# Green AI lesson and activity ideas

> Nine low-cost green AI activities for ages 9 to 18: energy estimates, paper tokens, data centre siting, debates and a rebound game. No paid tools.

Source: https://greenai.guide/schools/lesson-ideas/  
Updated: 2026-11-19

## In short

These nine activities teach the energy and environmental side of AI using paper, published figures and discussion, for pupils aged roughly 9 to 18. Each has an age band, timing, learning objectives and resources, and none needs paid software or pupil accounts on AI tools.

## Key points

- Every activity works without pupils using AI tools themselves, which respects the age limits many tools set.
- Estimation activities use published per-prompt figures from Google and OpenAI, quoted with their source.
- The activities fit existing computing, science, geography, maths and citizenship content rather than needing new curriculum time.
- Curriculum links are indicative: check them against your own national or state curriculum.

## How should I use these ideas?

Pick one, adapt the timing to your class, and quote any figure with its source and year. Background for every
activity is in the [teachers' guide](/schools/teachers/), and terms are defined in the [glossary](/glossary/).
None needs pupils to have accounts on AI tools.

Curriculum links use England as the example (KS2 is ages 7–11, KS3 11–14, KS4 14–16). They are indicative,
not official mappings; most countries have equivalent strands on computer systems, energy, natural resources and
responsible technology use.

## Activity ideas

### 1. How much energy does our class use on AI?

**Ages:** 11–16 · **Time:** 50 minutes · **Links:** KS3 science (energy, power, kWh, domestic fuel costs); maths
(estimation, units); computing (impact of technology)

**Objectives:** convert between Wh and kWh; make and justify a Fermi estimate; explain why small amounts add up.

**Resources:** paper, calculators, the published figures below.

**How:** pupils estimate how many AI text prompts the class makes in a week (anonymous show of hands works). Using
Google's 0.24 Wh for a median Gemini text prompt and OpenAI's 0.34 Wh for an average ChatGPT query, they work out
a range. For example, 30 pupils × 10 prompts × 0.24–0.34 Wh gives roughly 72–102 Wh a week. Then scale to the
whole school, the country and the world, and compare with appliances whose power rating is on the label. Finish
by discussing what the figures leave out: images, video, training and the devices themselves.

### 2. Tokens explained with paper

**Ages:** 9–14 · **Time:** 30 minutes · **Links:** KS2 and KS3 computing (how data is represented)

**Objectives:** explain what a [token](/glossary/#token) is; explain why longer text means more computing.

**Resources:** printed sentences, scissors, a stopwatch.

**How:** pupils cut sentences into chunks roughly the size of tokens. OpenAI's rule of thumb is that a token is
about four characters, or three-quarters of an English word, so long words split. A "model" (one pupil) must
read every chunk before adding one new word, timed. Doubling the sentence doubles the work. Older pupils can try
a free online tokenizer demonstration on the projector and notice that unusual words and some languages split
into more pieces.

### 3. Is AI the right tool?

**Ages:** 11–18 · **Time:** 40–60 minutes · **Links:** KS3 and KS4 computing (responsible use); citizenship

**Objectives:** weigh benefits and costs of a technology; argue from evidence; understand
[sufficiency](/glossary/#sufficiency).

**Resources:** task cards (for example: check a spelling, write a birthday poem, find the capital of Peru, summarise
a 40-page report, detect skin cancer in images, make 25 versions of a logo).

**How:** in groups, pupils sort cards into "AI is the best tool", "something simpler works" and "it depends", then
defend one choice to the class. Discuss what "frivolous" means and who gets to decide.

### 4. Where should a data centre go?

**Ages:** 11–16 · **Time:** 60 minutes · **Links:** KS3 geography (economic activity and natural resources;
human and physical interaction; maps and GIS)

**Objectives:** identify factors in siting decisions; weigh trade-offs; use maps and data.

**Resources:** printed or free online maps showing renewable generation, rivers, temperature, grid connections and
towns.

**How:** pupils act as a planning committee choosing between three fictional sites: cool and windy but remote;
near a city with water stress; next to a solar farm in a hot region. Score each on power, cooling, water, land,
jobs and local impact. Use Ireland as a real case: data centres used 22% of the country's metered electricity in
2024, up from 5% in 2015 (Central Statistics Office).

### 5. The efficiency and rebound game

**Ages:** 11–18 · **Time:** 40 minutes · **Links:** KS4 computing (impact of technology); economics; maths

**Objectives:** explain the [rebound effect](/glossary/#rebound-effect) and the
[Jevons paradox](/glossary/#jevons-paradox).

**Resources:** counters or tokens, a simple rule sheet.

**How:** each team has a fixed "energy budget" of counters. Each round, AI gets twice as efficient (a task costs
half as many counters), but teams then choose how many tasks to run. Record total energy each round. Pupils
usually find total use rises. Link to the economist William Stanley Jevons, who argued in 1865 that more efficient
steam engines would increase coal use, not reduce it. Ask: what rules would keep total use down?

