Context is everything: audience, purpose and the reasons behind your rules
The model knows the world but not your world. How to give it the background it needs, and why explaining "why" beats shouting "never".
In the first article we saw that a good prompt answers four questions: who, what, why and how. This article zooms in on the one people skip most often: context.
An AI model has read an enormous amount of text. It knows how emails, reports and lesson plans usually look. What it does not know is your situation: your audience, your constraints, what you have already tried. Context is everything that turns a generic answer into a useful one.
Three kinds of context
1. The audience
The same explanation should look very different for a child, a student and an expert.
Explain how vaccines work.
Explain how vaccines work to a 12-year-old who is nervous
about getting a shot next week. Use one everyday comparison
and keep it reassuring but honest.
The second prompt tells the model the reader's age, their emotional state and the purpose of the explanation. Each of those details changes the vocabulary, the length and the tone.
2. The purpose
Why do you need this? A summary of a report for your own revision is not the same as a summary you will present to your board.
Summarise this report.
Summarise this report for our board meeting on Thursday.
The board cares about cost and risk, not technical details.
They will have two minutes to read it.
Now the model knows what to keep, what to cut and how short to be.
3. The background
Give the facts the model cannot know: your product, your previous attempts, the constraints you work under.
Our newsletter has a 12% open rate. We send it on Monday
mornings to 4,000 small-business owners. We tried shorter
subject lines and it didn't help. Suggest five other things
we could test, and explain the reasoning behind each.
Notice the line "We tried shorter subject lines". Telling the model what didn't work saves you from getting that same advice back.
Explain the reason behind your rules
Many people write rules in capital letters: "NEVER use bullet points." A rule without a reason works, but only literally. The model follows the rule without understanding what you are trying to achieve.
Compare:
Never use bullet points.
Write in flowing paragraphs rather than bullet points,
because this will be read aloud on a podcast.
With the reason, the model also understands it should avoid tables, long lists of numbers and anything else that sounds odd when spoken. Anthropic's own prompting guide makes the same point: explaining the motivation behind an instruction helps the model make better decisions about everything related to it.
Use placeholders, not inventions
When you write a prompt for a document that needs real facts you don't have yet, ask the model to leave gaps instead of making things up:
Draft a job advert for a junior accountant. Where you need
details I haven't given (salary, location, start date),
write a placeholder in square brackets like [salary].
This keeps the draft honest and shows you exactly what information is still missing.
How much context is enough?
A simple rule of thumb: include anything a smart stranger would need to do the task well, and nothing they wouldn't. Context is not about length. A focused paragraph beats a page of loosely related background.
If you're unsure, try this: after writing your prompt, ask yourself "What would a colleague ask me before starting?" Then answer those questions in the prompt.
Try it now
Pick a task you do often at work or school. Write it once with no context, then again with the audience, the purpose and one piece of background. Run both through the coach and look at how the Context score changes.
Sources
- Anthropic, Best practices for prompt engineering (2026)