Every AI answer has a bill. Here's the receipt. 🌍
A funny, research-based receipt for every AI answer: electricity, water, CO₂ and tree-time.
Works with Claude, ChatGPT, Gemini, Mistral and local models. Type /planet-receipt.
Install · Commands · How it's calculated · Sources
At the end of an answer, you get this:
🧾 Planet Receipt (estimate) · standard tier · ~2,300 in / ~600 out tokens
⚡ 0.94 Wh · a 10 W LED bulb for 5.6 min
💧 0.94 mL cooling · 5.2 mL incl. power plants (~103 drops)
🌳 0.45 g CO₂ · a tree needs ~11 min to absorb it
A tree just worked 11 minutes overtime for this. Thank the tree, not me. 🌳
The AI pokes fun at itself, never at you. No guilt trips, no lectures, just numbers and a joke.
- Download
planet-receipt.skillfrom the latest release. - In Settings, turn on Code execution and file creation.
- Go to Customize > Skills, click +, then Create skill, and upload the file (rename it to
.zipif the uploader asks for a ZIP). Toggle it on.
Want a receipt on every answer? Paste this into Settings > Profile > personal preferences:
Always run the planet-receipt skill as the final step of every answer.
git clone https://github.com/menepet/planet-receipt.git
bash planet-receipt/skills/planet-receipt/install.sh # on demand: /planet-receipt
bash planet-receipt/skills/planet-receipt/install.sh --always # receipt on every answerCopy the block in portable-prompt.md into Custom Instructions (ChatGPT), a Gem (Gemini) or your system prompt.
| Type | Result |
|---|---|
/planet-receipt |
receipt for the last answer |
/planet-receipt <question> |
answer + one receipt |
/planet-receipt on / off |
receipt on every answer in this chat, or stop |
/planet-receipt why |
show the math with real numbers |
/planet-receipt habit 20 |
"if you did this 20 times a day for a year" |
/planet-receipt total |
text total for the whole chat |
/planet-receipt share |
image card of the last receipt (1080x1350) |
/planet-receipt share total |
image card for the whole chat |
E (Wh) = M × [ 0.0007 × T_out + 0.00022 × T_in × (1 + T_in / 100,000) ]
Water = E × 1.0 mL (cooling) | E × 5.5 mL (incl. power plants)
CO₂ (g) = E × 0.48 | Tree-minutes = CO₂ × 24
T_outincludes hidden reasoning tokens.T_inis everything the model reads this turn; cached context counts at 10%.Mis the model tier: small 0.3, standard 1, large 3.- Every model call counts. Agent loops re-read the whole context on each call, which is why tool-heavy answers cost far more.
Sanity checks against published figures
| Case | This skill | Published |
|---|---|---|
| Typical 500-token answer | 0.35 Wh | Google Gemini median 0.24 Wh (measured), OpenAI 0.34 Wh (stated), Jegham et al. GPT-4o 0.42 Wh |
| 10k-token input | 2.8 Wh | Epoch AI ≈ 2.5 Wh (before data center overhead) |
| 100k-token input | 44 Wh | Epoch AI ≈ 40 Wh (before data center overhead) |
| Water per Wh (cooling) | 1.0 mL | Google ≈ 1.1, OpenAI ≈ 0.9 (implied by their disclosures) |
Full derivations and links: sources.md.
- Estimates, not measurements. About ±3x for normal answers, more for the largest and reasoning models.
- Two water numbers on purpose. Providers report cooling water; researchers argue power-plant water must count too. We show both.
- Location-based carbon (world grid average). Providers with clean-energy contracts report lower market-based numbers.
- Excluded: training, your device, networks, hardware manufacturing. Lifecycle studies that include those come out higher.
- Model sizes are mostly undisclosed. Tiers are inferred; the large-tier multiplier is the least certain number here.
Found a better number? Open an issue with a source. That is the most valuable contribution.
See CONTRIBUTING.md. Factor updates with sources, translations and better jokes are all welcome.
Share your thirstiest prompt with #PlanetReceipt 🧾💧
Built by Menelaos Petousis, co-founder and CTO of dikaio.ai. Made in Athens 🇬🇷
Methodology builds on public work by Google, OpenAI, Epoch AI, Jegham et al., Luccioni et al. (Hugging Face), Li and Ren (UC Riverside), Lawrence Berkeley National Laboratory and Ember. Not affiliated with any of them.
MIT licensed.