### 6. Text, image or video?

**Ages:** 13–18 · **Time:** 40 minutes · **Links:** KS3 science (energy); computing

**Objectives:** compare the energy of different AI tasks; read a research figure critically.

**Resources:** the figures below, printed.

**How:** share one research finding: in a 2024 study, generating 1,000 images with the least efficient model
tested used 11.49 kWh, as much as 522 smartphone charges, while the most efficient text model used as much as 9%
of one charge for 1,000 runs (Luccioni, Jernite and Strubell). Pupils discuss why images cost more, what "most"
and "least efficient" mean, and how they would design a fair test.

### 7. Design a green AI label

**Ages:** 9–16 · **Time:** 50 minutes · **Links:** design and technology; computing; citizenship

**Objectives:** understand [energy labels](/glossary/#energy-label); decide what information is useful and honest.

**Resources:** examples of appliance energy labels and food labels, paper, pens.

**How:** pupils design a label to show next to an AI answer or app. What should it show: energy, water, carbon,
where it was computed? How would you stop companies making misleading claims
([greenwashing](/glossary/#greenwashing))? Older pupils can read [energy labels for AI](/articles/energy-labels-for-ai/).

### 8. Prompt golf

**Ages:** 13–18 · **Time:** 30 minutes · **Links:** computing; English (concise writing)

**Objectives:** write clear, specific instructions; connect brevity to efficiency.

**Resources:** paper; optionally one teacher-run AI demonstration.

**How:** pupils rewrite long, vague requests into the shortest prompt that still gets exactly what is needed (who it
is for, length, format). Count words, then estimate tokens. The teacher can test the best two live. Discuss why
one clear prompt beats five attempts.

### 9. Debate: should a country build its own green AI?

**Ages:** 15–18 · **Time:** 60 minutes · **Links:** citizenship; politics; computing

**Objectives:** construct arguments on technology policy; consider sovereignty, cost, energy and fairness.

**Resources:** a short briefing, such as the [sovereign AI and sustainability](/articles/sovereign-ai-and-sustainability/)
article.

**How:** the motion is "This house would build a national, publicly owned AI service powered by renewable
energy". Teams argue equity of access and control of data against cost, energy, infrastructure and political
risk. This question was debated at a 2026 UK green AI sandpit of mathematicians, computer scientists and
industry, where opinions were divided, which makes it a good open question.

## How can I assess learning?

Short written answers work well: "Explain why a single prompt uses little energy but AI as a whole uses a lot" or
"Give one way to make AI use more efficient and one way to make it more sufficient". Look for reasoning about
scale and for both halves of green AI.

## What to do next

- Try activity 2 or 3 this term; neither needs technology.
- Share [Green AI for pupils](/schools/for-pupils/) as pre-reading or homework.
- Check facts and misconceptions in the [teachers' guide](/schools/teachers/) before the lesson.
- Put the [glossary](/glossary/) on the board for key terms.
- Tell us how an activity went through the [contact page](/contact/), so we can improve these ideas.

## Frequently asked questions

### Do these lessons need computers?

Most need none. A few optionally use a free web page, such as a tokenizer demonstration or an online map, which the teacher can show on a projector.

### Which activity should I start with?

Paper tokens (activity 2) is the quickest: 30 minutes, no technology, and it explains why longer prompts and answers use more computing. Is AI the right tool? (activity 3) works well as a follow-up.

### Are these linked to the national curriculum?

They support existing content in England's computing, science and geography programmes of study and similar strands elsewhere, but they are not official resources and are not tied to any exam specification.

## Sources

- [Measuring the environmental impact of delivering AI at Google scale](https://arxiv.org/abs/2508.15734), Google (arXiv:2508.15734) (2025)
- [The Gentle Singularity](https://blog.samaltman.com/the-gentle-singularity), Sam Altman (OpenAI) (2025)
- [What are tokens and how to count them?](https://help.openai.com/en/articles/4936856-what-are-tokens-and-how-to-count-them), OpenAI Help Center (2025)
- [Energy and AI](https://www.iea.org/reports/energy-and-ai), International Energy Agency (2025)
- [Data centres metered electricity consumption 2024](https://www.cso.ie/en/releasesandpublications/ep/p-dcmec/datacentresmeteredelectricityconsumption2024/keyfindings/), Central Statistics Office, Ireland (2025)
- [Power hungry processing: watts driving the cost of AI deployment?](https://arxiv.org/abs/2311.16863), Luccioni, Jernite and Strubell (arXiv:2311.16863) (2024)
- [The coal question (1865), summary and extracts](https://energyhistory.yale.edu/w-stanley-jevons-the-coal-question-1865/), Yale University, Energy History (1865)
- [National curriculum in England: computing programmes of study](https://www.gov.uk/government/publications/national-curriculum-in-england-computing-programmes-of-study), Department for Education (2013)
- [National curriculum in England: geography programmes of study](https://www.gov.uk/government/publications/national-curriculum-in-england-geography-programmes-of-study), Department for Education (2013)
- [National curriculum in England: science programmes of study](https://www.gov.uk/government/publications/national-curriculum-in-england-science-programmes-of-study), Department for Education (2013)
