How to spot AI-generated fake news
Synthetic photos, cloned voices and fluent nonsense now travel as fast as reporting. A practical way to tell a generated claim from a reported one — without becoming a forensic lab.
A decade ago, fake news usually meant a badly written site with a stolen masthead, or a real photo given a new caption. You could often feel the cheapness. That advantage is gone. Image models produce faces and streets that look like places. Language models produce paragraphs that sound like desks. Audio models can borrow a voice. The problem is no longer that the fake looks amateur. The problem is that it looks finished.
You do not need to become a forensic analyst. You need a habit that is cheaper than panic and more reliable than "it looks real to me." How to spot AI-generated fake news is mostly the same skill as tracing a story back to whoever reported it first, with a few new tells layered on.
How to spot AI-generated fake news
Start with the claim, not the pixels. A generated image attached to a true event is a different problem from a generated event attached to a real-looking image. Most of the damage is in the second one.
Ask who reported it, not who posted it. A newsroom with a name, a dateline, a reporter and a page you can open is doing one kind of work. An account with a dramatic still and no outlet is doing another. If the first place you saw the story cannot name the desk that gathered it, you do not have news yet. You have a file.
Look for a second newsroom. One dramatic clip is a rumor. Two independent desks describing the same event, with their own pictures or their own quotes, is a story. When newsrooms in several countries cover the same event, they disagree about emphasis. They rarely invent the event itself. A claim that exists in only one place, in only one medium, is the one to slow down on.
Separate the picture from the caption. Generated images are often married to a real headline, or a real image is married to a generated caption. Reverse the pair. Does the same picture appear earlier, on a different day, in a different city? Does the caption name a place the background cannot support?
Prefer the publisher's own page. A screenshot of a homepage is trivial to fake. A live article on the outlet's domain, with a URL that matches, is harder. If someone is asking you to believe a paper said something, go to the paper.
That last step is why an aggregator that only shows you a card and never sends you to the source is a poor tool for this job. You need the original page.
What still gives a generated image away
The tells move. Anything published as a definitive list of "AI artifacts" will be stale in a year. A few families of mistakes are still worth a look, with the caveat that they are clues, not verdicts.
Text inside the picture. Signs, headlines, badges, license plates, newspapers held in a photo. Generated type still often collapses into almost-letters. If a protest sign is unreadable in a way that is not just motion blur, be suspicious of the still.
Hands, teeth, jewelry, crowds. Extra fingers are less common than they were. What remains is a softness, a symmetry, or a crowd in which no face will resolve into a person. Zoom. A real crowd is full of ugly specific detail.
Light that does not agree with itself. Two suns, a shadow that falls the wrong way, a reflection that shows a different room. Physics is still expensive for a model to keep consistent across a whole frame.
A place that is almost a place. Generated streets often look like a memory of a city rather than a block you could walk. Shop names that are not names. Architecture that mixes two continents. If the caption says "downtown Lagos" or "a courthouse in Ohio" and the building is a mash of both, you are not looking at a wire photo.
Too much finish. Real news photography is often slightly wrong: a bad crop, a blown highlight, a reporter's sleeve. A perfect, glossy, emotionally on-the-nose still of a tragedy is not proof of fakery. It is a reason to ask who the photographer was.
None of these survive contact with a better model. That is why the source check comes first. The pixels are the appeal. The provenance is the evidence.
Audio, video and the fluent paragraph
Pictures are not the only surface.
Voice. A short clip of a public figure saying something explosive is now cheap to attempt. Ask for the original appearance: a full interview, a podium, a broadcast with more than ten seconds of context. A six-second isolated line with dramatic music is a format that favors the fake.
Video. Look at the cut, not just the face. Generated or face-swapped video still struggles with ears, hairlines, glasses, and the moment a head turns. More useful than staring at the mouth: does any established newsroom have the same pictures from another angle?
Text. Language models produce confident, even copy. The tell is rarely grammar. It is specifics that do not exist: a law with the wrong name, a hospital that is not in that city, a quote no desk can place, a casualty figure that appears nowhere else. Search the distinctive phrase. If the only hits are copies of the same paragraph, you have a chain of repetition, not confirmation.
Homepages and documents. Entire fake articles in the costume of a known masthead circulate as images. Check the live site. Check the URL. A PDF of a "leaked memo" with perfect letterhead and no originating organization is a prop.
The common move in all of these is the same. Do not argue with the artifact. Leave it and look for the reporting.
A five-minute checklist
You will not do a full investigation on a train. You can do this.
- Name the source. Which newsroom, which reporter, which date. If you cannot fill those in, stop.
- Open the original page. Not the screenshot. Not the quote-tweet. The page.
- Find one more desk. A second independent account of the same event. If none exists an hour later on a supposedly huge story, wait.
- Uncouple media from words. Search the image, or a distinctive spoken line, without the caption. See what else it has been.
- Decide what you are willing to pass on. If you cannot explain the claim without the file, you are not ready to share it.
Waiting is allowed. The most expensive shares happen in the first twenty minutes, when the generated version is the only version.
What this has to do with how you read
AI did not invent credulity. It made credulity faster and better looking. The defense is not a new app that "detects AI" with a magic score. Those tools exist, they help sometimes, and they are wrong often enough that you should not outsource the whole job to them. The defense is the old one, applied on purpose: newsrooms, names, second sources, original pages.
A feed that mixes reporting, opinion, memes and synthetic media in the same gesture trains you to give them the same half-second. A card that shows the newsroom, then sends you to the publisher's own site, at least keeps the costume off. You can still be fooled. You are harder to fool in bulk.
How GetMyNews treats the problem
GetMyNews does not claim to detect generated media. No honest product should. What it can do is refuse to hide the source.
Every card names the newsroom. The app reads 1,200 public feeds from 22 countries — by default the English-language ones, 359 feeds across 14 countries — and it does not rewrite the article. Tap through and the publisher's page opens in your phone's browser, with your session. If a paper issued a correction, you will see it there, not in a copy we invented.
Clustering helps the second-source check: when several outlets cover the same event, they sit on one card, so you can see who else is reporting it. That is not proof. It is a faster way to notice a claim that has only one parent.
There is no account and no GetMyNews server. Personalization runs on the device. The app is free, funded by one full-screen ad every twenty stories, and it is in review with Apple and Google. We will not tell you to download it today.
If you want a reader that keeps the newsroom on the card and the article on the publisher's site, that is the design. The rest of spotting AI-generated fake news is still your job. It should be. A machine that sorted truth from invention for you would be another thing you would have to check.
Related: how to follow world news without drowning in apps for the habit of widening the lens, and why staying informed still matters for what the effort is actually for.