This guide brings together and replaces our earlier short posts from 2023 on ChatGPT, AI tools, recommendation engines and AI privacy, rewritten and expanded in October 2026 as a current, plain-English guide with privacy steps for the main AI assistants.
Generative AI went from a curiosity to an everyday tool in about three years. ChatGPT was released to the public on 30 November 2022, and by early 2025 a Pew Research Center survey found that 34% of U.S. adults had used it, roughly double the share in 2023. Among adults under 30 the figure was 58%. Most people now meet these tools at work, in search results, inside their phone's keyboard or in the apps they already use.
That makes it worth understanding what these systems actually do, where they go wrong and what happens to the things you type into them. This guide explains generative AI without the hype: what it is good at, which assistants are worth knowing, how the recommendation systems behind your feeds work, what privacy regulators have found, and which settings to change today.
In this guide
- What ChatGPT and other generative AI tools can (and can't) do
- The main alternatives to ChatGPT, and what they are not
- How recommendation engines decide what you see
- AI privacy: what regulators found, and the settings to change
- Where AI is already part of daily life
- Quick-start checklist
- FAQ
What ChatGPT and other generative AI tools can (and can't) do
ChatGPT, Gemini, Copilot, Claude and similar assistants are built on large language models. A language model is trained on an enormous amount of text and learns the statistical patterns of how words follow one another. When you type a request, it produces a response one small chunk of text at a time, choosing what is most likely to come next given everything before it. The result reads fluently because fluent text is exactly what it was trained to produce.
Two points follow from that. First, the model is not looking facts up in a database unless the product adds a web search or file-reading step on top. Second, fluent is not the same as correct. A confident, well-written answer can still contain invented details, a problem usually called "hallucination". Newer assistants reduce this by searching the web and showing their sources, but they do not eliminate it.
Where it genuinely helps
- Drafting and rewriting. First drafts of emails, cover letters, product descriptions or meeting agendas; rewriting something to be shorter, friendlier or more formal.
- Summarizing. Condensing a long report, contract or transcript you paste in or upload, so you know which parts to read closely.
- Explaining. Asking for a concept in plain language, then asking follow-up questions until it makes sense. This is one of the most popular uses: in Pew's survey, 26% of all adults had used ChatGPT to learn something new.
- Brainstorming. Lists of names, angles for a talk, gift ideas or ways to structure a project. Treat the output as raw material, not a final answer.
- Translation and tone. Rough translations and help understanding a message in another language. For anything legal, medical or official, use a qualified human translator.
- Coding and spreadsheets. Explaining an error message, writing a formula or a small script. Test everything before you rely on it.
- Images, audio and video. Many assistants now also generate pictures and voice. These carry their own risks, from copyright questions to realistic fakes, so label AI-made media when you share it.
Where it lets you down
- Facts, figures and quotes. Always check names, numbers, dates and citations against the original source. Ask the assistant for links, then open them.
- Recent events. Without a web search step, a model only knows what was in its training data, which stops at a cutoff date.
- Medical, legal and financial decisions. It can help you understand terms and prepare questions, but it is not a substitute for a professional who knows your situation.
- Anything it can't see. It cannot know what is on your screen, in your inbox or in a document unless you share it, and it may guess rather than say so.
How to get better answers
- Give context. Say who the result is for and why: "a two-paragraph note to my landlord about a broken heater, polite but firm."
- Show an example. Paste a sample of the style or format you want.
- Set limits. Word count, reading level, bullet points or a table.
- Iterate. Reply with what to change rather than starting over. The conversation is the tool.
- Ask it to check itself. "What in this answer might be wrong or out of date?" often surfaces weak spots worth verifying.
Common pitfall: pasting in confidential work documents or other people's personal details. Check your employer's AI policy first, and see the privacy section below before you share anything sensitive.
The main alternatives to ChatGPT, and what they are not
Many "ChatGPT alternatives" lists from 2023 mixed up two very different things. Tools such as TensorFlow, PyTorch and Keras are programming frameworks that developers use to build and train machine-learning models. Amazon SageMaker, IBM watsonx and Google Cloud's Vertex AI are cloud platforms for companies running AI at scale. They are important, but none of them is a chatbot you can open and talk to. If you want an assistant like ChatGPT, the real alternatives are other consumer AI assistants.
