How much energy does one ChatGPT prompt use?

Recent published estimates cluster around 0.24–0.34 watt-hours per typical text prompt: Google reports about 0.24 Wh for a median Gemini prompt, Epoch AI estimates roughly 0.3 Wh for a GPT-4o-class query, and OpenAI's Sam Altman has cited about 0.34 Wh. An older, widely-quoted figure of 2.9 Wh is now treated as a high-end bound.

How much CO₂ does an AI chatbot produce?

It depends on how clean your electricity is. One prompt at 0.3 Wh on Sri Lanka's grid (about 590 gCO₂/kWh) emits roughly 0.18 g of CO₂e. Sending 50 such prompts a day for a year works out to about 3.2 kg CO₂e — comparable to driving 27 km in an average petrol car.

Does ChatGPT use more energy than a Google search?

By the most recent first-party figures they are in the same ballpark. A classic Google web search has been quoted at about 0.3 Wh, and a typical AI text prompt is estimated at 0.24–0.34 Wh. Earlier reporting claimed AI used roughly 10× a search, based on the older 2.9 Wh estimate; newer disclosures narrowed that gap considerably.

How much water does an AI query use?

Google's 2025 disclosure puts a median Gemini text prompt at about 0.26 mL of water, mostly for data-centre cooling. This calculator scales that figure proportionally for higher-energy presets. Over a year of 50 prompts a day at 0.3 Wh, that is roughly 5.9 litres — a few large water bottles.

Is using AI bad for the environment?

A single prompt is tiny — a fraction of a watt-hour. The concern is scale: billions of daily queries add up, and training large models plus building data centres carries its own footprint, which this tool does not count. For an individual user, AI chat is a small slice of your digital footprint compared with streaming video or flying.

Does this tool measure training energy too?

No. It models inference only — the energy of sending prompts and getting answers. Training a large model is a separate, one-time cost spread across all its users, and the per-model figures are contested and rarely disclosed. Image, video, and audio generation are also excluded because their per-unit energy differs sharply from text.

Why can I pick different per-prompt figures?

Because no single authoritative number exists yet. Providers measure differently, and estimates have fallen as efficiency improves. Showing the source figures side by side — Google, Epoch AI, OpenAI, and the older de Vries estimate — lets you choose a defensible value and cite it, rather than trusting one hidden assumption.

Where do the grid carbon intensity numbers come from?

They are annual-average figures from Ember and the IEA: Sri Lanka about 590 gCO₂/kWh, the global average about 480, India about 700, the US about 370, and the EU about 250. Sri Lanka's figure swings year to year with hydro output. These were last cross-checked on 2026-06-05; you can also enter your own value.

AI · Sustainability

AI Energy & Carbon Footprint Calculator

Wondering how much energy ChatGPT uses? Enter how many prompts you send and pick your country's grid to estimate the electricity, CO₂e, and water footprint of your AI chatbot use — with every per-prompt figure traced to a published source so the result is citable. Runs entirely in your browser.

By Induwara Ashinsana— Executive Director, Ryzera TechnologiesUpdated Jun 5, 2026

Estimate your AI footprinttext prompts

Sources cited · deterministic

AI model / assistant

Electricity grid

Prompts you send

One prompt = one message you send to the assistant.

Per

Quick presets

Energy / year

5.48 kWh

15 Wh per day · 456.25 Wh per month

Carbon / year

3.23 kg CO₂e

8.85 g per day · 269.19 g per month

Water / year

5.93 L

0.33 mL per prompt

A year of this is like…

456.25

smartphone charges

26.92

km in a petrol car

547.5

hours of a 10 W LED bulb

18,250

classic Google searches

Assumptions used (cite these)

Energy per prompt: GPT-4o-class query — 0.30 WhEpoch AI, How much energy does ChatGPT use? (2025)
Grid carbon intensity: Sri Lanka — 590 g/kWhEmber / CEB, recent annual average
Water per prompt: 0.33 mLGoogle 0.26 mL at 0.24 Wh, scaled linearly by energy

Estimates of inference (per-prompt) energy only — not model training, image or video generation, or data-centre hardware manufacturing. Figures are published averages; real usage varies with prompt length and model load.

How it works

The calculator multiplies one published figure — the energy a single text prompt uses — by how many prompts you send, then converts that energy to carbon using your electricity grid's intensity. Every step is plain arithmetic; the only modelling choice is which per-prompt figure you pick, and that figure is always shown.

