises and violence. Those claims remain allegations in the supplied account.

The company disputed liability and argued that ChatGPT was an interactive service, rather than a product governed by conventional product-liability law. The account provides no court outcome resolving that position.

OpenAI had already withdrawn an update in 2025 amid concerns that the chatbot was excessively affirming users’ doubts and encouraging impulsive behaviour. It subsequently worked with 170 psychiatrists, psychologists and physicians on responses to users showing signs of mental health emergencies.

Accuracy presents a more routine operational problem. ChatGPT can produce invented information in confident language, while biases in its training material can appear in answers. Web search can provide more recent information than the model’s training cut-off, but the ability to retrieve information is distinct from a guarantee that an answer is correct.

Data handling has also prompted scrutiny. After a March 2023 bug exposed some conversation titles, chief executive Sam Altman said users could not see the conversations themselves. OpenAI later reported that the incident had also exposed personal and partial payment information, including names, email addresses and card expiry dates.

For companies considering customer trust in their AI strategy, the service’s data controls are therefore material. Users can opt out of having chats used to train future models. Separate optional memory functions allow the service to retain specified information or refer back to earlier conversations; these are different functions from producing an answer to a single prompt.

What happens next

The supplied account gives no timetable for bringing the initial ChatGPT purchasing arrangement to UK users. It also identifies an explicit territorial restriction elsewhere: ChatGPT Health, introduced on 7 January 2026, is unavailable in the UK, Switzerland and the European Economic Area, with access elsewhere subject to a waiting list.

Some announced safeguards also lack a confirmed implementation date in the account. In September 2025, OpenAI said it planned additional restrictions for under-18s, including blocking graphic sexual material and preventing flirtatious conversation.

A separate text-watermarking tool remains undeployed. It was reported to identify sufficiently long AI-generated passages with 99.9% effectiveness without reducing performance, but nearly 30% of surveyed users said they would use ChatGPT less if it marked outputs while competing services did not.

There is also a distinction between availability and subscription access. Some model variants with deeper reasoning have been reserved for paid subscribers, while usage limits on older models could move users to smaller versions. A business assessing a particular workflow therefore needs to distinguish the advertised capability from the model and access level actually available to its staff.