Stacie Porter Bilger is the Founder and President of Proof Digital. She brings more than three decades of experience helping business leaders build strategies rooted in data, ingenuity, and evolving technologies like AI.
TLDR: Every prompt you send an AI tool uses tokens, and tokens translate directly into energy, time, and cost. Simple habits like choosing lighter file formats and writing clearer prompts can meaningfully improve AI and energy efficiency for marketers and business owners who rely on these tools daily.
At a Glance:
- AI models process every input and output as tokens, and more tokens mean more energy, more wait time, and a higher bill.
- File format matters more than most people realize. Uploading a PDF or PowerPoint costs far more than a plain text or markdown file.
- Small workflow changes, like reusing chats and trimming uploads, add up to real AI energy efficiency gains over time.
- Being thoughtful about AI energy use does not mean giving up the tools. It means using them with intention.
AI tools have become part of daily work for a lot of marketers.
Drafting captions, summarizing reports, building outlines…none of it feels like a big deal one prompt at a time. But every one of those prompts costs something, and it is worth understanding what that cost actually looks like.
Here’s what you should know.
Does AI Use Energy? Yes, Every Prompt Has a Cost
Does AI use energy? It does, and the amount depends on what you ask it to do and how you ask it.
Every interaction with an AI model gets broken down into tokens, which are small chunks of text the model reads and generates. More tokens require more computing power, and more computing power requires more electricity.
That electricity has to come from somewhere, and data centers running these models draw real power at scale. A single prompt might feel weightless. Millions of prompts running across every business, every day, add up to a measurable footprint.
How much energy does AI use on an individual level is hard to pin down exactly, since it varies by model, task, and file type. What marketers, and other professionals, can control is how efficiently they prompt and what they upload.
Putting Tokens Into Perspective
A number like “50,000 tokens” does not mean much without something to compare it to.
Here is a rough way to picture it: 1,000 tokens is roughly 750 words of English text, so a file that costs 50,000 tokens to process is the token equivalent of reading or generating close to 37,000 words in one shot, well over the length of this entire blog post repeated several times over.
Estimates on the actual electricity behind each token vary widely depending on the model, the hardware running it, and the data center itself, and researchers are still refining those numbers. But the general pattern holds across most published research: more tokens require more computing cycles, and more computing cycles pull more electricity from the grid.
A prompt that processes 50,000 tokens is doing meaningfully more computational work, and burning meaningfully more energy, than one that processes 4,000.
How Much Energy Is AI Using? It Comes Down to Tokens
A major, yet often overlooked, player in AI energy use is file format.
Andy Crestodina, CMO and Co-Founder of Orbit Media, covered this in detail during Better Than Prompts: How to Build AI Assistants for Your Agency, a recent presentation powered by E2M.
Crestodina shared research from Orbit Media comparing estimated token usage across six common file formats, both reading a file (input) and generating one (output). The gap between the most and least efficient formats is bigger than most AI users would think.
The Token Cost of Common File Formats
Plain text (.txt) files sit at the bottom of the group, using roughly 3,900 tokens to read and 4,000 to generate. Markdown (.md) files run close behind at about 4,900 tokens in and 5,000 out. Both formats are stripped-down text with little to no extra formatting, so AI models process them quickly and cleanly.
From there, the cost climbs rapidly. Word documents (.docx) jump to around 13,800 tokens to read and 12,000 to generate, since the model has to work through embedded styling and layout data along the way.
HTML files land at roughly 16,000 tokens to read but only about 7,500 to generate, making HTML a reasonable choice when you need AI to produce something more structured than plain text.
PDFs are where things get expensive. Reading a PDF costs an estimated 45,800 tokens, even though generating one back only runs about 7,500. That input cost adds up quickly, especially since PDFs also cannot let AI process embedded images the way a human reader can.
PowerPoint files (.pptx) top the group on both ends, costing close to 49,500 tokens to read and 50,000 tokens to generate.
What This Looks Like in Practice
A few practical swaps make a difference:
- Instead of uploading a full brand guide PDF, paste the relevant section as plain text or save it as a markdown file.
- Instead of asking AI to generate a polished slide deck directly, ask it to draft the content in markdown first, then format the deck yourself or with a design tool.
- Instead of dropping a Word document into a chat for a quick edit, copy the text directly into the prompt.
