Task vs Context

Two slots, easy to confuse. Beginners often pile everything into the task slot — facts, history, examples, constraints — and end up with a 200-word run-on prompt that the model interprets as a single confused instruction. This lesson untangles them.

The clean definitions

Slot	Definition	Test for it
Task	The single action you want the model to perform. One verb.	"If I deleted this slot, the model wouldn't know what to do."
Context	Background facts the model needs to do the task well, but is not itself the action.	"If I deleted this slot, the model would still know what to do, but the answer would be more generic or wrong."

Task is one verb. Context is everything the verb needs to operate on intelligently.

Side-by-side example

Compare these two prompts. Both ask for the same thing.

Bad: everything mashed together

Help me write a 90-word warm email to my landlord that asks to extend
my lease by 6 months because I have lived here 2 years and pay rent
on time and have a stable job, and please sign as Sara.

Better: split into task and context

Task: Write a 90-word email to my landlord asking to extend my lease by 6 months.

Context:
- I have lived here 2 years.
- I have always paid rent on time.
- I have a stable job.
- Sign as "Sara".

Output: warm and professional tone. Email body only. No subject line.

The second is longer to read but easier for both you and the model. You can edit the context bullet points without touching the task line. The model sees the action clearly, then enriches it with the facts.

A common mistake

Beginners write context as if it's task. Example:

"Write an email and remember I've lived here 2 years and remember I've paid on time and remember I have a job."

The word "remember" is a clue you've put context inside the task. Strip those out. They go in the context slot.

The opposite mistake also happens: putting the task into context.

"Context: I have lived here 2 years. I want a 90-word email to my landlord. I pay rent on time."

Here the actual task ("write a 90-word email") is hiding among facts. The model has to infer what the action is. Promote it back to its own task line.

Quick checklist

Before you hit send:

1. Read your prompt and try to point at the single verb that names the action. If you can't, the task slot is missing or buried.
2. Look at every other sentence and ask: is this a fact the model needs, or is it telling the model what to do? Facts → context. Instructions → task.
3. If a sentence is doing both, split it.

This habit alone takes most "the model didn't understand me" frustration off the table.

Next: the output spec — the slot that decides whether the answer is paste-ready or a thing you have to reformat by hand. :::