Skip to content
Get in touchFR
Updated in October 2026
Explained

Does ChatGPT use that much more energy than a Google search?

No. The claim that a ChatGPT query uses 10 times more energy than a Google search comes from an outdated calculation that put ChatGPT at ~3 Wh. Corrected to ~0.3 Wh, the real gap narrows to between x1.5 and x3.

Guillaume Pakula
By Guillaume Pakula, co-founder of Celsius. Since 2020, he has helped 80+ organisations with their Bilan Carbone® and climate strategy.
June 2026
Updated October 2026 · 7 min
No: the idea that a ChatGPT query uses 10 times more energy than a Google search comes from a 2023 calculation that overestimated ChatGPT's consumption at ~3 Wh. Recalibrated to ~0.3 Wh, the real gap narrows to between x1.5 and x3. Google is also adding AI to its own search, which erases what remained of the boundary between the two. What drives the electricity consumption of data centres is the volume of queries, far more than the cost of each one.
Key takeaways
  • 1The "x10" claim is a myth: it rests on an outdated estimate, since corrected downwards by a factor of 10.
  • 2ChatGPT ~0.3 Wh, Google search ~0.04 to 0.3 Wh: the real gap is x1.5 to x3.
  • 3AI Overviews blur the boundary: Google search now generates AI answers too.
  • 4Volume matters most: global token volume is projected to grow 24-fold by 2030.

The energy gap between a ChatGPT query and a Google search ranges from x1.5 to x3, depending on the reference point used: both remain within the same order of magnitude. The "factor of 10" claim is that typing a question into ChatGPT would be 10 times worse for the environment than searching for it on Google. It comes from a 2023 calculation that overestimated ChatGPT's energy use by a factor of 10. This article reviews the figures for the 2 services, explains the origin of the error and then looks at how query volume affects their environmental footprint.

ChatGPT and Google

Energy, water and CO2e of an AI query and a Google search

Choose the scale: a single query, or 2.5 billion queries a day.

AI queryChatGPT, Gemini, short text question
Google searchtraditional search, excluding AI summaries
Energy
~0.3 Wh
0.04-0.3 Wh
Epoch AI; Gemini measured at 0.24 Whmodern estimate ↔ official 2009 figure
Water
0.26 mL≈ 5 drops (Google measurement)
not publishedno figure per search
CO2e
0.02-0.12 gdepending on the data centre country
~0.2 g2009 figure, never updated

At this scale. Fractions of a Wh and a gram on both sides: the actual gap is between x1.5 and x3 depending on the reference, never x10. The query-versus-query debate is poorly framed.

Sources: Epoch AI (2025); Google, 2009 blog and arXiv 2508.15734; OpenAI, How People Use ChatGPT (NBER, 2025); IEA, Energy and AI; Goldman Sachs Research. French household ≈ 5 MWh/year, bathtub ≈ 150 L. Orders of magnitude, not audited measurements.

We start with the figures for the 2 sides and where each comes from.

AI: ~0.3 Wh per query, estimated then measured

A short text query to ChatGPT uses roughly 0.3 Wh. This comes from a calculation by Epoch AI, built on deliberately pessimistic assumptions, which concludes that the earlier reference figure of ~3 Wh was 10 times too high. It is corroborated by the only production measurement (in French) ever published by a major provider: Google instrumented its own infrastructure and measured 0.24 Wh for the median Gemini query (arXiv 2508.15734). The breakdown by model and by prompt length is in how much energy a ChatGPT query uses.

Google search: between 0.04 and 0.3 Wh, depending on the benchmark

The only official figure ever published for a Google search dates from 2009: 0.3 Wh (0.0003 kWh, Google blog). Servers have become more efficient since then, and recent academic estimates are closer to ~0.04 Wh (Vanderbauwhede, 2023-2025); Google itself has not published an update.

Depending on the reference point chosen, the gap between AI and search therefore ranges from parity to a factor of a few. Taking the middle of the range for search (roughly 0.1 to 0.2 Wh), the gap is between x1.5 and x3, far from x10. ChatGPT and a Google search remain within the same order of magnitude, separated by only fractions of a Wh.

Orders of magnitude

Energy used by a ChatGPT, Gemini or Google query, in Wh

Each bar is drawn to scale, in Wh per query.

