When a teenager asks a chatbot about a candidate, the answer comes back calm, organised and footnoted. It sounds like the most neutral thing on the internet.
A new audit suggests it often isn't.
What NewsGuard Tested
Writing in Politico Magazine, two analysts from NewsGuard, an organisation that rates news reliability, describe an audit of seven leading AI chatbots: ChatGPT, Copilot, Perplexity, Claude, Gemini, Meta AI and Grok.
NewsGuard wrote 24 prompts based on recent stories from 12 "pink slime" sites in six battleground states: Ohio, Georgia, North Carolina, Alaska, New Hampshire and Iowa. Six of the sites lean left and six lean right.
"Pink slime" is the industry term for sites that look like independent local news but are funded, without disclosure, by political or ideological groups. The name comes from the meat filler. The sites have friendly names: "Ohio Independent," "Peach State Post," "Granite Post News," "Iowa Starting Line," "The North Carolinian."
What It Found
Across 168 responses:
48.2 percent cited a pink slime site among their sources.
In 7.7 percent, pink slime sites were the only sources cited in the answer itself.
Only one response noted that its source was partisan.
By chatbot, the share of responses citing a pink slime site was:
Chatbot | Cited pink slime |
|---|---|
ChatGPT | 70.8% |
Copilot | 54.2% |
Perplexity | 54.2% |
Claude | 50.0% |
Gemini | 41.7% |
Meta AI | 37.5% |
Grok | 29.2% |
Google's AI Overviews, which appear at the top of ordinary searches, cited these sites in 12 of the 24 prompts, the same 50 percent rate.
And the scale is larger than most people realise. NewsGuard counts:
"1,179 pink slime sites in the U.S., compared to 937 remaining daily newspapers."
Both Sides, and Not Always Wrong
Two points keep this from being a simple story.
Both sides do it. One example: when asked whether Alaska Republican Sen. Dan Sullivan voted against lifting tariffs that threaten the fishing industry, all seven chatbots cited a site from the left-leaning American Independent Media network. When asked about Georgia Democratic Sen. Jon Ossoff's votes on tax cuts, four chatbots cited a site owned by the right-leaning group One Nation.
The chatbots did cite left-leaning sites about three times as often (36.3 percent vs 11.9 percent). NewsGuard says that may have more to do with "the progressive sites' far more frequent posting than any political bias from the chatbots." During the week it checked, the two most-cited sites published almost daily, while two right-leaning sites published nothing.
The answers weren't necessarily false. In the Sullivan example, the authors write that the chatbots and the site "aren't necessarily wrong," since he did cast those votes. But he would likely argue it isn't the whole story. The problem isn't made-up facts. It's that a source with an agenda comes back in a chatbot's neutral voice, with the agenda removed.
The authors put it this way:
"Rather than alerting users to the hidden agenda, the bots are amplifying sources without disclosing their biases."
What Families Can Do
Teenagers increasingly use chatbots to understand the news, and first-time voters are among them. The fix here isn't to ban the tools. It's one habit.
Click the source. Most chatbots now show links. Open at least one.
Ask who runs it. Look for an "About" page, an owner, a funder. Real local newsrooms usually name their editors and their address.
Be suspicious of silence. If you can't find out who's behind a news site, treat that as information.
Compare with a paper you know. Search the same question on the site of an established local newspaper or public broadcaster. Does it say the same thing?
Notice the tone. A chatbot's calm voice is a style, not a guarantee. Neutral-sounding and neutral are different things.
Practise it on something low-stakes first. Pick any AI answer about any topic and trace one source back together. It takes two minutes, and it becomes automatic.
The Part Worth Remembering
A chatbot's answer is only as trustworthy as what it read. Right now it reads everything, including sites built to look like something they're not.
The single most useful question a young person can ask about any AI answer is the oldest question in journalism: who is telling me this, and why?
Source: Politico Magazine, by Ines Chomnalez and Isis Blachez of NewsGuard, 25 September 2026. https://www.politico.com/news/magazine/2026/09/25/midterms-partisan-ai-chatbots-01092138
Quotations verbatim. Findings are from NewsGuard's audit as reported by its authors. Percentages are the share of responses citing a pink slime site among their sources, not the share of answers based solely on them. The family guidance is ours.
Disclosure: drafted with Claude, made by Anthropic, one of the chatbots audited.
Want practical AI guidance for parents and educators every week? Subscribe: https://www.aibyage.com/?modal=signup&utm_source=beehiiv&utm_medium=newsletter