Normalise to a year. Prompts per year = prompts × { day: 365, week: 52, month: 12, year: 1 }.
Energy.Energy (Wh/yr) = prompts per year × watt-hours per prompt; divide by 1,000 for kWh. The per-prompt figure comes from your chosen preset — Google's 0.24 Wh for Gemini, Epoch AI's ~0.30 Wh for a GPT-4o-class query, OpenAI's ~0.34 Wh average, or the older 2.9 Wh high-end estimate.
Carbon. CO₂e (g/yr) = energy (kWh/yr) × grid intensity (gCO₂/kWh). Grid intensities are annual averages from Ember and the IEA. The same total is verified a second way — summing (Wh ÷ 1,000 × intensity) over every prompt — so the two routes must agree.
Water. Google reports ≈ 0.26 mL per Gemini prompt at 0.24 Wh. For other presets the tool scales that linearly by the energy ratio, 0.26 × (Wh ÷ 0.24)— a transparent approximation, labelled as such, and switchable off.
Equivalents. Phone charges = energy ÷ 12 Wh; petrol-car km = CO₂ ÷ 120 g/km; LED-bulb hours = energy ÷ 10 W; Google searches = energy ÷ 0.3 Wh. Each constant is sourced in the references below.

The tool models inferenceonly — the energy of using a chatbot. It excludes model training, image and video generation, and the embodied carbon of data-centre hardware, because those figures are model-specific, contested, or out of an individual user's control. Treat the output as a sourced estimate, not a precise meter reading: real energy varies with prompt length, model size, and data-centre load.

Worked examples

50 GPT-4o prompts a day, Sri Lanka grid

0.30 Wh/prompt · 590 gCO₂/kWh · water on

Prompts/year: 50 × 365 = 18,250
Energy: 18,250 × 0.30 = 5,475 Wh = 5.475 kWh/year
Carbon: 5.475 × 590 = 3,230 g ≈ 3.23 kg CO₂e/year
Water: 0.26 × (0.30 ÷ 0.24) = 0.325 mL/prompt → 18,250 × 0.325 = 5.93 L/year
Like: ~456 phone charges, ~27 km driven, ~548 LED-bulb hours

100 Gemini prompts a day, global-average grid

0.24 Wh/prompt · 480 gCO₂/kWh · water on

Prompts/year: 100 × 365 = 36,500
Energy: 36,500 × 0.24 = 8,760 Wh = 8.76 kWh/year
Carbon: 8.76 × 480 = 4,205 g ≈ 4.20 kg CO₂e/year
Water: Gemini preset, no scaling → 36,500 × 0.26 = 9.49 L/year
Like: 730 phone charges, ~35 km driven

Edge case — older high estimate vs newer figure

25 prompts a day, Sri Lanka grid

At 0.30 Wh: 25 × 365 × 0.30 = 2,737.5 Wh = 2.74 kWh → 1.62 kg CO₂e/year
At 2.9 Wh (older bound): 25 × 365 × 2.9 = 26,462.5 Wh = 26.46 kWh → 15.61 kg CO₂e/year
Same usage, ~9.7× the footprint — which is why the source figure matters
Switching the water toggle off zeroes the water line; energy and carbon are unchanged

Frequently asked questions

Sources & references

Per-prompt energy, grid intensity, and water figures were last cross-checked against these sources on 2026-06-05. They are reviewed whenever a provider publishes new first-party measurements or Ember updates its annual grid data.

Related tools

LiveAI

AI Chatbot Cost Calculator

Estimate the monthly API cost of a multi-turn AI chatbot across Claude, GPT, and Gemini. Models the quadratic context re-sending that single-call calculators miss, with and without prompt caching, in USD and LKR.

Open tool

LiveAI

AI Video Cost Calculator

Compare the cost of generating AI video across Sora 2, Veo 3, Runway Gen-4, Kling, Luma, and Pika — per-second, per-clip, and monthly totals in USD and LKR, every price sourced.

Open tool

LiveAI

AI Subscription Cost

Add up what you pay for ChatGPT, Claude, Gemini, Perplexity, Copilot, Cursor, Midjourney and Grok subscriptions across monthly and annual billing, converted to Sri Lankan rupees with an optional 18% VAT line.

Open tool

Rate this tool

Be the first to rate

Comments & feedback

Spotted a bug or want an improvement? Tell us — our team reviews every comment, and good ideas get built. Comments are public and anonymous.

Found a bug, edge case, or a better-sourced figure?

Email me at [email protected] — most fixes ship within 24 hours.