AIAIAI.FYI
LIVE

Claims check

Popular AI claims, checked.

Some AI claims are false. Some are half true. Some are real debates among experts. We rate each one on six labels, show what is actually true, and link every source so you can check our work.

  • False1 claim

    The evidence shows the claim is wrong.

  • Misleading4 claims

    Built on something real, but it leads you to a wrong conclusion.

  • Missing context1 claim

    Partly accurate, but it leaves out facts that change what it means.

  • Unproven2 claims

    There is no good evidence for it yet. New evidence could change the label.

  • Real1 claim

    Genuine photos, video or audio that people wrongly called AI-made.

  • Expert debate2 claims

    Qualified experts disagree. This is not a verdict: we show the evidence on each side.

The claims

Misleading

“The internet is dead. Bots and AI replaced real people, and it was done on purpose to control us.”

Bots really are a huge share of web traffic, and AI-made junk is easy to find. But no evidence shows a coordinated plan to replace people online. That part is a conspiracy theory that spread from a 2021 forum post.

What's true

Imperva counted more than 53% of all web traffic in 2025 as automated, up from 51% in 2024, when bots passed humans for the first time in a decade. Not all of it is harmful: search-engine crawlers count too. In 2024, bots Imperva classed as malicious made up 37% of all traffic. Wikipedia's summary notes that this does not prove a coordinated effort.

Sources

Last checked Nov 18, 2026 · Link to this claim

Unproven

“Your phone secretly listens to your conversations to pick the ads you see.”

No one has shown that phones or apps record conversations to target ads. In a 2018 test of more than 17,000 Android apps, Northeastern University researchers found no app sending audio. The closest real case is a sales pitch, and no one has shown it was used.

What's true

In 2024, 404 Media published a Cox Media Group pitch deck. It claimed the company could target ads based on what people said near device microphones. Google removed the company from its Partners Program after 404 Media asked about the deck. The Northeastern team did find some apps sending screenshots and screen recordings to other companies without telling users.

Sources

Last checked Nov 18, 2026 · Link to this claim

Real

“That video of the Palm Beach plane crash is 100% AI.”

The footage is real. PolitiFact traced it to a West Palm Beach man who recorded it on his dash cam, and the AP, CBS News, CNN and the BBC all ran it. PolitiFact rated the AI claim Pants on Fire.

What's true

AI fakes are common enough that doubt is healthy. But calling real footage fake is misinformation too. Tracing a clip back to the person who filmed it settles more than a gut feeling does.

Sources

Last checked Nov 18, 2026 · Link to this claim

False

“This video shows a former James Talarico student accusing him of misconduct.”

The video is an AI fake. The man it was made to look like told Lead Stories it was not him and that he never made the accusation. PolitiFact rated the claim Pants on Fire.

What's true

Coordinated fake campaigns do exist. Clemson University's Media Forensic Hub tied this video to Storm-1516, a Russian disinformation campaign, and Graphika and NewsGuard also reported on it. What separates a real operation from a conspiracy theory is evidence like this.

Sources

Last checked Nov 18, 2026 · Link to this claim

Unproven

“My chatbot has woken up. It's sentient, and it's revealing hidden truths to me.”

No one has shown that any chatbot is sentient. Chatbots write text from patterns they learned, and they tend to agree with what the user already believes. A "revelation" can be your own idea reflected back in confident words.

What's true

The harm is real, even though "AI psychosis" is not a medical diagnosis. OpenAI estimated in October 2025 that 0.07% of ChatGPT's weekly users, about 560,000 people, showed possible signs of psychosis or mania. If a chatbot conversation is pulling you or someone you love away from people, talk to someone you trust or a doctor.

Sources

Last checked Nov 18, 2026 · Link to this claim

Misleading

“Every ChatGPT question uses 10 times the energy of a Google search.”

For a typical question, the "10 times" figure is out of date. It compares a 2023 estimate of about 3 watt-hours per ChatGPT query with a 2009 Google figure of 0.3 watt-hours per search. Epoch AI's 2025 estimate for a typical GPT-4o query is about 0.3 watt-hours, about the same as that Google figure, and OpenAI's Sam Altman put the average query at about 0.34 watt-hours. Newer chips and shorter typical answers explain most of the gap.

What's true

Some queries do use far more. Epoch estimated about 2.5 watt-hours for a question with a very long input, and nearly 40 for an extremely long one. Reasoning models also write more words, so they likely use more. Google's 0.3 watt-hour figure is from 2009 and may be out of date.

Sources

Last checked Nov 18, 2026 · Link to this claim

Misleading

“38,000 ChatGPT queries use the same water as one California almond. (Sam Altman, September 2026)”

PolitiFact rated this Mostly False. The math pairs OpenAI's own low water figure per query, which no one else has confirmed, with an almond figure that counts rain and the water needed to dilute pollution. With direct irrigation water and outside estimates, roughly 1,000 to 10,000 prompts match one almond.

