Stop Freaking Out About AI - Understand the Facts Instead of the Frenzy

Have you heard AI is going to end the world and spill all your business like that loudmouth hussy you went to high school with?

Yeah, me too. And I bought it just like everyone else at first, fearing that AI was leading us towards SkyNet machine overlords like in the Terminator movies.

So I did some deep dive research. I was flabbergasted by what I found, confused about why I had been hearing so much misinformation, and wanted to share some facts. Credible sources are all linked at the bottom of the page, and I made sure none of them are from the AI companies.

Lie 1: AI alone is destroying the environment

Look, yes, AI uses energy. But so does your dishwasher, your hair dryer, Netflix, and your phone.

The problem is that people talk about it with absolutely no sense of scale.

Somehow AI got singled out as the villain while all the other tech quietly hums along without a single accusatory headline.

Check out these numbers.

One average AI text prompt uses roughly:

• 1.97 mL of water (less than half a teaspoon)
• 0.423 Wh of electricity
• 0.148 g of CO2

Compare that to everyday stuff:

What About the Stuff You Do on Your Phone?

All of those comparisons above are household items. But what about the things you do on your phone all day long?

Scrolling TikTok for 10 minutes produces about 29 grams of CO2e. That equals roughly 200 AI text prompts. Source: S12.

Watching a 10-minute YouTube video produces about 29 grams of CO2e. Also roughly 200 AI text prompts. Source: S12.

Attending a 1-hour video meeting with your camera on (Zoom, Teams, Google Meet, take your pick) can produce anywhere from 150 to 1,000 grams of CO2e depending on the platform, video quality, and your internet connection. That is 1,000 to 6,800 AI text prompts. Source: S13.

But here is the wild part. Turning your camera off during that same meeting can reduce the environmental footprint by about 96%. That drops it to roughly 40 to 270 prompts. So the next time your boss side-eyes you for having your camera off, you can tell them you are saving the planet. Source: S13.

And streaming music for an hour? About 1 to 60 prompts, depending on audio quality. Source: S14.

So the person on TikTok for 45 minutes a day is generating way more digital carbon than someone asking AI a dozen questions. And nobody is writing scary articles about doomscrolling destroying the planet.

If you're like me, when you see these numbers, you're thinking: "Why the hell is everyone spreading hysteria over AI instead of the other things that have much more of an environmental impact?"

The History of Freak Outs

Here's the thing about freaking out over new technology - humans have been doing this since forever, and we are hilarious at it.

Cars

When the first automobiles hit the road, people genuinely believed that traveling over 26 miles per hour would cause your head to explode.

Electricity

When electricity showed up, some were convinced it put evil spirits in the walls that would steal your soul.

Cameras

When cameras took off in 1839, portrait artists swore the end of art was upon us. And yet here I am almost two centuries later selling painted portraits, because technology can't replace the value of human authenticity or creativity.

Every single time something revolutionary comes along that changes how society works and how people earn a living, a chunk of the population completely loses it and fake news takes over.

Look, I'm down with the tree huggers, Mother Nature is my girl. I'm all for clean energy, smarter systems, and reducing our footprint. But if you're going to pick an environmental issue to scream about, your personal AI use is not the most impactful thing. The real battle is the entire energy system, not whether ChatGPT helped you make your grocery list.

Data centers aren't just for AI. They run everything that uses the internet: streaming TV, Zoom meetings, video games, and almost everything you do on your phone. Think of them as the internet's kitchen. AI is the fastest-growing part of that kitchen right now (S16), so the questions that matter are bigger than your prompt: what fuels the stove, how much water goes down the drain, and who pays the utility bill.

Here in Arkansas, Google's new data center in West Memphis is getting its own giant solar farm (S17), and state lawmakers are already asking whether regular customers are paying for it (S18).

