A word is a linguistic unit, while a token is a unit of text used by AI models to process information. One token can be a whole word (“tree”) or part of a word (“compu” + “ter”). For example, “carbon” is one token, but “decarbonisation” may be split into several tokens.

Tokens are how AI models understand and generate language. Every prompt or response is counted in tokens, which directly affect speed, cost, and energy use. The fewer tokens used, the more efficient the interaction.

Optimising tokens reduces unnecessary text processing. This means faster responses, lower costs, and less compute energy making your AI usage both efficient and climate-friendly.

No. Optimising tokens means using only what’s necessary to get accurate results, while increasing tokens adds more text or context, often causing waste. Optimisation is about smart efficiency, not quantity.

When prompts are concise and clear, the AI spends less time and energy processing extra tokens. This leads to quicker outputs, fewer errors, and lower energy use per interaction.

AI emissions come from the electricity and water used by data centres to run large AI models. Reducing them is vital because these models can have a large carbon and water footprint if left unoptimised.

Yes. Every optimised prompt saves compute power. When millions of users optimise, it significantly cuts energy demand and emissions, helping make AI more sustainable.

No — the opposite. DcarbonAI helps users and organisations reduce their AI-related carbon footprint by tracking and optimising token use, turning every AI interaction into a climate-positive action.

Free AI models often come with limited optimisation, inaccurate outputs, data privacy concerns, and higher chances of hallucination (false information). They prioritise access, not efficiency or reliability.

Hallucinations happen when an AI confidently generates incorrect or made-up information. They occur due to incomplete training data, ambiguous prompts, or model overconfidence in generating answers.

Yes, to varying degrees. Even the best models can hallucinate, especially when given unclear prompts or when generating beyond their training data.

More hallucinations mean less trustworthy results. They can mislead users, waste compute resources, and reduce efficiency making it vital to detect and minimise them.

Google ranks pages based on SEO, keywords, and engagement algorithms, often showing what’s popular rather than what’s verified or accurate.

Yes. Both may prioritise SEO-driven or engagement-based results. This can push unverified or biased content ahead of factual sources.

• Pros: Variety of perspectives and faster access to trending information.
• Cons: High risk of misinformation, bias, or inaccurate data, which can mislead users and affect decision-making.