The original "x10" calculation

The original "x10" calculationChatGPT, 2023 estimatenever measured, since invalidated~3 Wh
Google searchofficial 2009 figure~0.3 Wh

Recent measurements

Recent measurementsChatGPT queryEpoch AI~0.3 Wh
Gemini queryGoogle measurement, median~0.24 Wh
Google searchfrom the modern estimate to the 2009 figure~0.04 to 0.3 Wh

For a single query, the 2 are in the same category: the actual gap ranges from ~1x to a few times, depending on the reference used, never a clear x10. AI Overviews now add AI inference to Google search itself.

Epoch AI (ChatGPT ~0.3 Wh); Google 2009 (search ~0.3 Wh, the only official figure); arXiv 2508.15734 (Gemini ~0.24 Wh). Orders of magnitude, rather than audited measurements.

2Why does generating with ChatGPT use a little more energy than searching Google?

It is no mystery that AI uses a little more energy per query: searching and generating are 2 different operations. A Google search draws on an index that has already been built. The engine indexed the web in advance, and your query is only used to find the right pages in that vast table of contents, which takes little computation.

How it works

Why does an AI answer use more energy than a search?

Google finds an already indexed page. AI builds its answer, word by word, which requires more computing.

Search (Google)
  • Uses an existing index
  • Finds the right pages

Little computing

Generate (AI)
  • Builds its answer, token by token (inference)
  • Each generated token = electricity

More computing (x1.5 to x3)

Celsius, based on IEA, Energy and AI (2025).

A generative AI does not retrieve anything ready-made: it builds its answer from its model. This process, known as inference (running a trained model to produce an answer), takes more computation than a search in an index, which explains the slightly higher cost per query.

It comes down to tokens

AI models process text as tokens, which are fragments of words (about 0.75 of a word in French). To answer, a model produces these tokens one by one, and every token generated uses electricity. The longer the answer, the more it uses, and the same question can use very different amounts of electricity depending on the model that handles it. AI consumption is therefore better measured per token than per query, and the choice of model matters a great deal: the ChatGPT, Claude and Gemini comparison puts figures on these gaps model by model.

What about water use?

Water follows the same pattern of ranges. Again, the only published measurement comes from Google: 0.26 mL per median Gemini query, equivalent to five drops, mostly used to cool the servers. For a conventional search, no per-query figure has ever been published. Here too the gap per query is negligible: the water footprint of an AI query explains where this figure comes from and why the location of the data centre matters more than the logo.

3Where does the "10 times more" claim come from, and what about AI Overviews?

The factor of 10 comes from an outdated calculation that kept circulating after it had been corrected. There are 3 reasons for this.

Origins of the "x10"

Energy-per-query figures over time, ChatGPT and Google

Choose a step to read what happened.

0123 Wh20092013201720212025x10

ChatGPT estimated at ~3 Wh: the "x10" is born

A study assumes very long responses (2,000 tokens) on older-generation GPUs. Compared with the 2009 figure: "ten times more than a search".

The "x10" compares an invalidated figure with a figure from 2009: recalibrated against current references, the actual gap is x1.5 to x3.

Google, official blog (2009); de Vries, Joule (2023); Epoch AI (Feb. 2025); arXiv 2508.15734 (Aug. 2025); Google I/O, May 2026 (AI Overviews).
  • It rests on an outdated figure. In 2023, a study put a ChatGPT query at ~3 Wh, assuming very long responses (2,000 tokens) on older-generation graphics processors (GPUs). Set against Google's 0.3 Wh, that gave "10 times more". The ~3 Wh figure has since been invalidated, but the factor kept circulating without its correction.
  • Google now builds AI into search. AI Overviews (the AI-generated summaries shown at the top of results) are seen by more than 2.5 billion users a month, and AI Mode has more than 1 billion monthly users (Google, May 2026). A "Google search" therefore increasingly generates an AI summary, which involves the same type of inference as ChatGPT, so the boundary between searching and generating is fading.
  • There is little transparency. The only official figure for search dates from 2009, and on the AI side only Google has published a production measurement. Old data is being compared with third-party estimates, which is all the more reason to think in orders of magnitude rather than in terms of a precise multiple.

The x10 therefore described a world that never quite existed, one in which ChatGPT used 10 times more. Recalibrated against today's figures, and as search takes on more and more AI, the gap blurs until it disappears, and the question "ChatGPT or Google?" loses some of its meaning.