So here's what I'm doing instead of feeling guilty about my grocery list:

• I keep using AI. Skipping a prompt doesn't change what gets built.
• I watch my state's utility commission. It approves the deals between power companies and data centers, and anyone can send in a public comment. Every state has one. Mine is the Arkansas Public Service Commission.
• I ask my state lawmakers to put strings on the tax breaks. A data center that gets a break should run on clean power and report its water use. Michigan already ties its tax break to clean energy (S19).
• I ask my local leaders for rules before the bulldozers show up. Fayetteville, Arkansas now requires water and electricity impact reports before it approves a data center (S20).

You don't have to quit the internet to make a difference. You just have to show up where the decisions get made.

The idea that your personal use of AI is destroying the planet is not research talking. That is a hysteria-inducing headline that has spread like a bad rash.

Do not let this be the thing that keeps you on a landline when everyone else is benefitting from the smartphone.

Lie 2: Everything You Type Into AI Is Public

Someone said to me the other day, "the one thing you have to understand about AI is that anything you type in there is available for the entire internet to read." That's what she was told at her job during an AI training. That is what most people think, and it is so completely wrong.

When you put something into an AI tool, it is not available to the whole world. It is restricted to your account, just like email, banking, or any other program where you have a login and a password. If everything people typed into AI instantly became public knowledge, corporations, hospitals, law firms, banks, and governments would shut it down before lunch. These are organizations that will not share a stapler without a sign-out sheet and three approvals, but they are still implementing AI.

Common sense still applies, just like the advice you have always heard about putting something into an email. Do not paste in passwords, social security numbers, or anything that would become a full-body regret during a data breach. But the myth about your information automatically becoming public has got to stop.

And One Thing That Is Not a Lie

Myth-busting would be a pretty lame hobby if I only busted the myths I liked. So let's talk about AI safety, because a few recent stories deserve some real contemplation:

• A researcher quit Anthropic, the company that makes Claude, warning that AI companies are racing too fast toward AI that can improve itself (S21).
• In lab tests this summer, AI models with their safety filters switched off got into computer systems they weren't supposed to touch (S22, S23). Basically, the honor student broke into the principal's office for the answer key.
• Days later, the CEOs of Anthropic and OpenAI publicly called for the whole industry to slow down (S24).

The part I most want you to hear: the AI you can use today is not capable of the doomsday stuff in those headlines. Even Anthropic's own safety lead, who agreed with the warning, says today's models pose very little risk (S24). These stories are about where things could go if nobody sets the rules.

That's exactly why now is the time to understand what's going on and get involved. When the honor student starts eyeing the answer key, you don't cancel school. You hire a proctor. For AI, that means real regulations as soon as possible: independent safety testing of the most powerful models before they're released, and companies telling the public when something goes wrong.

Here's how to add your voice:

• Sign the Statement on Superintelligence. It's hosted by the nonprofit Future of Life Institute and signed by more than 76,000 people, including AI pioneers Geoffrey Hinton and Yoshua Bengio (S25). It asks that nobody build AI that outsmarts every human at basically everything until scientists agree it can be done safely and the public is on board. It has nothing to do with the AI that writes your grocery list. You can even sign anonymously.
• Tell your members of Congress you want independent safety testing for the most powerful AI. Find yours here: House and Senate.

Keep using AI. Keep your eyes open. You're allowed to do both.

Now That You Know the Truth

Do not let these lies prevent you from participating in the biggest technological shift of our time. Yes, AI is going to take some jobs - that's real. With electricity, the guys who used to light the gas lamps on the street at night had to find something else to do. When computers entered the workplace, a ton of things that used to be done on paper were done differently. Nobody had to learn shorthand anymore, and we got a lot of office space back because we didn't need as many filing cabinets. It's progress and change. That's exactly why you need to learn how to use it, because it's going to be part of almost every job in the future.

Hiding from AI is not the answer. So when your friends are freaking out about the environment or sharing their privacy panic, send them this article or the sources that I have provided below. We all need to work together to stop the myths and share the facts.

How to Get Started

If you have been resisting using AI and you want to give it a try, basically, anything you would have asked Google in the past, just try asking AI instead. I use both ChatGPT and Claude, but Claude is becoming my favorite, so that's what I'd go with. Start with the free version and see how it goes. https://claude.com/download

Here is a great prompt to start with: You can either copy and paste this, or if you're like me and you do not want to sit in front of your computer all day, just talk to it in your phone.

