Before You Repeat It

A video turns up on your feed. Something happened — a shooting, a collapse, a politician saying the unsayable. It looks real. Before you decide whether it is, understand what is actually being asked of you: not your belief. Your reach.

Nobody manufacturing a fake cares whether one person is privately convinced. A lie that stops with you has failed. It needs repeating, and you are the mechanism — which means the useful question is not do I believe this but am I willing to put my name behind it. That is a lower bar, it is answerable, and it is the one that stops the damage.

Why "just Google it" makes it worse

This is the advice everyone gives and it is the advice that backfires. Not sometimes — measurably, on average, across five experiments.

Researchers at NYU and Stanford ran the study and published it in Nature: "across five experiments, we present consistent evidence that online search to evaluate the truthfulness of false news articles actually increases the probability of believing them"[1]. Searching made people 19 to 22% more likely to rate a false story as true.

Their explanation is that searchers fall into "data voids, or informational spaces in which there is corroborating evidence from low-quality sources." And the number that makes it concrete: 38% of people searching a false article landed on unreliable sources, against 15% for a true one.

The reason is simpler and nastier than it sounds

Think about what you actually type. You saw a claim that John Smith died in a car accident, so you search did John Smith die in a car accident.

Those words, in that order, do not appear on any real news site — because it never happened. No newsroom wrote that sentence. So the only pages in the world containing it are the pages that invented it. There is no competition for the phrase, which means the fabrication ranks first, and it ranks first instantly.

You did not fail to find the truth. You searched the fake story's keyword. The specific wording of these stories is not incidental — it is the distribution method. It is SEO, and you are the traffic.

And filling that void is now nearly free

It used to take effort to stand up something that looked like a news site. It doesn't.

NewsGuard tracks sites it classes as AI content farms — and as of June 2026 the count is 3,749 of them, across 16 languages[3]. Their criteria describe exactly what you'd meet in a search result: "the site is presented in a way that an average reader could assume that its content is produced by human writers or journalists, because the site has a layout, generic or benign name, or other content typical to news and information websites" — published "without significant human oversight" and with no disclosure that any of it is machine-written.

So a result that looks like a news site is no longer evidence that a newsroom exists behind it. That is the single fact that broke "look it up," and it is why the rest of this article is about something else.

"Isn't there an app for this?"

No. Not one you can use, and it is better to know that than to trust a number a website hands you.

There is no tool available to the public that can tell you whether a YouTube video or a photograph was made by AI. That is a strong claim, so here is what it rests on.

The detectors disagree with each other. NewsGuard tested five of the leading ones on 45 images in spring 2026. They "collectively declared authentic images to be AI-generated 13.33 percent of the time, and one tool got it wrong 40 percent of the time" — and "in 35 of the 45 images tested, at least one tool reached a different verdict than the rest"[4]. On four images in five, the tools could not agree.

They collapse on exactly the files you have. The GenImage benchmark found detectors scoring above 98.5% on the generator they were trained on, but around 51% — a coin toss — on JPEG-compressed images[5]. Everything you see online has been compressed. The lab number and the number that applies to you are not the same number.

Video is worse, and the detection industry says so. One security vendor lists its peers' lab claims — 96%, 98%, 95–98% — and then concedes that in deployment "state-of-the-art detection systems dropped 45–50% in performance," because H.264 compression "creates particularly troublesome artifacts" that detectors misread as tampering. Their own conclusion: detection "cannot be your primary defense strategy"[6].

And they go stale in months. The Alan Turing Institute built a detector scoring above 99.8%, then tested it forward: it lost "over 30% recall when evaluated on deepfakes created with generation techniques from just six months later"[7].

Which is why the rest of this article is about corroboration rather than analysis. Not because looking closer is beneath you — because there is nothing to look at that will settle it, and a tool that answers confidently 13% of the time when it shouldn't is worse than no tool at all.

Why the standard advice is useless to you

Check the domain registration date. Look at the About page. See whether the site has been hijacked. Inspect the WHOIS record.

Ordinary readers cannot do this, and they should not have to. It is a technical audit wearing the clothes of common sense — and thousands of content farms are built specifically to survive the glance it produces. A recent registration date proves nothing; plenty of real outlets moved domains. An old one proves less; expired domains are bought precisely for their age.

There is a substitute that takes two seconds and needs no skills at all.

What to do instead

1. Search the event, not the caption. Not the sentence you were handed — the thing it claims happened, in your own words, with a place and a date. You are trying to find the reporting, not the claim.

2. Go to outlets directly, and go to more than one. Open the news organisations you already know and look for it there. A genuinely large story is covered independently by many large organisations, and that redundancy is the whole point: no one of them has to be trusted, because they did not coordinate.

3. One outlet is a lead. Several unrelated ones is corroboration. None is your answer. A major event that exists in exactly one place is not a scoop you found early. It is the shape a fabrication makes.

4. Treat social media as the thing you're checking, not the place you check it. It is where the claim came from. Fake accounts, recycled clips and coordinated posting are cheap there in a way they are not at an organisation with a name to lose.

5. If you cannot confirm it, don't pass it on. Not sharing costs you nothing. There is no penalty anywhere for being late to a true story, and no reward for being early to a false one.

The part that is actually being aimed at you

Everything above assumes you are trying to work out whether something is true. Often that is not the transaction taking place.

This is the age of social engineering used as a weapon — people going online with the specific goal of getting other people to do things. And the most reliable way to get somebody to spread something without checking it is to make them angry. Anger is not a side effect of these stories. It is the delivery system. An outraged person shares immediately, adds their own endorsement, and does not pause at any of the five steps above.

There have always been people like this. Propagandists, agitators, rumour-mongers — none of it is new. What is new is the reach. A person who once needed a printing press, a broadcast licence or a crowd now needs an account and an afternoon. The character has not changed; the distribution has. That is a claim about infrastructure, not about human nature, and it is why this is worth taking seriously without treating it as the end of the world.

Don't be used. Be informed.

Nobody can tell by looking any more, and that is fine — you were never going to win that contest, and you do not have to enter it. The question was never whether you can spot a fake. It is whether you will lend it your name.

You have every right to make up your own mind. Making it up well means knowing that fabrications are out there, that they are engineered to be found, and that the search box is one of the places they are waiting for you. It does not mean believing less. It means checking differently — going to the people who put their name on their reporting, checking more than one of them, and staying quiet when they are all silent.

Being slow is not a failure. Being used is.