By Rachel Hurley
The anti-AI mafia is in my comments again. I don’t have any interest in refuting their claims one by one—but if they wanna talk about AI, let’s talk about AI.
None of this will change any minds-but these are the facts.
The biggest issue I see is that people’s entire arguments are based on a few headlines they’ve read—and many of them are out of date.
The claim that chatbots use a ton of energy came from a guess somebody published in 2023, before there was real data. Researchers redid the math and found a number ten times smaller than they had originally thought. Not to mention they get more efficient every time a new model is released. Google published a study of its own systems and published the real figure: 0.24 watt-hours per question.
In terms of power—that’s one second of your microwave.
And the pictures—one AI image takes between 0.3 and 1.2 watt-hours. Call it four chatbot questions at the high end. Less than 30 seconds of Netflix. Every image on my page costs less power than the autoplay video you just watched. Every image in my campaign used less energy to make than 2 hours of scrolling Facebook—and that’s text. If you’re scrolling Reels—the number jumps way higher.
And let’s talk about Facebook specifically—it uses generative AI to feed you ads and generate everything you see in your feed. You use it every time you search. Using FB to criticize the use of generative AI does not make any sense.
Now let’s look at what else nobody feels guilty about.
One hour of Netflix in HD burns about 400 chatbot-question-worth of power. Hulu, Prime, Max—same thing, because an hour of streaming is an hour of streaming. An hour of Zoom, Google Meet, or Microsoft Teams with your camera on is worth about 200 AI questions. TikTok tested worst of all the apps—almost double YouTube’s drain, thousands of questions’ worth every hour.
Music is the one cheap thing on your phone. Streaming a song moves about 20 times less data than streaming video. But if you’re streaming hours at a time—on Apple or Spotify—you’re using AI and racking up a lot of power usage.
The part that really gets me is how much AI the “I’d never use AI” crowd uses every single day. Search Engines create AI summaries on billions of searches now. Generative AI is baked into almost all word processors and spreadsheet creators. It’s in almost all graphic design software and video editing software. It’s the backend of most website hosting. Not to mention any of your home smart devices.
If you’re scrolling Instagram, Threads, WhatsApp, Pinterest, X, Reddit, LinkedIn, Telegram, Nextdoor, Quora, or Snapchat—SURPRISE—every scroll of your feed runs AI deciding what you see next.
Oh—and if you shop online—Amazon, Walmart, Target, Home Depot, Lowe’s, Kroger, Best Buy, eBay, Etsy, Wayfair, Stitch Fix, Saks, David’s Bridal, Guitar Center, Ashley Furniture, Sephora, and Ikea—to name a few—all have generative AI baked in.
If you order anything from Instacart, DoorDash, or Uber Eats—same story.
Wendy’s FreshAI takes drive-thru orders with generative AI (built with Google), Taco Bell runs AI drive-thrus, and Starbucks runs its Deep Brew AI across the app and stores.
Your government uses it too. Memphis put AI cameras on city buses that found 63,000 potholes. Police dash cams automatically read license plates. The DA’s office uses AI to sort through evidence. Gallup says 43 percent of government workers now use AI on the job.
The comeback I always get is training. Sure, one question is cheap, but training the model ate a whole city’s worth of power. Except answering questions is now 80 to 90 percent of what AI computing does.
Water is the same math. One question is a few drops. An hour of scrolling uses most of a water bottle.
I’m not saying there are no issues to be addressed—there are plenty. And if anyone thinks I have not thought through them—they are wrong. But nobody ever mentions that I am using AI to educate people.
When I originally started writing about corruption, I wanted to reach people who usually don’t pay attention. That’s why I came up with the RATC character—as a detective—and sometimes an image is the only thing that gets someone to stop and pay attention. The character is based on a hand-drawn image a friend created over a decade ago—one that uses a generic comic book style. I created it specifically so it wouldn’t be misleading. I weighed the positives and negatives—and made a pragmatic decision—one that I will continue to stand by. If it helps spread the information—it’s worth it.
And I am not using AI to write my posts—if you know anything about me—I’ve been writing long-winded posts for decades—and I have been told I need an editor for that long, too. I use it for research and fact-checking, and sometimes to make my essays shorter and easier to understand, with an AI disclaimer on my website.
I’ve said it before—and I will say it again—the argument about whether or not to use AI is moot—we are all using it. The focus needs to be on how to regulate it to minimize harm. But for some reason, we can’t seem to do that.
— Rachel Hurley
Peer Reviewed Sources:
https://dl.acm.org/doi/10.1145/3630106.3658542
https://arxiv.org/abs/2505.09598
https://www.science.org/doi/10.1126/science.aba3758
https://arxiv.org/abs/2204.05149
https://www.cell.com/patterns/fulltext/S2666-3899(21)00188-4
https://oceanrep.geomar.de/id/eprint/58910/
https://www.nature.com/articles/s41467-024-47621-w
https://www.sciencedirect.com/science/article/pii/S2542435123003653
https://proceedings.mlsys.org/paper_files/paper/2022/hash/462211f67c7d858f663355eff93b745e-Abstract.html
Chatbot Energy (2023 guess, the redo, Google’s 0.24)
https://epoch.ai/gradient-updates/how-much-energy-does-chatgpt-use
https://www.datacenterdynamics.com/en/news/google-median-gemini-prompt-uses-024-watt-hours-of-power-and-consumes-026ml-of-water/
https://hannahritchie.substack.com/p/ai-footprint-august-2025
Netflix/Hulu/Prime, Zoom/Meet/Teams, TikTok, Facebook scrolling
https://www.iea.org/commentaries/the-carbon-footprint-of-streaming-video-fact-checking-the-headlines
https://www.forbes.com/sites/johnkoetsier/2025/12/03/new-data-ai-is-almost-green-compared-to-netflix-zoom-youtube/
https://greenspector.com/en/social-media-2021/
https://www.purdue.edu/newsroom/archive/releases/2021/Q1/turn-off-that-camera-during-virtual-meetings,-environmental-study-says.html
Music streaming
https://factually.co/fact-checks/environment/music-streaming-co2-emissions-per-listener-a5aec0
AI search summaries, & Meta AI apps
https://blog.google/products-and-platforms/products/search/generative-ai-google-search-may-2024/
https://techcrunch.com/2026/05/12/threads-tests-a-meta-ai-integration-that-works-similarly-to-grok/
Retailers and delivery
https://www.retaildive.com/news/a-look-at-retails-year-of-ai-news/807946/
https://www.aboutamazon.com/news/retail/how-to-use-amazon-rufus
https://www.modernretail.co/operations/chatbot-cheatsheet-a-guide-to-the-ai-assistants-from-walmart-loreal-amazon-and-more/
Government (potholes, plate readers, DA, Gallup)
https://publicsector.google/ai/data-driven-solutions-helped-memphis-find-and-fix-63-000-potholes/
https://www.actionnews5.com/2026/02/24/shelby-county-da-partners-with-memphis-ai-company-manage-largest-caseload-history/
https://www.gallup.com/workplace/702983/adoption-rapidly-growing-public-sector.aspx
Training vs answering (80-90%)
https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/https://www.sciencedirect.com/science/article/pii/S2666389925002788
https://www.pnas.org/doi/full/10.1073/pnas.0906974107
https://www.pnas.org/doi/10.1073/pnas.1006388108
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.