All of the assistants below have a free tier and paid plans with higher limits and newer models. Features and model names change every few months, so compare them on the tasks you actually do rather than on headlines.
| Assistant | Made by | Where you find it | Worth knowing |
|---|---|---|---|
| ChatGPT | OpenAI | chatgpt.com, iPhone, Android, Windows and Mac apps | The most widely used; web search, file uploads, image generation and voice |
| Gemini | gemini.google.com, Android, the Google app, Gmail and Docs for many accounts | Tied into Google services; on many Android phones it replaces the old Google Assistant | |
| Microsoft Copilot | Microsoft | copilot.microsoft.com, Windows 11, Edge, Microsoft 365 apps | Built into Windows and Office; work accounts may have separate data protections |
| Claude | Anthropic | claude.ai, iPhone, Android, Windows and Mac apps | Often used for long documents, writing and coding |
| Meta AI | Meta | WhatsApp, Instagram, Messenger, Facebook and meta.ai | Appears inside apps many people already use |
| Perplexity | Perplexity AI | perplexity.ai and mobile apps | Search-first: answers with numbered source links |
Availability varies by country, age and account type. Check each service's own help pages for current limits.
Built-in AI on your phone and computer
You may not need a separate app at all. Recent iPhones, iPads and Macs include Apple Intelligence writing and summarizing tools, and Apple lets you choose whether some requests can be passed to ChatGPT. Android phones increasingly ship with Gemini, and Windows 11 includes Copilot. Built-in tools are convenient, but each has its own privacy settings, covered below.
How to choose
- Pick two assistants and give both the same three real tasks from your week, such as summarizing a PDF, drafting a reply and answering a factual question.
- Check which one shows sources you can click and verify.
- Read the privacy settings before you commit, especially whether your chats are used for training by default.
- Only pay once a free tier's limits actually get in your way.
Common pitfall: third-party apps in app stores with names or icons that look like a well-known assistant. Download apps from the developer's own site or verified store listing.
How recommendation engines decide what you see
Long before chatbots, AI was already choosing much of what you see online. A recommendation engine predicts what you are most likely to watch, buy, read or click next, and ranks content accordingly. Your YouTube home page, Netflix rows, Amazon's "customers also bought", Spotify playlists and every social media feed are built this way.
How they work, briefly
- Collaborative filtering looks for people whose behavior resembles yours and suggests what they liked.
- Content-based filtering looks at the item itself, such as its topic, genre, length or creator, and finds similar items.
- Signals include what you watch and for how long, what you skip, search, like, share, buy or mark "not interested", plus context such as time of day and device.
Modern systems blend these with machine-learning models trained on huge numbers of interactions. Companies such as Amazon (with Amazon Personalize) and Google Cloud sell recommendation tools so that smaller shops and apps can do the same. Generative AI is now being layered on top, for example to write a summary of reviews or to let you ask a shopping assistant for "a waterproof jacket under a certain price".
Why it matters to you
Recommendations save time, but they optimize for engagement, which is not always what you want. They can narrow what you see, keep you scrolling longer than you planned and show you more of something after a single curious click. The good news is that you have more control than most people use.
Take back control on YouTube
- On a recommendation you don't want, open the More (three-dot) menu and choose Not interested. You can add a reason.
- To stop a creator appearing at all, choose Don't recommend channel from the same menu.
- To stop new viewing shaping your suggestions, pause or clear your watch and search history in your YouTube history settings. Deleting single entries removes their influence.
- On the Home feed, choose Show fewer Shorts from the menu above a Shorts row if you want fewer of them.
- To undo "Not interested" feedback later, go to My Activity at myactivity.google.com, open Other Google activity and delete the YouTube "Not interested" feedback.
Other apps
- Most feeds, including Instagram, TikTok, Facebook and X, have a "Not interested" or "See less" option in each post's menu. Using it consistently works better than ignoring content.
- Many apps offer a chronological or "Following" feed that shows posts from accounts you follow in time order instead of a ranked feed. Look for it at the top of the feed or in the app's feed settings.
- Streaming services usually let you remove titles from your viewing history and create separate profiles, which keeps a child's viewing out of your recommendations.
- In the European Union, the Digital Services Act requires the largest platforms to offer at least one recommendation option that is not based on profiling you.
AI privacy: what regulators found, and the settings to change
In spring 2023, privacy regulators in several countries began asking how ChatGPT had been built and what it did with personal information. Those cases have now reached conclusions, and they explain why today's assistants have the privacy controls they do.
