The web has quietly filled up with websites that look like local papers or niche news outlets but are mostly written by a language model and published with little or no human review. NewsGuard, which has tracked the phenomenon since 2023, counts thousands of these "AI content farm" sites across more than a dozen languages, and the number climbs every month. Most exist to run programmatic ads against cheap, high-volume text; some exist to launder misinformation into something that reads like reporting.
You don't need a detector to catch most of them. The giveaways are structural β who's behind the site, whether the bylines are real, how the articles are sourced β and they hold up far better than any automated score. Here's the sequence I run when a site I've never heard of shows up in a search result or a group chat.
The 7-step check
Read the About page and the masthead first
A real newsroom tells you who runs it. Look for named editors, a physical address, an ownership statement, a corrections policy. Content farms skip all of this or fill it with boilerplate β "a team of passionate writers dedicated to bringing you the latest" with no actual names. If the About page could describe any website on earth, that's your first flag. NewsGuard's own definition of an AI content farm hinges partly on this: the layout is built to look like human journalism while hiding that almost no human is involved.
Click a byline and try to find the reporter
Pick any article, click the author's name, and search for them elsewhere. A working journalist usually has a trail β other bylines, a LinkedIn, past corrections, a photo that survives a reverse image search. Content farms often invent authors with generic names and AI-generated headshots (run the headshot through Google Lens; a face that appears on stock-photo and GAN-portrait pages is a giveaway). If every "writer" on the site is a dead end, you're not reading a newsroom.
Look for leftover machine text
The crudest content farms publish straight from a chatbot without editing, and the seams show. Search the site for phrases that no editor would leave in: "as an AI language model," "I cannot fulfill that request," "as of my last knowledge update," "here is a 600-word article on." Newer models have mostly been trained out of the obvious openers, so also watch for the subtler residue β placeholder brackets like [Insert City], repeated boilerplate transitions, or a date that's permanently "recent" with no real timestamp.
Check how old the domain is and who owns it
Most AI content farms are weeks or months old, not years. Paste the domain into a WHOIS lookup (whois.com, or ICANN Lookup) and check the registration date. A site claiming to be an established regional paper but registered last spring, behind a privacy shield, with no corporate filing anywhere, is wearing a costume. Fresh registration isn't proof on its own β plenty of good new outlets exist β but paired with anonymous ownership it moves the needle hard.
Read three articles back to back
Sameness is the tell a detector can't fully capture. Content farms churn out dozens of pieces a day, so open three and notice the rhythm: identical structure, the same hedging padding ("it is worth considering that"), confident summaries with no named sources, no quotes from real people, no reporting that required leaving a desk. Human coverage of the same story will cite a specific official, a document, a place. If every article is a smooth, source-free summary of other people's work, you've found a mill.
Cross-check one concrete claim it makes
Pull the single most checkable fact from an article β a statistic, a quote, an event β and verify it independently. This is where FAXTR's check tool earns its place: search the claim and see whether established fact-checkers have already addressed it, and whether reputable outlets reported the same thing. AI farms frequently rephrase real news, but they also hallucinate, mangle numbers, and recycle debunked claims. One failed cross-check tells you not to trust the rest.
Run the text through an AI detector β but treat it as a hint
As a last pass, paste a few paragraphs into an AI-text detector. A high "AI-generated" score adds weight to everything above, but never lean on it alone: detectors throw false positives on plain, well-edited human writing and can be fooled by light paraphrasing. NewsGuard pairs its analysts with an automated detector precisely because neither is reliable by itself. Use the score to confirm a suspicion the other six steps already raised, not to start one.
The six tells at a glance
| Red flag | Why it matters |
|---|---|
| No named editors or address | Real outlets are accountable to someone; farms hide who's responsible. |
| Authors you can't find anywhere | Invented bylines, often with AI-generated or stock headshots. |
| Leftover chatbot phrasing | Unedited model output published straight to the page. |
| Brand-new, privacy-shielded domain | Thrown up fast and cheap, with ownership deliberately obscured. |
| Dozens of source-free articles a day | Volume no human desk could produce or verify. |
| Claims that fail a cross-check | Rephrased, hallucinated, or recycled misinformation. |
What this check is not
A news site using AI somewhere in its pipeline is not automatically a content farm. Established outlets now use models to draft summaries, translate wires, and transcribe audio, usually with an editor signing off. The line that matters isn't "did a machine touch this," it's "is anyone accountable for whether it's true." A site can pass every tell above and still be thin; a site can fail one and still be a real newsroom having a bad week. Treat the checklist as a way to calibrate trust, not to hand down a verdict β and keep your conclusion to your own sharing decision rather than publicly branding a site.
The flip side is just as important: a clean AI-detector score proves nothing. These sites are specifically built to look like human journalism, and the better-funded ones edit enough to slip past both detectors and a quick glance. That's exactly why the structural checks β ownership, bylines, sourcing β do more work than any single tool.
When it matters most
Run the full sequence when a story is breaking, emotionally charged, or being used to make a decision β what to buy, who to vote for, whether to panic. Content farms move fastest in exactly those windows, because that's when people share before they check. If the only outlets carrying a dramatic claim are sites you can't trace to a real newsroom, that absence is the story.
Cross-check a claim in seconds
FAXTR searches 100+ fact-checking organizations in one query β free, no login required. Pull the most checkable fact from a suspicious site and see who else has reported it.
Sources: NewsGuard, AI Tracking Center and Rise of the Newsbots; Wikipedia, Content farm. Published 2026-10-02.