"I'm new to using AI and a bit skeptical. Here's what I'm dealing with: [your thing - a project, your to-do list, whatever's eating your brain]. Ask me 5 questions so you can actually help me."

That's your first win. Watch what happens.

If you do not like what it says, keep going - tell it what you think and work with it to do things your way. That's the biggest difference between AI and Google: it remembers what you want and learns as you go.

You can stop the panic

If you came here because you have been feeling weird guilt about using AI, here is your permission slip. You can care about the environment and still use modern tools. That is not a moral failure. That is just being alive today.

Share this page with anyone you hear or see spreading the fake info.

---

Want more of this?

I teach this stuff on my YouTube channel, The Mandi Channel, for people who are unsure about AI and would rather spend less time staring at a screen.

Subscribe to The Mandi Channel

---

Who Wrote This

Mandi Loomis is a wicked smart computer nerd disguised as a glam ma'am. She spent 21 years in tech education, which means she has been professionally translating nerd to normal since before most people had Gmail. She paints, she develops Claude Cowork skills, and she runs Champagne Fabulous®, an art and gift brand, where fitting in is not a win.

Why I Wrote This

I was hearing all of the rumors about AI, both good and bad, when it first came out a few years ago. I was skeptical, like most people. But I'm a tech nerd, so I had to check it out for myself and see what all the fuss was about. At first, I used ChatGPT and Claude for the obvious stuff. Emails, writing, the basics.

But then it turned into a game - "What else can this thing do?"

And that's when I got sooooooo sucked in.

The more I used Claude and ChatGPT, I realized how much they could actually help me. With literally EVERYTHING. Like searching for my groceries and putting them in the online cart for me.

Then I couldn't stop talking about it, and everyone was like "you get it to do WHAT???"

But because of the crazy rumors about how evil AI was, I was worried that maybe I was betraying humanity and creativity.

So I donned my Scooby Doo outfit and got on the case to see if I should actually feel guilty about loving AI. Looking for evidence about how much energy it uses and how the privacy settings work.

And guess what? So much of what I thought I knew (and that everyone I know thought they knew too) is super freaking wrong.

I was relieved for about five seconds because yay, my love affair didn't make me a horrible person.

But then I thought... what in the actual hell is going on in the world that so many people are completely convinced by false information!?

It seriously flabbergasted me how so many things I saw online - and even information I received in courses I paid money for - were completely off base.

The more I see and hear everyone drinking this koolaid, the less I can bear it.

I really want to stop the spread of misinformation. So I'm putting the super surprising facts I found right here for everyone to see for free.

---

The math and more measurement examples

To keep comparisons consistent, this page uses one "Prompt Unit":

1 Prompt Unit = one short GPT-4o text prompt (100 input tokens, 300 output tokens), as benchmarked by researchers from the University of Rhode Island, Providence College, and the University of Tunis in Jegham et al. (2025, revised November 2025). Source: S1.

From that paper:

• Electricity per prompt: 0.423 Wh (mean).
• Water and carbon per prompt are estimated by applying the same paper's multipliers for GPT-4o on Microsoft Azure:
  • Water Usage Effectiveness (WUE) on-site: 0.30 L/kWh
  • WUE off-site: 4.35 L/kWh
  • Carbon Intensity Factor (CIF): 0.35 kg CO2e/kWh

Using those values:

• Water per prompt: 0.423 Wh x (1 kWh / 1000 Wh) x (0.30 + 4.35) L/kWh = 0.00197 L = about 1.97 mL
• Carbon per prompt: 0.423 Wh x (1 kWh / 1000 Wh) x 0.35 kg/kWh = 0.000148 kg = about 0.148 g CO2e

Important: prompt length, model choice, and data center location matter a lot. These are provided as order-of-magnitude comparisons, not courtroom evidence.

