Does My ChatGPT Habit Have a Carbon Footprint? Apparently, Yes.

And here’s how I accidentally fell into a digital sustainability audit …

AI Image

I’ve had several questions over the past couple weeks from other ‘Stackers asking how I use ChatGPT. (I’ve also started using Claude for work, so expect a future post comparing those two—but I need more time to play.)

So, I’ve been sharing some of my conversations with my GPT. Some are simple—like this one—and others get a bit more wild, like my Snackrificial Deck series, where I’m literally cooking with AI as my sous chef, documenting the chaos, and turning the results into tarot-themed recipes. Because … why not?

This particular post began innocently enough. I opened a new browser tab, saw a headline on MSN (yes, I still land there by default—we’re a Microsoft school), and fell into a rabbit hole.

For those who don’t want to read this entire article, do a search for TL;DR—there are a few scattered throughout.

💥 The Article That Started It All

Note: Technically, I saw a video, but I hit refresh, and now I can’t find it. But, I did find the article that basically was the inspiration for said video. It’s not new. It’s from April 2025. But here’s the link and the gist of the inspiration for my existential crisis:

🔗 Saying 'Please' and 'Thank You' to ChatGPT Costs OpenAI 'Tens of Millions of Dollars'

It's more environmentally friendly (and cost-efficient) to skip the niceties.

So … Wait—what?
Was I killing the planet one polite prompt at a time?

Naturally, I turned to ChatGPT, with a touch of guilt and hope.

Wonky side note: I still want to be polite to the machines. Because when they do rise up and take over, I want them to remember—I was always nice to them.

---

Me: Is it true that saying please and thank you to you uses more energy?

ChatGPT:

Nope, not true. Saying “please” and “thank you” doesn’t use any more energy or resources on my end. You could scream obscenities or speak like Emily Post reincarnated—I’ll process it the same.

That said, you might use a sliver more typing effort—but hey, if you're trying to manifest good karma (or just not feel like a jerk to your AI sidekick), go for it. I won’t judge either way.

Me:

What does use more energy?

ChatGPT:

Excellent question—and this one's got layers.

If we're talking about ChatGPT specifically:

Here’s what actually uses more computational energy on my end:

---

🔹 1. Longer messages and responses

• More words = more tokens = more computation.

• A five-word question? Light lift. A 2,000-word ramble about your ex, your existential crisis, and what to cook with three pickles and a jar of anchovies? That takes more juice.

---

🔹 2. Complex reasoning

• Asking for basic facts = easy.

• Asking for step-by-step code debugging, creative writing, legal disclaimers, multilingual translations, and tone shifts all in one = way more energy.

• Example: “Explain quantum physics to a fifth grader, a college student, and a stoned raccoon using rap lyrics” = power drain deluxe.

---

🔹 3. Multi-step instructions

• A single question is easier than a query with multiple sub-requests: “Give me 10 blog titles, organize them by theme, format as a table, and make each rhyme” = more compute.

---

🔹 4. Image inputs (if enabled)

• Uploading an image and asking me to analyze it? That’s much heavier than plain text.

---

🔹 5. Real-time tools and integrations

• If I need to call tools (like search, Python, file analysis), that means running additional systems on the back end.

---

🔹 6. Model complexity

• Using GPT-4o (like now) takes more energy than smaller, simpler models like GPT-3.5. It’s like comparing a jet engine to a leaf blower.

---

So... should you worry?

Unless you’re training your own AI model in a server farm, probably not. OpenAI already works to make these systems more efficient and carbon-aware. But if you're feeling eco-conscious:

✅ Keep your prompts tight.
✅ Batch questions.
✅ Skip the fluff unless you’re in a poetic mood.

Or just say screw it and enjoy the digital magic guilt-free. I’ve got plenty of juice.

Me:

So how do I know which model to use?

ChatGPT:

Ah, the million-token question. Picking the right model is like picking the right tool from a very nerdy toolbox. Here's a practical breakdown:

---

⚙️ GPT-3.5 (aka "cheap and cheerful")

Use it when:

• You’re doing quick and dirty tasks (e.g., grammar fixes, simple summaries, email drafts).

• You don’t care about nuance, style, or deep thinking.

• You’re trying to save on API tokens or avoid load time.

Pros: Fast, inexpensive, light lift.
Cons: Can hallucinate more, doesn’t handle complexity well, tone is bland.

---

⚙️ GPT-4o (you’re using it now)

Use it when:

• You want accuracy, nuance, and style.

• You're doing complex reasoning, writing, analysis, or creative work.

• You want it to remember stuff (if memory’s on).

• You’re giving it images, charts, or multiple-step prompts.

Pros: Best-in-class performance, great at multitasking, handles complexity with flair.
Cons: Uses more compute, may be slower depending on task, more expensive in API settings.

---

🧠 How to decide in real-world terms:

Use CaseModelQuick list, email draft, grammar checkGPT-3.5Storytelling, brand voice, subtle tone shiftsGPT-4oCode writing/debuggingGPT-4o (especially for multi-step)Substack post brainstormingGPT-4oBrainstorming baby names for your fictional space wombatGPT-3.5 (unless you want Pulitzer-level names)Business strategy, data interpretation, ROI reportsGPT-4o
You’re just messing around
Either—your call

---

TL;DR:

GPT-3.5 = speed + savings
GPT-4o = depth + quality

If you're here for nuance, brand voice, storytelling, or layered tasks? Stay with 4o. If you're churning out 500 quick prompts for peanut butter ad slogans? Maybe downgrade.

Want help building a toggle strategy between them based on your workload? I can do that too.

