What AI can actually do in 2026: a plain-English tour

Somewhere in your life is a person who keeps asking what the AI fuss is actually about. Maybe that person is you. You’ve heard it will change everything and that it’s all hype, that it will take the jobs and that it can’t count the r’s in “strawberry.” What you haven’t heard is a plain answer to the plain question: what does it actually do, for a normal person, on a normal Tuesday?

This post is that answer. No jargon, no doom, no sales pitch: a tour of eight things modern AI does well, each anchored to one concrete example you could try tonight, followed by the two things you genuinely need to know before you trust it, because one of them is that you shouldn’t always trust it. If you finish this page, you know enough to start. If you’re the family’s designated explainer, this is the page to send instead of explaining it again.

Half the country quietly started using it

First, the number that reframes the whole conversation. In Pew Research’s June 2026 survey, 49% of US adults said they use AI chatbots, up from 33% in 2024, and about a quarter of the country now uses one daily. ChatGPT alone passed 900 million weekly users in February 2026. Whatever this is, it stopped being a tech-industry curiosity a while ago.

More telling is what people do with it. When researchers at OpenAI and Harvard analyzed real usage for a 2025 NBER study, more than 70% of messages had nothing to do with work. The three biggest categories were practical guidance, looking things up, and help with writing: everyday life, in other words. Nobody took a course first. They typed a question the way they’d ask a knowledgeable friend, and that turned out to be the entire manual. Keep that in mind through everything below: if you can describe what you want, you already know how to use this.

It writes the email and reads the fine print

Start with the ability that half the world discovered first: it writes. Not “writes” in the school-essay sense, but in the life-admin sense. The complaint to the airline. The message to the landlord that needs to be firm without burning the relationship. The condolence note you’ve rewritten four times because every version sounds wrong. Tell it who the message is for, what happened, and the tone you want, and it hands you a competent draft in seconds. You then make it yours: cut the stiff sentence, add the detail only you know, and send. It’s a draft machine, not a signature machine, and used that way it removes the dread from a whole category of chores. If the first draft misses, don’t start over. Reply the way you’d brief a colleague: “shorter, warmer, and don’t apologize so much.” The second version is usually the keeper, and learning that back-and-forth on a landlord email teaches you the skill everything else on this page runs on.

The mirror image is less famous and, for many people, more useful: it reads. Paste in the 40-page lease, the insurance policy, the school district’s new policy letter, and ask for the plain-English version. Then interrogate it: What’s unusual here? What am I agreeing to that most leases don’t include? What happens if I cancel in month three? The unread pile on your counter is full of documents someone else wrote to be skimmed past. A tool that reads them carefully on your behalf, and answers follow-up questions without sighing, is quietly one of the most practical things AI does.

It can look at what you photograph and talk out loud

The part that still surprises people: the newer AI tools have eyes and ears. Photograph your fridge shelf and ask what you can cook in 30 minutes; our meal-planning guide runs that exact play for a full week of dinners. Photograph the houseplant with the yellowing leaves, the error message on the washing machine, the form that arrived in a language you don’t read. Then ask questions about what’s in the frame, in ordinary words, and get ordinary answers back.

This is not a party trick. For people who are blind or have low vision, apps like Be My Eyes now use the same image-describing AI to read labels, decipher menus, and describe unfamiliar rooms on demand. The same ability that suggests dinner from a photo gives someone their groceries’ labels back.

Voice crossed a similar line. You can hold a spoken conversation with an AI assistant at something close to human pace, which matters for everyone who thinks better out loud than in a text box. And live translation went mainstream: Apple’s Live Translation now sits inside Messages, FaceTime, and phone calls, translating as you talk, with AirPods speaking the translation into your ear. If you’d rather learn the language than lean on the earpiece, AI works as a tireless conversation partner too; we wrote a full guide to that. One term worth knowing so the marketing doesn’t intimidate you: tools that handle text, images, and sound together are called multimodal, and our short explainer covers what that means. That’s the whole secret.