Italy: a temporary block, then a fine
In March 2023, Italy's data protection authority, the Garante, imposed an immediate temporary limitation on ChatGPT in Italy over concerns including the legal basis for collecting data to train the model and the lack of age checks. The restriction was lifted in less than a month after OpenAI agreed to changes. On 20 December 2024 the Garante issued its final decision and fined OpenAI 15 million euros. It found that OpenAI had trained ChatGPT without an adequate legal basis, had not been transparent enough with users, had no adequate age verification, and had not notified the regulator of a March 2023 bug that exposed some ChatGPT Plus subscribers' chat histories and payment information. It also ordered a six-month public information campaign in Italian media. OpenAI said it planned to appeal.
Canada: a joint investigation
The Office of the Privacy Commissioner of Canada opened an inquiry in April 2023 after a complaint that OpenAI had used personal information without consent. In May 2023 it widened this into a broader investigation, joined by the privacy regulators of Quebec, British Columbia and Alberta. Their findings, published on 6 May 2026, concluded that the complaint was well-founded. The regulators found that OpenAI's early models had collected large amounts of personal information without valid consent, that many users did not know their chats could be used for training, that warnings about inaccurate answers were insufficient, and that people did not always have an easy way to access, correct or delete their information.
The federal commissioner treated the matter as conditionally resolved because of commitments OpenAI made, including filtering personal information from training data, publishing more about its data sources, warning signed-out users before their first prompt that chats may be reviewed and used for training, and reporting on progress every quarter. The British Columbia and Alberta offices found the consent issue unresolved.
- 30 Nov 2022
ChatGPT is released to the public. - Mar 2023
Italy's Garante imposes a temporary limitation on ChatGPT; it is lifted within a month after OpenAI agrees to changes. - Apr 2023
Canada's Privacy Commissioner opens an inquiry after a complaint about consent. - May 2023
The inquiry becomes a joint investigation with Quebec, British Columbia and Alberta. - 20 Dec 2024
The Garante fines OpenAI 15 million euros and orders a six-month information campaign. - 6 May 2026
Canadian regulators publish findings: the complaint is well-founded; OpenAI commits to further changes.
What this means for you
The lesson from both cases is simple: assume that what you type into a consumer AI assistant may be stored, reviewed by people and used to improve future models unless you change the settings. Never paste passwords, ID numbers, bank details, health records or other people's private information into a chatbot.
ChatGPT
- On the web, open your account menu and choose Settings > Data controls. Turn off Improve the model for everyone. On iPhone and Android, open the sidebar, tap your profile to open Settings, then Data controls. When you are signed in, the choice applies on all your devices.
- For a one-off sensitive question, start a Temporary Chat. It doesn't appear in your history, doesn't create memories and isn't used to train models, though OpenAI may keep it for up to 30 days for safety.
- Turning off training does not delete past chats. Delete old conversations you no longer need from your chat history.
Google Gemini
- Go to Gemini Apps Activity at myactivity.google.com/product/gemini.
- Turn off Keep Activity if you don't want future chats saved to your account or used to improve Google's AI. Google says chats are still kept for up to 72 hours so it can respond and keep the service safe.
- If you leave it on, activity is deleted after 18 months by default; you can change that to 3 or 36 months. Google notes that chats already reviewed by people can be kept for up to three years, and warns not to enter anything confidential you wouldn't want a reviewer to see.
Microsoft Copilot, Claude, Meta AI and others
Each service has a privacy or data section in its settings that controls chat history and whether conversations are used for model training, and the defaults differ. Open the assistant's settings, search for "privacy" or "data", and read what each toggle does before you change it. If you use an assistant through work or school, your organization's account terms usually apply instead of the consumer ones.
| Assistant | Training or activity setting | Private one-off chats |
|---|---|---|
| ChatGPT | Settings > Data controls > Improve the model for everyone | Temporary Chat (kept up to 30 days for safety) |
| Gemini | Gemini Apps Activity > Keep Activity | Turning Keep Activity off; chats kept up to 72 hours |
| Copilot, Claude, Meta AI | Search Settings for "privacy" or "data" | Varies by service; check its help pages |
Settings names as described on each provider's help pages in 2026. Menus can change; if a label has moved, search the app's settings.
Common pitfall: assuming a browser's private window makes an AI chat private. If you are signed in, the conversation is tied to your account regardless of the browser mode.
Where AI is already part of daily life
Most of the AI that affects you isn't a chatbot. It sits behind services you use without thinking about it. Stanford's 2025 AI Index, an annual independent report, gives a sense of how quickly that has happened: 78% of organizations reported using AI in 2024, up from 55% the year before, and the cost of running a system that performs at the level of GPT-3.5, the model behind the original ChatGPT, fell more than 280-fold between November 2022 and October 2024.
