How to brief AI for content: the three layers of context it needs

Most people who say AI writes bland, generic content are half right. The output is generic. The cause usually is not the model. It is the brief.

You asked a system that knows nothing about your organisation, your standards, or your reader to produce something specific. It gave you back the average of everything it has seen.

Generic AI output is almost always a briefing failure, not a model failure. This guide is for content teams, writers, and anyone in government or a regulated field who uses AI and keeps getting drafts they have to rewrite from scratch. It sets out the three layers of context AI needs to produce something you can actually use: who you are, the rules you work to, and the specific task in front of you.

Before you start

The principle is simple. AI does not lack capability. It lacks context.

A capable person on their first day, handed a one-line request with no background, would also produce something generic. You would not brief a new starter with "write us a page about rebates." You would tell them who you are, the rules they have to follow, and exactly what this page has to do.

AI needs the same three things. Write the first two once, reuse them, and change only the third each time. That single habit does more for output quality than any prompt trick.

I built this for my own work. Any AI model I use reads a standing brief before it does anything: who I am, how I work, the standards I hold, and what I will not accept. It took an afternoon to write and has saved weeks of rewriting since.

Step 1: Write the organisational brief (who you are)

Write a short, standing document that describes your organisation, your audience, your voice, and your non-negotiables. Who you serve. What you sound like. What you never do.

This is the layer that stops AI sounding like everyone else. "Write in plain English" produces a generic result. "We write for Queenslanders applying for a disaster payment, often stressed, on a phone, who need to know if they qualify in one screen" produces something shaped to a real person.

A useful organisational brief usually covers four things: who your readers are and what state they are in, the voice you hold to, the things you refuse to do, and one or two examples of content you already consider good. Examples teach an AI faster than adjectives.

Write it once and reuse it every time. This is the layer most people skip entirely, and it is the one that does the most work.

Step 2: Write the rules and frameworks brief (the standards the output must meet)

List the specific rules, standards, and frameworks the output has to meet. Not "make it accessible": name the standard. WCAG 2.2 AA. The Australian Government Style Manual. Your own content model or governance rules.

AI follows explicit rules far more reliably than implied ones. "Follow the Style Manual" is vague. "Sentence case for headings, spell out acronyms on first use, sentences under 25 words, follow the Style Manual on numbers and dates" is a checklist it can apply and you can check against.

Link to the real source where you can. The Style Manual and the WCAG 2.2 quick reference cover most of what content teams need. Like the first layer, most of these rules do not change between tasks, so this brief is reusable too.

Step 3: Write the task brief (this specific output)

Now, and only now, describe the specific thing you want. The format. The audience for this piece. The one job it has to do. The length. What to include, and what to leave out.

With the first two layers in place, this brief can be short. Without them, no amount of detail here will fix the output, because the AI still does not know who you are or what you are working to.

A usable task brief looks like this: "Draft a 300-word web page explaining who can apply for the home energy rebate. Reader: a homeowner checking if they qualify. Answer 'can I apply' in the first two lines. Group the eligibility conditions under one heading, and keep it to no more than three."

Step 4: Give all three together, in order

Hand the AI the full stack: organisational brief, then rules, then task. Keep the first two as saved, reusable assets: a document, a saved prompt, or a project file. Only the third changes each time.

The order matters less than the completeness. What matters is that the standing context is always present. The most common failure is treating every request as a fresh, contextless task: retyping half a brief, forgetting the rules, then blaming the result.

What good looks like

When all three layers are in place, the output stops needing a full rewrite. You edit, you do not restart. The voice is recognisably yours because you described it. The rules are followed because you named them. The task is done because you set it out.

The test is quick. If a draft comes back generic, do not reach for a cleverer prompt for that one task. Look at what standing context you never gave it. The fix is almost always in the first layer, not the third.

So before your next AI request: which of the three layers are you missing?

Photo by Amélie Mourichon on Unsplash.