Our Environmental Footprint
We use AI to help you and your clients after every session. Here's an honest look at what that costs the planet — and what we're doing about it.
Yes, AI uses energy. Here's how much ours uses.
What does one session cost the planet?
When Parts Companion processes a therapy session — transcribing it, identifying parts, writing summaries, creating images — it uses about as much energy as leaving a light bulb on for five minutes. That's real, and we think you deserve to know about it. But it's also quite small: about 45-seconds of streaming Netflix.
You might be wondering whether using AI-powered tools in your practice comes with an environmental cost. The short answer is yes — but it's a small one, and we're committed to keeping it that way.
This page explains where that energy goes, puts it in context against things you already do every day, and shares the specific steps we're taking to reduce our footprint over time.
What happens after your session
When a session recording is processed, it moves through several steps. Each step uses AI to read, understand, and create something useful for you and your client.
Transcription
The audio recording is converted to text so AI can read it.
Part Identification
AI reads the transcript and identifies the parts that showed up in the session.
Understanding & Detail
Each part is explored more deeply — what it said, how it felt, what it needs.
Summaries & Notes
Session summaries are written for both you and your client, each in the right voice.
Part Images
Unique images are created for each part, giving clients a visual connection to their inner world.
Additional features that use AI
Beyond the initial session processing, Parts Companion also uses AI when you chat with it for clinical consultation, when it writes SOAP notes, when it creates guided meditations, and when it generates practitioner feedback. Each of these features uses a small amount of additional energy.
Three layers of AI energy
There are three different kinds of energy cost in AI, and it's helpful to understand the difference.
Building the AI
- 🏗️ The AI models we use (like Claude) were built once by companies like Anthropic
- 🌍 Building them used a lot of energy — but it's a one-time cost, like building a road1
- 👥 That cost is shared across millions of people and thousands of apps that use the same AI
- 📊 Our tiny share of this cost is negligible — thousandths of a gram of CO₂ per session
Building with AI
- 💻 We use AI coding assistants (Claude, GPT-4, Gemini) and voice synthesis (ElevenLabs) to build Parts Companion itself8
- ⚡ This is a one-time development cost — once a feature is built, the energy is spent
- 📊 Estimated total development energy: ~150 kWh over the project lifetime, or about 60 kg CO₂
- 🔄 That's roughly equivalent to driving 150 miles — spread across the entire product
Using the AI
- 🔄 Every time we process a session or answer a question, we use a small amount of energy
- 🎯 This is the part we control — and the part we're focused on reducing2
- 📈 It grows as more practitioners use Parts Companion
- 📊 Currently about 2 grams of CO₂ per session — the weight of a small paperclip
Think of it like building a house
Someone manufactured the bricks (training). We used those bricks to build the house (development). Now people live in it every day (inference). The brick factory's energy is shared across every building in the world. Our construction cost was a one-time investment. What we focus on is making the house as energy-efficient as possible for everyone who uses it.
What one session costs the planet
Here's the environmental footprint of processing a typical 50-minute therapy session through Parts Companion.
Energy Used
About 5 watt-hours3. A watt-hour is a small unit of energy — your phone uses about 15 watt-hours to fully charge. So processing one session uses about a third of a phone charge.
Carbon Produced
About 2 grams of CO₂4. For comparison, driving your car one mile produces about 400 grams. So one session equals about 20 feet of driving.
Water Used
About 10–20 milliliters — roughly a tablespoon5. AI data centers need water for cooling, much like your computer's fan keeps it from overheating, but with water instead of air.
How does that compare to everyday life?
These comparisons help put the numbers in perspective.
Where the energy goes
Different features use different amounts of energy. Here's a breakdown of what each part of Parts Companion costs the environment when you use it.
Session Processing
Chat Consultations
SOAP Notes
Guided Meditations
Practitioner Feedback
Part Images
If you use every feature on a single session
If you process a session and then use the chat, generate SOAP notes, create a meditation, and run the feedback assessment, the total comes to about 12–20 watt-hours of energy and 5–8 grams of CO₂. That's still less than 25 minutes of a light bulb, or about 6 minutes of streaming video.
What happens as we grow?
Each session's footprint is small, but as more practitioners use Parts Companion, the total adds up. Here's what that looks like — and why we think the numbers are reassuring.
For perspective
Even if Parts Companion were processing tens of thousands of sessions every month, the total annual carbon footprint would be less than a single round-trip flight from New York to London. A single person commuting to work by car for a year produces about 25 times more CO₂ than Parts Companion would at that scale.
AI is getting greener, fast
One piece of genuinely good news: AI technology is becoming dramatically more efficient every year. The energy needed to perform the same task has dropped by over 100× in just a few years7, and that trend is continuing. Global data centers currently use about 1.5% of the world's electricity6 — but the AI tools we'll be using a year or two from now will do the same work with far less energy than they use today.
What we're doing about it
We believe in being upfront about our environmental impact — and in actively working to reduce it. Here are the specific steps we're taking.
We measure everything
Every time Parts Companion uses AI, we log exactly what it used and how much it cost — in both dollars and energy. You can't improve what you don't measure, so we track our environmental footprint with the same rigor as our financial costs.
We use the smallest AI that works
Not every task needs the most powerful AI. For simpler tasks, we use smaller, faster models that use 40–60% less energy. We reserve the more powerful models only for tasks that truly need them, like understanding the nuances of a therapy session.
We avoid doing the same work twice
When we've already processed something — like generating your SOAP notes — we save the result. The next time you view it, we serve the saved version instead of generating it again. This eliminates unnecessary energy use entirely.
We advocate for transparency
The AI companies we work with (like Anthropic) don't yet publish detailed environmental data. We believe they should, and we created this page to signal support for greater disclosure. In the meantime, we're being as transparent as we can with the data available to us.
Small footprint, real responsibility
Using Parts Companion has a real environmental cost — but it's a modest one. Processing a therapy session uses about as much energy as leaving a light bulb on for five minutes. Even if every therapist in the country used Parts Companion for every session, the total would be a tiny fraction of what we all produce just by driving to work.
We share this not to dismiss the impact, but because we believe you deserve to make informed choices about the tools you use in your practice. Environmental responsibility matters to us, just as it matters to many of you.
We're committed to keeping this footprint as small as possible — by choosing efficient technology, by avoiding wasted computation, and by being transparent with you about the trade-offs. As AI technology continues to get more efficient, these numbers will only go down.
References & methodology
- 1.Patterson, D., et al. "Carbon Emissions and Large Neural Network Training." arXiv:2104.10350, 2021. arxiv.org
- 2.Rillig, M.C., et al. "Risks and Benefits of Large Language Models for the Environment." Nature Scientific Reports, 2024. nature.com
- 3.Amri, S., et al. "How Hungry is AI? Benchmarking Energy and Carbon Footprint of LLM Inference." arXiv:2505.09598, 2025. arxiv.org
- 4.Luccioni, A.S., et al. "Power Hungry Processing: Watts Driving the Cost of AI Deployment?" ACM FAccT, 2024. doi.org
- 5.OECD. "How Much Water Does AI Consume? A Review of the Evidence." 2024. oecd.org
- 6.International Energy Agency. "Energy and AI." IEA, 2025. iea.org
- 7.Li, J., et al. "TokenPowerBench: Benchmarking LLM Energy Efficiency." arXiv:2512.03024, 2024. arxiv.org
- 8.Development energy estimated from API usage logs across Anthropic, OpenAI, Google, and ElevenLabs during Parts Companion development (2025–2026). Internal data.