Overview
ChatGPT is the product that kicked off the generative AI gold rush. Launched in November 2022, it went from zero to 100 million users faster than any consumer product in history. OpenAI says weekly active users passed 1 billion in August 2026, up from 900 million in February and 400 million a year earlier. That is roughly one in eight people on the planet returning to a single company’s chatbox every week.
The product itself is genuinely useful. It can draft emails, debug code, explain complex topics, and help you think through problems. By mid-2026 it does more than answer questions. Agent Mode and the ChatGPT Work product it powers can browse the web, run code, and hand back a finished spreadsheet, slide deck, or document with limited supervision. The better it gets at doing things for you, the more of your judgment it quietly takes over.
What It Knows About You
Every conversation you have with ChatGPT is stored by default. OpenAI’s privacy policy gives them broad rights to use your inputs for model training unless you explicitly opt out, and even then, your data is retained for 30 days for “abuse monitoring.”
The system now also keeps a standing memory profile, editable at chatgpt.com/settings/personalization, that stores facts it extracts from your conversations: your preferences, your past decisions, your personal context. OpenAI has not published an expiration for it. It persists until you delete it yourself.
OpenAI updated its US privacy policy on February 9, 2026, then again on April 30, formalizing an advertising pilot inside ChatGPT. Marketing cookies are now on by default for free-tier users, and OpenAI shares limited identifiers and purchase data with ad partners for targeting; conversation content is excluded, but paid subscribers are the only ones exempted from the cookies themselves. A separate class action, consolidated in the Northern District of California, alleges ChatGPT’s own tracking code shared query topics, account identifiers, and email addresses with Meta and Google. In late 2025 the judge found the plaintiffs had standing and let the case proceed past OpenAI’s motion to dismiss.
OpenAI has suffered data breaches before. In March 2023, a bug exposed other users’ chat histories and payment information. In 2024, internal Slack messages were compromised. Italy banned ChatGPT temporarily. The FTC investigated. None of this has slowed adoption.
The Real Risks
Privacy has gotten worse, not better. Free users now carry default-on ad tracking on top of training-data collection, and a court has allowed a suit over data sharing with Meta and Google to move forward. Enterprise customers get contractual protections. Individual users on the free tier are the product twice over, once for training and once for ads.
Job displacement is accelerating. ChatGPT directly threatens copywriters, customer service agents, junior developers, tutors, translators, and paralegals. OpenAI’s own July 2026 research, drawn from over 800,000 business messages, found 43.5% of occupation-specific requests already cross job-role lines, meaning workers routinely use ChatGPT to do parts of jobs that are not theirs. Freelance writing rates on platforms like Upwork collapsed 30-40% within a year of ChatGPT’s launch, and that trend has not reversed.
The bias problem is real but improving. ChatGPT exhibits measurable biases in its outputs, political, cultural, and demographic. OpenAI has invested heavily in alignment, but the fundamental issue remains: the model reflects the internet it was trained on, and the internet is not a fair place.
Autonomy erosion is no longer subtle. Agent Mode lets ChatGPT plan and execute multi-step tasks largely unsupervised: browsing, writing and running code, producing finished documents. That is a step past “the more you use it to think, the less you practice thinking.” Now it can act on your behalf entirely, and the persistent memory behind it means it increasingly decides what context matters without asking you each time.
Alternatives
- Claude (Anthropic): Similar capability with stronger privacy defaults and a focus on safety research. Still a large AI company with its own risks.
- Perplexity: Better for research tasks where you want cited sources rather than generated text.
- Local models (Llama, Mistral via Ollama): Run entirely on your hardware. No data leaves your machine. Capability is lower but improving fast.
- Just… Thinking: For many tasks people use ChatGPT for, a few minutes of focused thought works better.
Our Verdict
ChatGPT still earns a D, and the numbers behind it moved in the wrong direction this year. Since March, OpenAI turned on default ad tracking for free users, is defending a court-approved class action over sharing query data with Meta and Google, and shipped an agent mode that acts on your behalf instead of just answering you. Privacy, autonomy, and lock-in all moved up a point. Use it if you must. But understand what you are trading now: your data twice over, your cognitive independence, and increasingly, direct control of the actions taken in your name.