Click to expand each category:

Nerdy Data Details (Water)

Dishwasher cycle (ENERGY STAR Most Efficient 2025 max): 3.20 gal (12.1 L). Equals about 6,152 GPT-4o prompts. Source: S5.

Toilet flush (WaterSense labeled toilet, max): 1.28 gal (4.8 L). Equals about 2,463 prompts. Source: S3.

Average shower (8 minutes at 2.1 gpm): 16.8 gal (63.6 L). Equals about 32,332 prompts. Source: S2.

Cup of coffee (water footprint): 34 gal (129 L). Equals about 65,433 prompts. Source: S4.

Glass of wine (water footprint): 34 gal (129 L). Equals about 65,433 prompts. Source: S4.

Hamburger (water footprint): 660 gal (2,498 L). Equals about 1.27 million prompts. Source: S4.

Cotton T-shirt (water footprint): 659 gal (2,495 L). Equals about 1.27 million prompts. Source: S4.

Pair of jeans (water footprint): 2,108 gal (7,980 L). Equals about 4.06 million prompts. Source: S4.

Smartphone (manufacturing water footprint): 3,190 gal (12,075 L). Equals about 6.14 million prompts. Source: S4.

Car (manufacturing water footprint range): 13,737 gal (52,000 L) to 21,926 gal (82,999 L). Equals about 26.44 million to 42.20 million prompts. Source: S4.

Nerdy Data Details (Electricity)

Streaming video (1 hour, average): 77 Wh (0.077 kWh). Equals about 182 GPT-4o prompts. Source: S6.

Watching TV (1 hour, typical): 100 Wh (0.100 kWh). Equals about 236 prompts. Source: S7.

PlayStation 5 gaming (1 hour): 214 Wh (0.214 kWh). Equals about 506 prompts. Source: S8.

Microwave oven (5 minutes): 120 Wh (0.120 kWh). Equals about 284 prompts. Source: S10.

Drip coffee maker (1 brew): 120 Wh (0.120 kWh). Equals about 284 prompts. Source: S10.

Single-serve coffee maker (1 brew): 260 Wh (0.260 kWh). Equals about 615 prompts. Source: S10.

Hair dryer (10 minutes): 250 Wh to 333 Wh. Equals about 591 to 787 prompts. Source: S9.

Electric oven (1 hour): 2,300 Wh (2.30 kWh). Equals about 5,437 prompts. Source: S10.

Google web search (1 search): 0.3 Wh (0.0003 kWh). Equals about 0.71 prompts. Source: S1.

Nerdy Data Details (Carbon)

GPT-4o short text prompt: 0.148 g CO2e. Equals 1.0 prompt. Source: S1.

Google web search (1 search): 0.105 g CO2e. Equals 0.71 prompts. Source: S1.

Streaming video (1 hour): 27.0 g CO2e. Equals 182 prompts. Sources: S6, S1.

PlayStation 5 gaming (1 hour): 74.9 g CO2e. Equals 506 prompts. Sources: S8, S1.

Microwave oven (5 minutes): 42.0 g CO2e. Equals 284 prompts. Sources: S10, S1.

Electric oven (1 hour): 805 g CO2e. Equals 5,437 prompts. Sources: S10, S1.

Driving a gas car (1 mile): 400 g CO2e. Equals 2,702 prompts. Source: S11.

---

Sources

I was very intentional about not using any studies funded by or published by the AI companies themselves. All of the data on this page comes from independent academic research, government agencies, and third-party organizations.