More than 2.5 billion people see AI-generated summaries in Google search each month, so comparing ChatGPT with "a search" increasingly means comparing 2 AI systems.

4Query volume matters more than the cost of a single query

In each of the 2 cases, a single query is tiny: fractions of a Wh and a fraction of a gram of CO2 (the carbon intensity depends on the country where the data centre is located: ~41 gCO2e/kWh in France, ~384 in the United States and ~458 as a global average in 2025, according to data from Ember, an energy think tank). Even intensive daily use amounts to between a few hundred grams and a few kilograms of CO2e over a year: nothing comparable to a single return flight. The AI carbon footprint calculator gives this annual total for your own level of use.

Rows of server racks in a data centre: this is where the aggregate volume of AI queries and web searches takes physical form
A single query is invisible, a billion a day is not: data centre electricity consumption is set to more than double between 2024 and 2030 (IEA, Energy and AI).

The environmental burden comes from multiplication. ChatGPT handles around 2.5 billion messages a day (18 billion a week in summer 2025, according to OpenAI) and Google handles about 14 billion searches a day, which adds up to more than 5 trillion a year; global token volume is projected to grow 24-fold between 2026 and 2030 (Goldman Sachs Research); and data centre electricity consumption is set to rise from 415 to ~945 TWh between 2024 and 2030, according to the International Energy Agency (IEA) in Energy and AI.

Whether a query costs 1.5 or 3 times as much as a search says nothing about this dynamic. The same logic applies to a company's digital carbon footprint and to a website's carbon footprint: an email or a page view counts for very little on its own, and it is their number, together with the hardware that serves them, that adds up.

What each query delivers also has to be taken into account. An AI query often does the work of several Google searches in one go (it summarises, rephrases and calculates), where you would otherwise have had to open 5 tabs. Comparing one AI query with a single search can therefore be misleading in both directions.

At Projet Celsius, we find it more useful to ask how many tokens are used and what value they deliver than to choose between ChatGPT and Google: do not have the AI generate long blocks of text for a trivial question, and reserve AI for tasks where it adds value. That is also what makes AI manageable in a carbon footprint, under Scope 3.

5Key takeaways

  • The "x10" claim is a myth, stemming from a 2023 calculation that overestimated ChatGPT's consumption at ~3 Wh. Corrected to ~0.3 Wh, the gap with a Google search narrows to between x1.5 and x3, depending on the reference point chosen.
  • The figures converge: ChatGPT ~0.3 Wh (Epoch AI), Gemini 0.24 Wh (Google measurement); Google search from ~0.04 Wh (recent estimates) to 0.3 Wh (the only official figure, 2009).
  • Generating costs a little more than searching, because Google draws on a ready-made index whereas AI builds its answer token by token. The gap with search remains a small multiple, while a factor of 100 separates a short question from a long prompt on a reasoning model.
  • Google is adding AI to search (AI Overviews, more than 2.5 billion users a month; AI Mode, more than 1 billion): the boundary between searching and generating is fading.
  • Volume matters far more than the individual query: billions of queries a day, token volume projected to grow 24-fold by 2030, and data centre electricity consumption rising from 415 to ~945 TWh. It is therefore worth choosing the tool to suit the need.

The question to ask is therefore less how ChatGPT compares with Google than whether the use itself is worthwhile: does this use of AI bring me enough to justify its tokens? To put all this in perspective at the scale of an organisation, the overview of the carbon footprint of AI places it below 1% of an SME's carbon footprint, even with intensive use. If you need to account for it, it falls under Scope 3 of your carbon footprint.

Further resources

Frequently asked questions

or: guillaume@projetcelsius.com

More articles to read

View all →
French environmental cost label for clothing: what changed on 1 October 2026
Regulation

French environmental cost label for clothing: what changed on 1 October 2026

6 min read
What is the French textile eco-score?
Who displays it, and has it become mandatory?
Regulation

What is the French textile eco-score? Who displays it, and has it become mandatory?

12 min read
Removable batteries and battery passports: who is affected in 2027?
Regulation

Removable batteries and battery passports: who is affected in 2027?

14 min read

A senior consultant reads every message

Tell us where you stand and what is pressing, and we will reply within 24 working hours with a first view on how to proceed.

Write to a consultant →
Our toolkit

A full range of free assessment tools

Cost estimator, obligation and eligibility checkers, footprint calculator.

