The Power Is Yours… Sort Of: AI Writing and the Environment

My generation grew up on Captain Planet and The Lorax. Commercials in the middle of our Saturday morning cartoons told us to recycle, reduce, reuse. In adulthood, news reports with many fewer friendly cartoon dinosaurs informed us about the rise and dangers of global warming, while over the last few years, generative AI has joined planes, cars, factory farms, and other ecological nightmares.

Some people have responded by refusing to use AI at all, while others have embraced the technology with no concern for its potential effect on the planet (or much of anything else). Particularly as more and more companies incorporate AI, many people want to find compromises that let them keep their jobs while being as environmentally friendly as possible.

The good, the bad, and the bigger picture

The good news: A 2024 study in Scientific Reports estimated that an AI writing tool emits 130 to 1,500 times less CO₂ per page of text than a human writer (counting a share of the writer’s everyday footprint for the hours spent writing). Text is far lighter than images or video: a typical AI image takes about 60 times the energy of a short text answer, according to a 2026 UN University report. Independent researcher Andy Masley estimates that one prompt uses about as much energy as ten to twenty minutes of Netflix, and a whole conversation about a minute of microwave use.

The bad news: These measurements don't take into account the environmental impact of training the model in the first place, which is by far the most destructive part of any AI.

The good news: That means you don't have to beat yourself up over writing a business email with Claude or using the AI summary tool your boss wants, any more than you need to feel bad about flying across the country once in a while or putting your laundry in the dryer when it's gross out. As with so many issues, it's companies, not individuals, that need to make serious change.

"Use of AI to generate text or imagery probably accounts for a mere sliver of any given person's environmental footprint. And experts stress that the onus is on tech companies to reduce AI's resource consumption, from creating smarter, energy-saving algorithms to building more efficient hardware."

—Katarina Zimmer, "Using AI chatbots? Here are four ways to reduce the energy drain", Knowable Magazine, July 21, 2026

The bad news: Companies respond to consumers. If you use AI a lot, they'll see demand and probably build more datacenters or roll out more of the enormous models that present real problems.

"Individual choices are not meaningless, and some are more powerful than people realize," says computer scientist Ivana Drobnjak of University College London.

So how do you make good choices if you have to use AI for your writing?

Think of it like your boss.

Wait, what?

No, I'm not saying we should surrender to the machines now.

I also hope your actual boss isn't relevant. (For the sake of my continued employment, I should probably note that none of mine are!) Instead, I'm talking about Your Boss in the cartoon or sitcom style, and the RL versions most of us sadly do run into at some point. Think Malory Archer, Monty Burns, or those guys in The Flintstones and The Jetsons. (While Ms. Archer represents a growing female iteration of the type, I’m going to stick with male pronouns to fit the larger demographic.)

Your Boss is pretty lazy, he's not too bright, and he doesn't have a lot of time or energy to spend on your project. If you want to get his approval or, God help you, his opinion, you're going to have to come prepared…and I don't necessarily mean with reasoning or details.

First, know what you need

You don't interact with Your Boss any more than you absolutely have to, right? Violating that rule is a good way to get fired, asked to work Saturdays, or at least told way too much about his new macrobiotic diet and how it's affecting his love life. You figure out where you need his involvement and where you can't avoid it, and you plan accordingly.

Same here: Does your company want you to summarize three articles with their just-implemented tool? Do you need to send out a quick email about a new product? What are you trying to write, and why do you need AI to do it?

If your task is relatively simple, like a summary, a standardized email, or a search for information in a specific field, a small language model might work just fine. These forms of generative AI don't have access to the wide array of content ChatGPT does, but they also don't use as much power and water or emit as much carbon, and when you need to summarize a chat log or sum up the advantages and disadvantages of a particular product, a smaller model could do the job very well. (Just remember to double-check the results—big or small, no model gets it right all the time—and triple-check them if you're sending an email!)

Also consider how specific and precise you need to be. The more an AI model "thinks," the more tokens it takes and therefore the more environmental damage it does. If you're asking your LLM for help with a technical document, maybe it really does need to come back with "a chip 3.23 nm on its longest side." If you're writing an interior decorating plan, you'll be fine with a model telling you the table should be "about six feet."

Second, know how to ask

Your Boss wants facts, figures, and bullet-point summaries (or crayon drawings, in some cases). Interactions with him aren't the place for any more context or nuance than you can possibly avoid.

Treat AI likewise. Every extra word you put into the prompt, and every extra prompt you have to use, takes more power and water and emits more carbon.

The open access CoDesignS AI Framework (2024) offers the ROCKS guide to composing efficient prompts: Role, Objective, Community, Key, and Shape.

R, O, and C are pretty obvious. "Key" means style, tone, and parameters like "optimize for applicant tracking systems" or "account for the role of Napoleon in the conflict." "Shape" is the format: do you want the results in an email, an academic paper, a resume, or a screenplay?

Unlike Your Boss, an AI model will also keep its responses short if you ask. (It won’t even get huffy or bring it up at your mid-year review!) Take advantage of that difference, both because it’s a lovely and rare experience for most of us and because it helps the environment: in Knowable Magazine, Katarina Zimmer points out that cutting an AI’s answer in half saves about ten times as much energy as shortening the prompt.

Tell your AI to keep its reply below a certain number of words, specify five bullet points, request lists or tables (which automatically lower word count), or just practice your best keep it quick, I don’t have all day in a New York accent. It’ll help the planet and it may give you a delightful little feeling of power as well.

Try to fit multiple tasks into one prompt, so that your interface doesn't have to keep calling the AI. "Revise the attached resume to emphasize tasks appropriate to the job below, then write a corresponding cover letter in the style of George Orwell," can all be one request, if you've planned far enough ahead to know you'll need all of that.

Speaking of that attached resume and the job description, make sure you've cut them down beforehand so you're only asking the LLM to parse relevant information. Don't ask it to summarize an entire history book when you only need to know about the Battle of Hastings. Remember: you're trying to get the input you need without making Your Boss fret about missing his three-martini lunch.

Third, pick your time

When you have a choice, you don't ask Your Boss to review your proposal too close to the lunch I just mentioned—he's not going to do his best work when he's worried about his reservation at the Four Seasons. You don't bring problems up on a Friday at 4:30 unless you want to slide them under the radar, and you don't ask anybody anything on Monday before they've had at least two cups of coffee. (What can I say? Sometimes stereotypes exist for a reason.)

Timing applies to environmentally conscious AI use too, it's just less fun and/or dramatic. The time of day can influence how easy a datacenter is to cool (and thus water use), the drain on the power grid, and whether that grid is using renewable energy or has switched to a more destructive fuel source like coal.

You can check data to find out when particular systems or datacenters are running at their environmental best. If you have a choice about when to work, especially if you need to tackle a big project, avoid the times with peak impact.

See beyond your actions, look beyond the environment

Individual actions are great, but there's only so much you can do when the structure fails. Not to go too 1990s Earth Day here, but if you're concerned about how writing with AI could hurt the planet, take action on a larger scale. Talk to your legislators about regulating energy sources and mandating more sustainable cooling tactics. (That's not just an AI thing, either—datacenters have been around for decades, using natural resources to perform calculations, play cat videos, and everything in between.) Talk to your actual boss about how the company sources and applies LLMs.

Don't forget non-environmental issues, either: job displacement, deceptive labeling, and ethical concerns about source material and training practices. AI may not be going away any time soon, it may be valuable if you use it carefully, and there may well be no ethical consumption under capitalism…but that doesn't exempt any of us from the responsibility to do what we can for the world and each other.