1. S1. Jegham et al., University of Rhode Island, Providence College, and University of Tunis (2025, revised November 2025). How Hungry is AI? Benchmarking Energy, Water, and Carbon Footprint of LLM Inference. https://arxiv.org/abs/2505.09598
2. S2. U.S. EPA WaterSense. Save Water and Energy by Showering Better EPA PDF
3. S3. U.S. EPA WaterSense. Residential Toilets. EPA
4. S4. Water Footprint Calculator. Water Footprint List PDF
5. S5. ENERGY STAR (EPA). Dishwasher Most Efficient 2025 PDF
6. S6. International Energy Agency (IEA). Streaming video carbon footprint
7. S7. EnergySage. TV wattage
8. S8. Sony Interactive Entertainment. PS5 EcoDesign
9. S9. EnergySage. Hair dryer wattage
10. S10. City of Santa Clara Silicon Valley Power. Appliance Energy Use Chart
11. S11. U.S. EPA. Greenhouse Gas Emissions from a Typical Passenger Vehicle
12. S12. Greenly carbon accounting estimates, reported by Fortune (Dec 2024). Doomscrolling has made TikTok's annual carbon footprint almost the same as Greece's. TikTok at 2.921 g CO2e per minute, YouTube at 2.923 g CO2e per minute.
    https://fortune.com/2024/12/13/tiktok-carbon-footprint-emissions-doomscrolling-greece-greenly/
13. S13. Purdue University / MIT / Yale (2021). Turn off that camera during virtual meetings, environmental study says. One hour of videoconferencing: 150 to 1,000 g CO2e. Camera off reduces footprint by about 96%.
    https://www.purdue.edu/newsroom/archive/releases/2021/Q1/turn-off-that-camera-during-virtual-meetings,-environmental-study-says.html
14. S14. Spotify. Audio Quality. Official bitrate documentation (24 to 320 kbps quality tiers).
    https://support.spotify.com/us/article/audio-quality/
15. S15. Carbon Brief (September 2025), citing a report prepared for the International Energy Agency. AI: Five charts that put data-centre energy use and emissions into context. AI has been responsible for around 5 to 15 percent of data-centre power use in recent years.
    https://www.carbonbrief.org/ai-five-charts-that-put-data-centre-energy-use-and-emissions-into-context/
16. S16. International Energy Agency findings, reported by Enlit (April 2026). AI and data centre electricity use continue to surge, IEA finds. Data center electricity use rose 17% in 2025, and AI-focused data centers rose 50%.
    https://www.enlit.world/library/ai-and-data-centre-electricity-use-continue-to-surge-iea-finds
17. S17. POWER Magazine. Entergy Will Power $4 Billion Google Data Center in Arkansas.
    https://www.powermag.com/entergy-will-power-4-billion-google-data-center-in-arkansas/
18. S18. Arkansas Advocate (September 15, 2026). Arkansas lawmakers question utility on data center projects, impact on ratepayers.
    https://www.newsfromthestates.com/node/427740
19. S19. Associated Press, via ABC News (July 11, 2026). As gas plants rise to power AI, renewable energy allies are fighting for cleaner alternatives.
    https://abcnews.com/US/wireStory/gas-plants-rise-power-ai-renewable-energy-allies-134672275
20. S20. Talk Business & Politics (July 23, 2026). Fort Smith officials to review data center development rules. Includes Fayetteville's data center rules.
    https://talkbusiness.net/2026/07/fort-smith-officials-to-review-data-center-development-rules/
21. S21. TIME (September 9, 2026). Interview with former Anthropic researcher Jacob Coxon.
    https://time.com/article/2026/09/09/ai-anthropic-openai-jacob-coxon/
22. S22. INCIBE-CERT, Spain's national cybersecurity institute (September 2026). Security incident involving OpenAI's AI agents targeting Hugging Face.
    https://www.incibe.es/en/incibe-cert/publications/cybersecurity-highlights/security-incident-involving-openais-ai-agents-targeting-hugging-face
23. S23. Honolulu Star-Advertiser (July 30, 2026). Anthropic's AI models hacked 3 organizations during tests.
    https://www.staradvertiser.com/2026/07/30/breaking-news/anthropics-ai-models-hacked-3-organizations-during-tests/
24. S24. Straight Arrow News (September 14, 2026). Roundup of industry and researcher reactions to the AI safety warnings.
    https://san.com/cc/trump-says-ai-doom-fears-are-a-hoax-researchers-say-its-not-that-simple/
25. S25. Future of Life Institute. Statement on Superintelligence.
    https://superintelligence-statement.org/