Health care
AI helps radiologists flag possible problems on scans, transcribes doctor-patient conversations into notes and predicts which patients may need extra attention. The U.S. Food and Drug Administration approved 223 AI-enabled medical devices in 2023, up from six in 2015, according to the AI Index. These tools assist clinicians; they do not replace a diagnosis from one.
Transportation
Navigation apps use machine learning to predict traffic and arrival times, and driver-assistance features use it to keep lanes and spot hazards. Fully driverless taxis are now a normal sight in some U.S. cities: the AI Index reported that Waymo was providing more than 150,000 autonomous rides each week.
Banking and payments
Fraud detection is one of the oldest everyday uses. Card networks and banks score transactions in real time and block or flag ones that don't fit your usual pattern. The same techniques help scammers too, so voice clones and AI-written phishing are now common. A call or message that pressures you to act fast is a warning sign, however real it sounds.
Customer service
Chatbots on websites and in messaging apps answer routine questions, track orders and book appointments. If you need a human, asking for "agent" or "representative" usually works, and you should never give a full password or one-time code to any chat window.
Shopping, media and search
Recommendation engines (covered above) choose much of what you see. Search engines increasingly put an AI-written summary at the top of results. Treat those summaries as a starting point and click through to the sources for anything important.
Work and school
Office software now builds in AI for drafting, summarizing meetings and analyzing spreadsheets. Pew found that 28% of employed U.S. adults had used ChatGPT for work in early 2025. Rules vary widely between employers and schools, so check the policy before you use AI on assignments or confidential material.
Quick-start checklist
Do these this weekend, in this order. Together they take under an hour.
| Priority | Task | Time needed | Impact |
|---|---|---|---|
| 1 | Check the training or activity setting in every AI assistant you use | 10 minutes | High: controls what is stored and reused |
| 2 | Delete old chats that contain personal or work details | 10 minutes | High: reduces what is kept on file |
| 3 | Agree a family rule: no passwords, ID numbers or other people's private details in chatbots | 5 minutes | High: prevents the worst leaks |
| 4 | Try two assistants on the same three real tasks and keep the one that cites sources best | 20 minutes | Medium: better answers |
| 5 | Clean up YouTube and social feeds with "Not interested" and chronological feeds | 10 minutes | Medium: less doomscrolling |
| 6 | Bookmark the habit: verify any fact, figure or quote before you reuse it | 1 minute | High: avoids embarrassing errors |
FAQ
Is ChatGPT free?
Yes, there is a free tier, and most other major assistants have one too. Paid plans add higher usage limits, newer models and extra features. Try the free versions on your real tasks before paying for anything.
Can I trust what an AI assistant tells me?
Trust it the way you would a quick answer from a knowledgeable stranger: useful, but worth checking. Assistants can state wrong facts confidently. Prefer tools that show sources, open those sources, and verify anything you will act on or publish.
Does ChatGPT use my conversations for training?
For consumer accounts, it can unless you turn off Improve the model for everyone in Settings > Data controls. Temporary Chats are not used for training. Work and education plans often have different terms set by your organization.
Is it safe to let my kids use AI chatbots?
Check each service's minimum age first; many require users to be at least 13, with parental consent for younger teens. Regulators in Italy specifically criticized the lack of age checks. Sit with younger children while they use it, keep personal details out of chats, and explain that the AI can be wrong.
Will AI replace search engines?
It is changing them rather than replacing them. Search engines now add AI summaries, and AI assistants now search the web. For anything important, the original source still matters more than the summary.
Sources
- Pew Research Center: 34% of U.S. adults have used ChatGPT, about double the share in 2023 (June 2025)
- Office of the Privacy Commissioner of Canada: Summary of joint investigation into OpenAI's ChatGPT (May 2026)
- Office of the Privacy Commissioner of Canada: Commissioner's statement on the joint investigation (May 2026)
- PPC Land: Canadian regulators find ChatGPT privacy rules broken from the start
- Lewis Silkin: OpenAI faces 15 million euro fine as the Italian Garante strikes again (January 2025)
- Seeking Alpha: OpenAI plans to appeal 15 million euro fine by Italian regulator
- OpenAI Help Center: Data controls FAQ
- Google Gemini Apps Privacy Hub
- YouTube Help: Manage your recommendations and search results
- Stanford HAI: The 2025 AI Index Report
- AIhub: Key takeaways from the 2025 AI Index Report
Top image: "Humanoid robot at Science Square Tsukuba" by Syced, CC0 1.0, via Wikimedia Commons (resized). Charts by Thank You Robot.
){kind=link}
0 Comments