It draws the invitation and films the 30-second clip

AI also makes pictures from descriptions, and this has already gone thoroughly mainstream: in the same Pew survey, 24% of US adults said they use chatbots to create or edit images or videos. The recipe is plainer than the results suggest. Describe the subject, the style, and the mood: “a birthday invitation for a dinosaur-obsessed five-year-old, bright and cheerful, with room for the date at the bottom.” Ten seconds later you have something that would have taken an hour in a design app, and if it’s not right, you say what to change and try again. People use this for invitations, flyers for the yard sale, a rough logo to react to, a picture of the kitchen repainted in the blue they’re debating. The describing skill transfers directly from the writing sections above: say what you want, look at what you got, say what to change. One honest caveat before you print anything for a business: the rules on using generated images commercially vary by tool, so check the terms of whichever one you pick.

Video is real too, with honest limits. Type a sentence, get a short clip with sound: convincing for product shots, mood pieces, and social posts, and still prone to physics glitches and mangled on-screen text when you push it. Clips run seconds, not minutes, and good results take retries. We ran the current video models head to head with real clips and real costs if you want receipts rather than promises. The fair one-line summary of both abilities: the first attempt is a draft, the tenth minute of trying is where the good results live, and hands, text, and logos are still where the illusion slips.

It plans the trip, the dinners, and the fractions homework

The quietest big win is delegation of thinking-work: the planning, sorting, and list-making that fills evenings. Give it your dates, budget, and the fact that you hate museums before noon, and it drafts a day-by-day itinerary you then argue with until it’s yours. We planned eight days in Japan this way, including the arithmetic mistake it made until we checked its math. The same move plans the week of dinners around what’s on sale, drafts the packing list, untangles the family calendar, and turns “we should really deal with the garage” into a Saturday plan with steps. Parents in particular are using it as a second brain for the household’s mental load; that post covers where the line sits between help and homework-doing.

And it teaches. This might be the most underrated ability on the page: it is an endlessly patient tutor that explains fractions five different ways at 9 pm without a flicker of frustration, and tutoring consistently ranks among the top uses in the NBER data. The trick is one sentence: say “don’t give me the answer, help me understand it.” That single instruction turns an answer machine into a teacher, for your kid’s math homework and for the thing you always meant to learn yourself. Ask it to explain compound interest with your actual numbers, quiz you on the driving-test rules, or walk you through what a word in your medical results means in plain English (and then confirm anything that matters with your doctor, for reasons the next section makes clear).

It makes things up, and knowing that is half the skill

Now the honest paragraph this whole page has been building toward. AI sometimes states things that are flatly false, in the same confident tone it uses when it’s right. It will invent a book title, a statistic, a court case, a quote, complete with convincing detail. This isn’t lying, and it isn’t a bug that next month’s update fixes. These systems work by predicting what plausible text looks like, and most of the time plausible and true point the same way. When they don’t, you get fluent nonsense, and nothing about the wording warns you. Our deeper explainer covers why, but the one-paragraph version is enough to protect you.

The stakes are not hypothetical. A public database maintained by a legal-data analyst has now logged nearly 1,600 court cases worldwide in which AI-fabricated citations or content made it into legal filings, with the count more than doubling in the first half of 2026. Those are trained professionals getting burned by trusting fluent output. The household rule that follows is simple: anything that matters, verify. Use AI freely for drafts, ideas, summaries, and explanations. The moment an answer would change what you sign, spend, take as medicine, or repeat as fact, check it against a source that isn’t the AI. People who internalize that one rule get almost all of the benefit and almost none of the embarrassment.

Where your words go, and where to start

One more thing to know, in one breath. When you type into a free AI chatbot, your words travel to the company’s computers, are processed there, and may be kept under policies you’ve never read. Pew found 71% of Americans expect AI to erode the security of personal information, and the sensible response isn’t fear, it’s a habit: don’t paste anything you wouldn’t email to a helpful stranger. Medical records, financial documents, and other people’s secrets stay out, or get their names and numbers covered first. It also helps to know that AI which runs entirely on your own computer exists and is getting good fast; Apple runs its live translation on the device for exactly this reason, and we’ve written about that shift for when you’re curious. For your first experiments, the stranger rule is all you need.

So, where to start? Not with the year’s best model or the perfect prompt. Start with a chore you already have, small enough that a wrong answer costs nothing:

Here’s the tour in three sentences. AI in 2026 is eight everyday abilities: it writes, reads, sees, speaks, draws, films, organizes, and teaches, and the price of entry is describing what you want in plain words, which you have been doing your whole life. It makes things up sometimes, so anything that matters gets verified, and anything private stays private. Everything else on this site goes deeper on some corner of that map, but the map itself now fits on one page, and you’ve just read it.