Bias in the Media: What We Actually Find in Real Articles
July 2026 — based on production data from BiasChecker.ai
Most writing about media bias is theoretical: definitions, taxonomies, hypothetical examples. We're in an unusual position to do better. BiasChecker.ai has now run hundreds of AI bias analyses over real news articles — from the BBC, CNN, NOS, Fox News, USA Today, Al Jazeera and many others — and every analysis records exactly which bias categories it found, with cited evidence from the text.
So instead of telling you what bias could look like, here is what we actually find, how often — and how to spot each one yourself, without any tools.
The seven biases we find most often
Share of analysed news articles where at least one AI model flagged the category (223 unique articles, 746 analyses, July 2026):
Framing bias
79% of articlesThe same facts, packaged to lead you to one interpretation — word choice, emphasis, and context doing quiet editorial work.
How to spot it: Swap the charged words for neutral ones ("slams" → "criticises", "chaos" → "disagreement"). If the story suddenly feels smaller, the frame was doing the talking.
Emotional bias
57% of articlesLanguage chosen to make you feel — outrage, fear, sympathy — rather than to inform.
How to spot it: Notice what you feel after the first two paragraphs. If you’re angry or alarmed before you’ve learned a single verifiable fact, the text is working on your emotions.
Selection bias
45% of articlesThe slant is in what got covered: which facts, which events, which voices made the cut.
How to spot it: Ask what an editor with the opposite view would have led with. If you can name it and it’s absent, you’ve found the selection.
Source bias
43% of articlesWho gets to speak: quotes stacked on one side, officials treated as neutral, critics anonymous or absent.
How to spot it: Count the named sources for each side of the dispute. A 4-to-1 quote ratio tells you more than any single quote does.
Omission bias
33% of articlesThe most invisible one: key context, caveats, or counterarguments simply left out.
How to spot it: Hardest to catch by definition — you can’t see what isn’t there. Ask: what would I need to know to check this claim? Is it in the article?
Narrative bias
31% of articlesFacts arranged into a story arc — heroes, villains, momentum — because stories are more compelling than information.
How to spot it: If the article reads like it has a protagonist, ask whether events actually happened that neatly, or whether the arc was fitted afterwards.
Political bias
25% of articlesThe classic partisan slant — and notably, only seventh place. Most bias we find is structural, not partisan.
How to spot it: Check whether policy positions are described in their proponents’ terms or their opponents’ terms. Consistent asymmetry is the tell.
Full definitions for every category are in our bias types guide.
How serious is it, typically?
Finding a bias doesn't make an article propaganda — framing and emphasis are part of all writing. What matters is degree. Across our analyses, the overall verdict breaks down like this:
25%
no notable bias
10%
low concern
43%
medium concern
21%
high concern
Two readings of the same numbers, both true: a quarter of mainstream news articles come back clean — and roughly two in three carry enough slant that a careful reader should notice it. The point isn't cynicism about the press; it's that "published by a reputable outlet" and "free of slant" are different claims.
The pattern worth noticing
The most common biases are structural, not partisan. Explicit political slant appears in a quarter of articles — but framing, emotional language, and selective coverage appear in half to four-fifths. That inverts how most people scan for bias: we look for the other team's colours, while the mechanisms that actually shape our understanding are quieter and present almost everywhere, including in outlets we trust.
This is also why bias is hard to see in sources you agree with. If a frame matches your expectations, it doesn't register as a frame at all — it just reads as "accurate". The categories above are worth checking especially when an article feels obviously right.
A practical reading routine
Five habits, one per top finding — each takes seconds:
- De-frame the headline: restate it in the most boring words possible before reading on.
- Check your pulse: if you feel something strong by paragraph two, ask what fact caused it. Often there isn't one.
- Name the missing editor: what would the opposite outlet have led with?
- Count the voices: named sources per side, in ten seconds.
- List what you'd need to verify: if the article doesn't contain it, treat the conclusion as unproven.
About these numbers
Based on 746 bias analyses covering 223 unique news articles, run through BiasChecker.ai's bias lens (29 categories, findings cited to the text) as of July 2026. The corpus is articles our users chose to analyse — mostly mainstream English-language and Dutch outlets — not a random sample of all media. Many articles were analysed by several AI models; a category counts for an article when at least one model flagged it. AI analyses are a critical-reading aid, not ground truth: models have their own biases and disagree at the margins, which is why every finding cites the passage it rests on, so you can judge it yourself.
Check the next article you read
The habits above work unaided — but you can also just ask. Paste any article into our free analyser and get the same 29-category bias analysis these numbers come from, with every finding anchored to the exact passage. Or join the waitlist for the Chrome extension and read with it, side by side, on any page.