Task, Action, Goal — The simplest framework — three words, one clear instruction.
TAG is for genuinely small tasks where anything more elaborate would be overkill: a single email, a short list, a quick calculation, a one-off question. If you find yourself wanting to add a fourth section, the task probably needs RACE or ORACLE instead.
What needs to get done, named directly.
The specific step or output you want, not just the topic.
What success looks like for this specific task.
Bad prompt:
“Help me with my to-do list.”
Structured with TAG:
Task: Organize my to-do list for tomorrow. Action: Group these 8 items into "must do," "should do," and "if time allows." Goal: I want to walk into tomorrow knowing exactly what the top 3 priorities are.
The eight items sorted into the three named buckets, with the top three must-do items called out explicitly at the top so the priority is unambiguous at a glance.
TAG works equally well across ChatGPT, Claude, and Gemini — its brevity means there's little room for model-specific phrasing to matter.