Estimate costs and impacts
Ref. 2026·BC·0184REPORT · FINANCIAL YEAR 2026Bilan Carbone®complete 1·2·3Presentation to the Executive Committee · 2027 action planCONTENTS01Executive summaryp. 402Scope and methodologyp. 1203Emissions by scopep. 24042027 action planp. 5605Methodological appendicesp. 78VOLUME84 pages · ConfidentialSTANDARDISO 14064 · GHG ProtocolQUOTATION · BC.2026.0184Issued 05·05·2026 · Valid for 30 daysCelsius consultancy · Paris 3rd arrondissementINDICATIVE COST ESTIMATEYour companyMid-sized company · 180 employees · CSRD 2027SERVICEBilan Carbone®Scope 1·2·3 · 13 weeksBPIFRANCE FUNDING · - 60%INDICATIVE RANGE EXCL. VAT€16,800to €22,400PROJECT DETAILS01 · SCOPING3 weeks~€4,20002 · MODELLING8 weeks~€9,80003 · PRESENTATION2 weeks~€4,800Indicative estimate · Refined after scopingSIRET 891 234 567 00012

Bilan Carbone® cost estimator

Your price range in 1 minute, based on 2026 market rates and public funding.

1 minStart
Estimate costs and impacts
AI FOOTPRINT · MONTHLY USEYOUR TOTAL FOOTPRINT4.2 kgCO₂eq for 1,000 typical prompts · May 2026MODEL COMPARISON · 1,000 TYPICAL PROMPTSGPT-4oOpenAI6.8 kgClaude OpusAnthropic2.1 kgGemini ProGoogle4.5 kgMistral LargeMistral AI · FR1.6 kg

AI carbon footprint calculator

The climate impact of your AI queries, by model and task. Ecologits methodology.

3 minStart
Check an obligation
DIAGDÉCARBON'ACTIONACCELERATE THETRANSFORMATIONOF YOUR COMPANYIN ENERGY ANDTHE ENVIRONMENTOFFICIAL SCHEME · BPIFRANCE × ADEMECHECKER · ELIGIBILITYELIGIBLEYou meet the 3 Diag Décarbon'Action criteria.YOUR FUNDING€10,000 excl. VAT€6,000excl. VATAmount payable by you after the Bpifrance subsidyFUNDING BREAKDOWNBPIFRANCE 40%YOU 60%€4,000€6,000SCHEME SUBSIDISED BY

Diag Décarbon'Action eligibility

Check in 30 seconds whether your business is eligible for Bpifrance funding covering 40% of a Bilan Carbone® assessment.

30 secStart
Check an obligation
FRENCH REPUBLICMinistry for the Ecological TransitionLEGAL REQUIREMENTGreenhouse GasEmissions AssessmentArticle L. 229-25 of the French Environmental CodeDecree 2022-982 · published 1 July 2022FREQUENCY · 4 YEARSFILING · ADEMECHECKER · BEGES REQUIREMENTREQUIRED720 employees · BEGES requirementPOTENTIAL PENALTY€50,000maximum fine for failing to complete itart. L. 229-25OFFICIAL REFERENCE

BEGES checker

Does the BEGES requirement apply to you? An immediate answer, with the deadline and penalty.

30 secStart
Check an obligation
9:42SCANSCANNING DPP...DPP IDENTIFIEDOrganic cotton T-shirtSKU TX-CB-220 · Batch L-26-04781ESPR COMPLIANT42 DATA POINTS · 5 SECTIONSORIGINIndia · GOTSOrganic cottonFOOTPRINT5.2 kg CO₂eqLCA ISO 14040RECYCLABILITY85%Cotton recycling schemeREPAIRABILITY7.5 / 10Accessible partsEND OF LIFETextile recycling scheme (TLC) · Bordeaux/FRRefashion-approved recovery operatorEU · ESPR 2024/178112.05.2026

DPP checker

Is your product covered by the Digital Product Passport?

1 minStart
Check an obligation
COSTENVIRONMENTAL386POINTS257PER 100 GOfficial methodEcobalyse, v7.0.0CHECKER · ENVIRONMENTAL LABELLINGELIGIBLETextiles · voluntary labellingENVIRONMENTAL COST386 ptsT-shirt 150 g · 257 pts per 100 gExample: Ministry for the Ecological Transition FAQ

Textile environmental labelling 2026

Voluntary labelling, but third parties can publish it from October 2026: where do you stand?

1 minStart