- If you need something more structured than plain text, HTML is a lighter option than PDF or PowerPoint for both reading and generating.
None of these swaps require new software or a steep learning curve. They just require noticing which format you reach for out of habit and choosing a lighter one when the task allows it.
Crestodina put it plainly: if a file opens in Notepad, it is an efficient format for AI. If it needs dedicated software to open, it is an expensive one.
Uploading a bulky slide deck or scanned PDF to an AI assistant is one of the slowest, most token-heavy ways to use the tool, and it often does not even deliver a better result.
Simple Swaps for AI Energy Efficiency
Improving AI for energy efficiency does not require giving up your favorite tools. It just requires a few smarter habits.
Choose the Right File Format
When you have a choice, feed AI plain text or markdown files instead of PDFs, PowerPoints, or Word docs. If you need AI to generate a document, ask for markdown first and convert it later. This can cut token usage dramatically.
Skip the Giant Uploads
Not every AI assistant needs your entire brand guide or a 40-slide deck to answer a question. Pull the relevant section and share just that. Smaller inputs mean faster answers and a lighter load on the model.
Write Clear, Specific Prompts
Vague prompts often lead to multiple back-and-forth exchanges before you get a usable answer. A clear, specific prompt on the first try saves tokens and saves your own time.
Reuse and Reference Instead of Re-Uploading
If you are working in a tool with memory or project features, reference earlier context instead of re-uploading the same files. Repeated uploads mean repeated processing for information the AI has already seen.
How to Use AI Ethically Beyond Energy
Discussions around ethical AI use cover more ground than energy consumption alone, and it is worth thinking about the full picture.
AI ethics start with transparency. If AI helped draft content, a report, or an image, disclose it where it matters. Fact-check anything AI generates before it goes out under your name or your brand’s name. Models can sound confident and still be wrong.
Being thoughtful about data is another way how to ethically use AI. Do not paste sensitive client information, financial details, or proprietary strategy into public AI tools without understanding where that data goes.
And keep a human reviewing the output, since AI is a tool for speeding up work, not a replacement for judgment.
Why This Matters for Marketers and Business Owners
Digital and marketing trends move fast, and AI tools have made keeping up with that speed possible in ways that felt unimaginable a few years ago.
That speed does not have to come at the cost of being thoughtless about how these tools get used.
Concern about AI and energy efficiency is growing among businesses, consumers, and regulators. You do not need to overhaul your entire workflow to respond to that shift. Small, consistent habits, like the ones above, make a real difference without slowing your team down or draining morale.
Use AI Well, Not Just Fast
AI is not going anywhere, and neither is the conversation around its environmental and ethical impact.
Marketers who pay attention to AI energy use now will be better positioned when clients, partners, and platforms start asking harder questions about how these tools get used.
You do not need to choose between using AI and using it responsibly. A few smarter habits around file formats, prompt clarity, and data handling go a long way.
Curious how Proof Digital builds AI-informed strategy without losing sight of efficiency or ethics? Talk to our team.
FAQs
Does AI use energy?
Yes. Every prompt sent to an AI model requires computing power to process, and that power comes from electricity used by data centers.
How much energy does AI use?
Energy use of AI depends on the task, the model, and especially the file format involved. Text-heavy uploads like PDFs and slide decks require far more processing power than plain text or markdown files.
How much energy is AI using compared to a normal search?
Generative AI tasks generally require more computing power than a standard search query, particularly when large files or images are involved in the request.
How can I improve AI energy efficiency at my company?
Use lightweight file formats like markdown or plain text when possible, avoid uploading large PDFs or slide decks unnecessarily, and write clear prompts that reduce back-and-forth exchanges.
How do I use AI ethically at work?
When important, disclose when AI contributed to content or decisions, fact-check outputs before sharing them, protect sensitive data from public tools, and keep a human reviewing the final result.
Related Links
- AI in Digital Marketing
- How to Write AI Prompts for the Best Results
- AI Resources
- Small Business Marketing Meets AI: Trends, Tools, and Strategies
- AI and Digital Marketing: Helping Small Businesses Stay Competitive
- The Blog Structure Checklist: How to Write for SEO, AI Search, and Real Human Readers
- Inside Vistara 2026: AI Trends, Workflows, and Business Strategy
- How to Future-Proof Your Business with AI: A Practical Guide for SMB Leaders
- Our Work
- Let’s Talk
