What is Digi Prompt Checker?
A quick way for Oxford students and staff to understand what a prompt will cost before using it in ChatGPT Edu, Copilot or any other AI tool.
The quick check runs entirely in your browser; nothing you type is sent anywhere unless you ask for an AI review. Digi Prompt Checker is a prototype from the Digital Innovation team: its figures are estimates to support reflection, not measurements.
Token use
Counts your prompt exactly and estimates the answer it is likely to generate
Environmental impact
Shows a likely energy and CO₂ range, scaled up to the whole University
Responsible use
Flags personal data, assessment rules and research data, and shows what triggered each flag
Improve my prompt
Suggests a safer, leaner rewrite you can use with one click
Check your prompt
Your results
Tokens
Environment
Ethics
Tokens
Start typing to see an estimate.
Environmental impact
Energy, carbon and an everyday comparison will appear here.
Ethics
Checks for personal data, academic integrity, bias, deception and high-stakes decisions.
This sends your prompt to Claude for a closer reading. The quick check above uses keyword rules and can miss context. The review itself uses a small amount of energy.
Improve my prompt
How the scores are worked out
These are estimates meant to build awareness, not precise measurements. Real usage depends on the provider, data centre, hardware and grid.
| Criterion | Green | Yellow | Red |
|---|---|---|---|
| Tokens (prompt + estimated answer) | under 800 | 800 to 2,500 | over 2,500 |
| Energy per request (middle estimate) | under 0.3 Wh | 0.3 to 1.5 Wh | over 1.5 Wh |
| Ethics | no flags | needs care | serious concern |
Tokens. Your prompt is counted exactly with o200k, the open tokenizer used by GPT-4o and later OpenAI models (the models behind ChatGPT Edu and Copilot), loaded from jsDelivr. If it can't load, the page falls back to a rule of thumb. The answer length is guessed from words like "essay", "summarise" or "in one word" in five languages, and shown as a likely range. Reasoning models add hidden thinking tokens, assumed to be one to four times the answer.
Energy. The range uses 0.05–0.4 Wh, 0.2–1.5 Wh and 0.8–6 Wh per 1,000 generated tokens for fast, standard and reasoning models, with middle estimates of 0.15, 0.6 and 2.5 Wh. Reading the prompt costs about a tenth as much. Public estimates for a typical chatbot query span roughly 0.2 to 3 Wh, so the light follows the middle estimate and the bar shows the spread. Carbon assumes 400 g CO₂ per kWh, because the data centres behind ChatGPT Edu and Copilot are not on the UK grid. Oxford-scale figures assume about 27,000 students and an 8-week term; a kettle boil is taken as 100 Wh.
Ethics. Rules look for patterns in English plus the prompt's detected language (Chinese, Spanish, French or German). Each flag shows the exact words that triggered it, and you can mark a flag as not applying. Phone numbers must look like real UK or international numbers, so lists of years or figures aren't flagged. The rules reflect the University's published position: AI use in summative assessment is unauthorised unless permitted in writing, AI should support rather than replace learning, University work should use SSO-protected tools, and AI use should be acknowledged. They prompt reflection; they are not a ruling. Your tutor's, supervisor's and department's guidance takes precedence.
Improve my prompt. Rewrites use fixed templates, not AI: names become Student A, B…, emails and phone numbers are removed, "write my essay" becomes "help me plan it", decisions become requests for criteria, and long answers get a word limit. Nothing is sent anywhere. Always read a rewrite before using it.
Learn more about responsible AI at Oxford
Read the University's guidance, book training, or contact the Digital Innovation team to discuss your use case.