The AI chat that plants trees: how Treechat works

Treechat is an AI chat service that funds one newly planted tree each month for every paid member. It also routes everyday questions towards smaller models and shows an estimated energy, carbon and water receipt for each answer. Planting certificates are published in batches as they land.

Those are three separate parts of the model: reduce likely computation through model choice, make estimated operational impact visible, and fund restoration through paid membership. Treechat does not claim that a planted tree erases a message, that its receipts are direct meter readings or that using the service is impact-free. The aim is a more transparent everyday chat with a simple, checkable funding commitment.

Everyday questions use right-sized models

Many chat tasks—rewriting a paragraph, explaining a concept or brainstorming a short list—do not need the largest available model. Treechat routes routine questions towards smaller, efficient models and uses greater capability when the task requires it.

Smaller models generally perform fewer operations and move less weight data per token than larger dense models under comparable conditions. That makes model choice a practical reduction lever, although it is not a guarantee. Hardware, batching, context, output length and retries also matter. What are small language models? explains those trade-offs.

The product principle is “smallest capable”, not “smallest regardless of outcome”. A weak answer that must be generated repeatedly can waste resources and time. Important answers should still be checked, whichever model produces them.

Model routing is only one part of efficiency. Focused context and proportionate answers also matter, so users retain a useful role without being asked to calculate the footprint themselves.

Each answer has an estimated receipt

Treechat estimates operational energy from the model used and the number of input and output tokens. It adds a stated allowance for data-centre overhead, converts energy with an average grid carbon factor and derives a water estimate from a published factor.

The receipt is not connected to a power meter, cooling tower or named data-centre location. Model providers do not expose those details per response. It also does not allocate the full impact of training, chip manufacture, construction or the user’s device. The current factors, calculation and omissions are published on the methodology page.

Visible estimates are still useful when their limits are clear. They show that a long response is not the same job as a short one and make model choice tangible. They should not be treated as audited life-cycle measurements.

The calculation can change as better public evidence becomes available. Publishing the method gives readers a stable place to see current assumptions rather than freezing an estimate into a timeless promise.

One tree is funded for every month of membership

Every paid Treechat member funds one newly planted tree for every month of membership through verified reforestation partners. The commitment is a direct membership benefit, not a calculated claim that one month of AI use equals one tree’s lifetime carbon storage.

That distinction is essential. Tree growth and survival vary by species, site, climate, maintenance and disturbance. A newly planted seedling does not instantly remove a fixed amount of carbon, and a funding certificate is not an offset credit. Treechat therefore does not subtract projected tree carbon from message receipts or label conversations carbon neutral.

Planting happens in partner batches rather than on each member’s billing date. Certificates are published as those batches land, so users can inspect the evidence available at that stage. How verified tree planting works describes what a certificate can establish and what longer-term monitoring still needs to show.

Batching means membership payments and planting events need not happen at the same moment. Documentary evidence follows the partner’s delivery cycle.

Why planting is kept separate from reduction

Efficient AI and restoration address different things. Smaller models can reduce the computation used for a task. Tree funding directs money towards reforestation. One changes the operational workload; the other supports an environmental project.

Keeping them separate prevents a common accounting shortcut. The Green Software Foundation’s Software Carbon Intensity specification allows a software score to fall through lower energy, lower-carbon electricity or lower hardware emissions—not through offsets. Treechat’s receipt follows that spirit: planting is not deducted from the estimate.

Restoration can bring biodiversity, soil, water and community benefits when it is well designed. It can also fail if trees are unsuitable, land rights are ignored or maintenance stops. Funding a tree is a commitment to support the work, not a guarantee about every ecological result.

This separation makes correction easier. A revised energy factor need not rewrite the planting record, and new survival evidence can update the restoration story without pretending past computation changed.

What “verified” should mean in practice

Verification starts with evidence that funding reached a real partner and was assigned to a planting batch. Useful records identify the partner, project or planting location, date or period, and number of trees funded. They should be clear about whether the count means seedlings planted, trees surviving a later check or something else.

Longer-term quality requires more than paperwork. The UN Food and Agriculture Organization’s forest restoration guidance recommends checking planting quality, monitoring growth and survival, maintaining sites and replanting failed areas where required. Partner selection should also consider ecological suitability and local communities.

Treechat publishes partner certificates as batches land. A certificate documents that stage of the chain; it does not prove future survival or convert the planting into a carbon credit. Any stronger result needs its own evidence over time.

The language should mature with the evidence: funded first, planted when confirmed, and monitored or established only when a partner provides the relevant follow-up.

What Treechat does not promise

Treechat does not promise zero-impact AI, exact per-message measurement or a universal energy saving. It does not claim that every funded seedling survives forever, that planting neutralises fossil carbon or that message volume measures tonnes removed.

The International Energy Agency projects substantial growth in data-centre electricity use in its Energy and AI report. Individual model choices sit inside that larger infrastructure challenge; they do not solve electricity supply, water use or hardware impacts on their own. Providers and policy makers retain the strongest system-level levers.

Treechat’s narrower proposition is meant to be auditable: use proportionate models, state how the receipt is estimated, fund one tree monthly for every paid member and publish batch evidence. For the broader set of choices, see a practical guide to sustainable AI and what green AI means.

An AI chat that plants trees is not a licence to generate without limit. It is a way to make everyday computation visible and connect actual use to a separate restoration contribution—without pretending either part is more certain than the evidence supports.