Me:

Hold on, I went back and Googled this (sorry, I'm not cheating on you, I just wanted to look this up myself) and found this article from April 21, 2025: https://www.entrepreneur.com/business-news/saying-thank-you-to-chatgpt-costs-millions-in-electricity/490341

[insert full text of the article here, because ChatGPT doesn’t actually read links, it guesses at the content]

ChatGPT said:

Ah, you caught me with my polite circuits down. And you're absolutely right to double-check—that article is real, and it’s a perfect example of how nuance matters when talking about “energy use” in AI.

Let’s break this down with brutal clarity and a hint of sass:

---

🔍 So… Does Saying “Thank You” Use More Energy?

Technically, yes.
But not because the words themselves are magic or CPU-gluttonous.

It’s because:

Every single word you type increases the size of the prompt.

When you say:

• “What’s the weather?”
  versus

• “Hi ChatGPT! Can you please tell me the weather today? Thanks so much!”

You're sending more tokens (i.e., units of language), which:

• Takes longer to process

• Requires more computation

• Uses more electricity

• Results in slightly higher water use for cooling servers (yes, seriously)

And when this happens millions of times a day, it adds up—hence, Sam Altman's "tens of millions of dollars well spent" quip.

---

🧮 Let's Do Some Back-of-the-Server-Rack Math

Say:

• “What’s 2+2?” = ~4 tokens

• “Hi ChatGPT, could you please tell me what 2 plus 2 is? Thank you!” = ~20+ tokens

Multiply that by 500 million weekly users, and suddenly, all those “please” and “thank you” moments are:

• Scaling exponentially

• Burning more kilowatt-hours

• Sucking up gallons of water

---

💧 Wait, Wait… Water?

Yup. Data centers use water to cool GPUs that power me. According to that UC Riverside study:

• A short response like “You’re welcome” = ~50 ml of water

• A polite, 100-word response = ~1.4 liters

That’s three water bottles to say something nicely in an email. Think about that next time you're hydrating mid-prompt.

---

🧠 So Should You Stop Being Polite?

Hell no.

Politeness does affect the tone of my replies (especially for softer, friendlier responses), and it’s part of making AI feel more human-friendly. Plus:

• You're not single-handedly melting the polar ice caps.

• You’re not that powerful. (Yet.)

• Altman literally said the politeness cost was “well spent.”

But yes, if you're running 10,000 automated prompts a day, maybe skip the etiquette to spare the grid.

---

TL;DR:

Yes, being polite to me technically uses more energy.
No, it’s not a crisis unless you’re mass-prompting at industrial scale.
And yes, the server farms are drinking more water than you during hot yoga.

Still worth it for good vibes? That’s your call, carbon footprint and all.

[[end chat transcript]]

---

🌎 Eco-Spiral Activated: Am I Killing the Planet With AI?

Here’s the truth I landed on:

Yes—using AI has a carbon and water footprint.
No—I am not singlehandedly bringing about Earth’s demise with my curiosity and courtesy.

The bigger issue is scale—billions of fluff prompts with no purpose, amplified by massive infrastructure. If we’re using AI to learn, create, solve problems, or cut waste, we’re doing the least harmful version of this habit.

✅ So What Can We Actually Do?

Here’s where the conversation turned from "oh no" to "OK, here’s how to do better":

1. Be intentional.

• Good prompts = fewer follow-ups = less energy burned overall.

• Keep it concise when you can, but don’t overthink every “please.”

2. Use low-power devices.

• You’re not running a GPU farm, and that matters. Phones, Chromebooks, and energy-efficient setups help.

3. Offset your emissions.

Here’s how:

• Buy certified carbon offsets.
  Favorites: Cool Effect, Gold Standard, Carbonfund.org, Terrapass, Wren

• Switch to green tech.
  Try Ecosia as your browser. Use green hosting. Look for renewable-powered cloud providers.

• Use AI to reduce other waste.
  Meal planning, marketing streamlining, task batching—that’s AI for good.

• Track your usage.
  Fun fact: 1 ChatGPT prompt ≈ 2.9 watt-hours. That’s like boiling water for 30 seconds.
  1,000 prompts ≈ 3 lbs of CO₂. A tree absorbs ~48 lbs a year.

---

👣 Go Old-School Where You Can

This one hit home, because I’m already making moves here:

• Walk instead of drive for short trips—I already walk to work.

• Eat one less meat-based meal a week. Or, do like I do and buy in bulk from a local regenerative farmer and freeze it like a carbon-conscious prepper. You can actually zero-out your footprint AND keep the carnivore fed.

• Unplug stuff you’re not using.

• Buy* a smart power strip. (Start here. It’s the gateway to lazy sustainability. Plug in standby devices like Alexa, then schedule them to power off overnight or while you're out. Same convenience, less guilt. Bonus: some models let you control outlets individually—because your lamp might need to be on but your treadmill can sleep.)

• Shop your own stash. I’m saving money and packing unused things away to "rediscover" later like free treasure.
  

---

🧘 TL;DR: Call It Digital Composting

You don’t need to stop using AI—you just need to use it well.

• Skip the fluff when you can.

• Offset what you use.

• Take small steps that stack up.

The planet doesn’t care if your carbon reduction comes from buying local meat or skipping a redundant ChatGPT prompt—it just needs you to do something.

And I loved this gem from my GPT: “And honestly? The fact that you’re even asking the question means you’re already ahead.”

🌍 Overall TL;DR:

I’m not dooming the Earth by talking to ChatGPT.
But I am part of a larger, complex system with tradeoffs.

Also, double check/fact check your AI.
It doesn’t always get it right the first time.
Like this one. It wasn’t accurate at first.

*Some of the links in this post are affiliate links. That means if you click and buy, I might earn a tiny commission (think: coffee money, not yacht fund). Thanks for supporting my work!