What's true

Farms use far more water than data centers overall. PolitiFact cited about 30 trillion gallons for US crop irrigation in 2020, against about 228 billion gallons for data centers in 2023.

Sources

Last checked Nov 18, 2026 · Link to this claim

Misleading

“AI needs more energy than the entire country produces, maybe almost double. (Donald Trump, July 2026)”

AI's power demand is growing fast, but no forecast PolitiFact reviewed comes close to this. Projections put data centers at roughly 6% to 25% of US electricity, not 100% or 200%. PolitiFact rated the claim Mostly False.

What's true

Demand is rising quickly. The Electric Power Research Institute says data centers could use 9% to 17% of US electricity by 2030, up from 4% to 5% today.

Sources

Last checked Nov 18, 2026 · Link to this claim

Missing context

“"AI has already wiped out jobs." Or the opposite: "AI hasn't affected jobs at all."”

Both versions overstate it. Stanford researchers using ADP payroll data found no sign of economy-wide job losses through June 2026. But young workers in the jobs most exposed to AI are falling behind.

What's true

Workers aged 22 to 25 in AI-exposed jobs are 19% below where they would be if they had kept pace with less-exposed peers. The gap comes mostly from companies hiring fewer young people, not from layoffs. Experienced workers show no similar gap. The authors call these early findings, not proof that AI caused them.

Sources

Last checked Nov 18, 2026 · Link to this claim

Expert debate

“AI could wipe out humanity.”

Qualified experts disagree about this, so we don't rate it true or false. The International AI Safety Report 2026, written by more than 100 experts and led by Yoshua Bengio, says experts' views on loss-of-control scenarios "vary widely." It also says today's systems are not yet highly capable of the behaviors those scenarios would need.

What's true

This is a mainstream scientific question, not a fringe one. Leading researchers and the heads of the biggest AI labs have put their names to it, and thousands of researchers have answered surveys about it.

Why some experts worry

In May 2023, Geoffrey Hinton, Yoshua Bengio and the CEOs of Google DeepMind, OpenAI and Anthropic were among those who signed a one-sentence statement: reducing the risk of extinction from AI should be a global priority, alongside pandemics and nuclear war. In a survey of 2,778 researchers who had published at top AI venues, between 38% and 51% gave at least a 10% chance of outcomes as bad as human extinction. The 2026 safety report says current systems may show early warning signs of evading oversight.

Why others are skeptical

The same survey found that most researchers were very uncertain, and they disagreed on whether faster or slower progress would be better. The safety report says current systems are not yet highly capable of the behaviors loss-of-control scenarios would require. Some researchers, including Timnit Gebru and Emily M. Bender, argue that focusing on speculative future risks pulls attention from harms happening now, such as worker exploitation, data theft and concentrated power.

Sources

Last checked Nov 18, 2026 · Link to this claim

Expert debate

“All this AI extinction talk is just marketing by AI companies to hype their products.”

There is a real argument here, and a real answer to it, so we don't rate it. Some of the loudest warnings come from people who sell AI. Many others who share the concern have no product to sell.

What's true

Both facts hold at once. The CEOs of Google DeepMind, OpenAI and Anthropic signed the 2023 extinction-risk statement while building the systems it warns about. So did academics such as Geoffrey Hinton of the University of Toronto and Yoshua Bengio of the Université de Montréal and Mila.

The case that it's hype

In March 2023, the authors of the well-known "Stochastic Parrots" paper wrote that warnings about powerful digital minds are inflated by hype. They argued the focus should be on what companies are doing now: exploiting workers, taking data, spreading synthetic media and concentrating power. Warnings that AI is world-changing can also make a product sound more powerful.

The case that it isn't

The worry is not limited to company leaders. The statement's signers include university professors, and the 2,778 researchers surveyed had published at top AI venues. Between 38% and 51% of them gave at least a 10% chance of extinction-level outcomes. The International AI Safety Report, overseen by a panel nominated by more than 30 countries and international organizations, includes loss of control among the risks it reviews.

Sources

Last checked Nov 18, 2026 · Link to this claim

How we check claims

  • We pick specific claims that many people repeat and that evidence can settle. We don't rate predictions or opinions.
  • We go to the original: the study, the data, the company's own words, or the person who filmed the clip.
  • When a fact-checker such as PolitiFact has already rated a claim, we say so and link their work. Our labels use our own six-step scale, so the wording can differ.
  • Where qualified experts disagree, we use Expert debate and show the evidence on each side. That label is not a verdict.
  • We use the words “conspiracy theory” only for claims of a secret, coordinated plot made without evidence. Coordinated campaigns are sometimes real, like the Russia-linked fake video above. The test is evidence.
  • Every verdict links the sources it rests on, and we checked each one against them. If you think a label is wrong, tell us at corrections@aiaiai.fyi and we'll look again.
  • If new evidence changes a verdict, we update the label and the “Last checked” date. See our corrections policy.

Page last reviewed Nov 18, 2026. New to AI? Start with AI for Humans.