Context & Reference Material

By the end you will wield the single highest-leverage lever in all of prompting.

See it first

Two prompts, same request. One says "Write our Q3 update." The other pastes in the actual Q3 numbers, last quarter's update for tone, and three bullet wins, then says "Write our Q3 update from this." The second is not a little better. It is a different category of result.

This is the highest-leverage skill in prompting, and the one most people skip: giving the model your own context beats clever wording almost every time. The model is not short on language ability; it is short on your specifics. Supply them.

Why context beats cleverness

Recall how models work: they generate likely text from patterns. Without your material, "likely" means "generic average of the internet." With your documents, data, and examples, "likely" becomes "shaped by your actual situation." You are not coaxing a smarter model; you are giving the same model the raw material only you have.

What context to provide

The useful kinds:

Source documents: the report, contract, transcript, or data to work from.
Your own writing: samples so it matches your voice and tone.
Background: who the audience is, what the goal is, what constraints apply.
Examples: one or two instances of the output you want (the next lesson goes deep on this).

What to leave out

More is not always better. Padding the prompt with irrelevant material can dilute focus and bury the signal.

Include

Material directly relevant to the task
The specific data the answer depends on
A short style sample if voice matters
The real audience and goal

Leave out

Tangential documents "just in case"
Confidential data not needed for the task (recall the paste-safe rule)
Huge dumps where a focused excerpt would do
Contradictory or outdated versions

Working with long documents

When inputs are large, a few moves help: paste only the relevant sections, ask the model to summarize a long document first and then work from the summary, or split the work into a chain (summarize → analyze → write). Modern tools with large context windows can take a lot, but focused beats firehose.

What you can now do

Explain why providing context beats clever wording
Choose what material to include for a given task, and what to omit
Handle long documents with summarize-first and excerpting moves
Treat context as your highest-leverage prompting lever

Check your understanding

1 / 2

context-beats-cleverness

Why does providing your own context usually beat clever wording?

What's next

Context is the biggest lever. A close cousin is showing the model examples of what you want. Next: Zero-, One-, and Few-Shot Prompting.