AI-generated content

You search for something ordinary — how long to roast a chicken, which of two car seats is better, what a word means in a school letter — and the results read as though they were written to be found rather than to be read. The reviews under a product all have the same shape and the same enthusiasm. A clip arrives in a group chat showing something that would be significant news if it happened. None of it announces itself. You simply notice that more of your day now goes on deciding whether things are real.

What makes this an objection rather than a mood is the price. Producing a paragraph that reads convincingly, or a photograph of something that never happened, used to cost a person's time and some amount of skill. That cost was the only thing limiting how much of it existed — not taste, not rules, not anyone's policy. It has now fallen close to nothing. Anything whose price falls that far arrives in volume, and that is not a claim about this technology in particular. It is how supply behaves.

The worry also gets waved away with history, and it should not be. The web has had junk since it had pages: content farms built for search engines, chain emails, reviews written by people who never owned the thing. That record usually gets deployed to say nothing has changed. People being wrong about the details of an earlier wave is not evidence that they are wrong now, and each of those waves did change what the web was good for. A parent saying the feel of the thing has changed is describing something, and “this existed before” is not an answer to “there is much more of it now”.

Generated text, images, and video are cheap and abundant. Someone with no particular skill can produce in an afternoon what would once have taken a small team, and can do it again tomorrow, and again the day after. Nothing about that is contested by the people building these systems or by the people criticizing them. It is the thing both sides agree on before they start arguing.

Low-quality generated pages are made specifically to capture search traffic. This is a business rather than an accident: a page exists to rank for a question people ask, to carry advertising, and to be replaced once it stops working. The model long predates these tools. What the tools changed is the cost of running it at a scale that used to require hiring people, which means the pages between you and a real answer are no longer limited by anyone's payroll.

Fabricated reviews and engagement-farming images circulate widely. Reviews get written to order for products the writer has never held. Images get built to be shocking, or heartbreaking, or impressive enough that a thumb stops moving, and the reaction is the thing being sold. Neither of these is unusual now, and neither is aimed only at people who are careless with technology. They work on readers who consider themselves careful, which is most of the point.

Voice cloning from a short sample is practical. A few seconds of recorded speech is enough — a voicemail greeting, a clip posted publicly, a video from a school concert that a hundred other families also have. What comes out is not a perfect copy of a person. It does not need to be. It needs to be good enough for a phone line and a listener who is frightened.

It is used in impersonation scams aimed at families. The call sounds like someone you know, it concerns money or a pickup or an emergency, and it is built so that stopping to check feels like a cruelty. That combination — a familiar voice plus manufactured urgency — is the whole mechanism, and it works on people who would never open a badly written email. This is the part of the subject that is worth knowing about beforehand rather than afterwards.

And the volume makes verification harder for everyone, not only for children. An adult who checks things carefully now has more things to check, and each check takes longer than it did. This is not a children's problem with an adult solution standing ready beside it. The person who is supposed to be the backstop is working in the same conditions, with the same amount of time, against more material than before.

Whether provenance standards get adopted widely enough to matter is unknown, and it is not mainly a technical question. The pieces exist. Adoption is a coordination problem across camera makers, phone software, editing tools and everywhere files get passed around, and there is no mechanism that makes all of them do it. A standard that half the chain honors may end up worse than none at all, because a record that is present in some places and missing in others is easy to read backwards. Which way that lands is not knowable from here.

What happens to the usefulness of the open web is a genuine question rather than a dramatic way of putting one. It worked because pages written by people for people were the bulk of what existed and search could sort them. What it becomes when the cheapest thing to produce is a page shaped like the answer you wanted is not clear. One expectation is a retreat into smaller, closed, paid or vouched-for spaces. Another is that the sorting adapts, the way it has before. The second has been right in the past, which is not the same as being right about this.

Whether a generation that grows up assuming media is synthetic ends up better calibrated or simply cynical is the one that keeps people awake. There is a version where children who never believed a photograph was proof develop a steadier relationship with evidence than their parents ever had. There is another where the lesson that lands is that nothing can be known, which is the more comfortable conclusion and the one that serves everybody who benefits from doubt. Which of those arrives depends on things nobody is steering, and it will not be visible for years.

A parent who looks at all of this and concludes that the information environment has got worse is reading it accurately on the evidence available. Deciding to limit what your family is exposed to — fewer feeds, fewer forwards, more things read from places with a name attached — follows from that reading rather than avoiding it. It is a legitimate conclusion and not a retreat from the question. It costs something, and that should be said too: some of what you filter out would have been worth seeing. The trade is real, it is yours to make, and nothing here is settled enough for anyone to tell you that you have made it wrongly.