AboutServicesResultsWorkshopsResourcesClient PortalWork with Us
Home / Insights / Agency workflows
Agency workflows

What context should you give AI before asking it to write?

Daniela Valencia, founder of Reclaim CreativeBy , Founder, Reclaim Creative, Australia's leading AI consultancy for agencies and founders
Published 18 November 2026 · Updated 18 November 2026 · 9 min read
Short answerBefore asking AI to write, give it six things: who the audience is, your brand voice, the goal and channel, the source material, an example of good work and the limits it must respect. Without that context the model fills the gaps with generic, average language, so a briefed prompt gives your team drafts it can actually use.
Key takeaways
  • AI output is only as good as its human input, so brief it like a sharp new freelancer rather than typing a one-line task.
  • The six context layers are audience, brand voice, goal and channel, source material, examples and limits.
  • Save stable client context once in a Claude or ChatGPT project so the whole account team prompts from the same brief.

What context should you give AI before it writes anything?

Give AI the same brief you would give a sharp new freelancer: who the reader is, how your brand sounds, what the piece must achieve, the facts to work from, an example of good work and the limits it must respect.

AI output is only as good as its human input. A model has read an enormous amount of text, but it has never met your client, sat in your strategy meeting or seen last quarter's results. When you leave those gaps open, it fills them with the most average answer available. Six layers of context close most of those gaps.

Context layerWhat to includeAgency example
AudienceWho reads it, what they already know, what they care aboutTime-poor owners of Melbourne physio clinics who distrust marketing jargon
Brand voiceTone, words you use, words you never use, spellingWarm and plain, Australian spelling, no exclamation marks, never "solutions"
Goal and channelThe one action you want, the format, the lengthA 120-word LinkedIn post that drives bookings for a free assessment
Source materialFacts, notes, transcripts or data the draft must rely onThe client's call notes and three customer reviews
ExamplesOne or two pieces that show what good looks likeThe client's best-performing post from last month
LimitsWhat to avoid, claims it cannot make, compliance rulesNo health outcome promises, no invented statistics

Most people give the first line of that table and skip the rest. "Write a LinkedIn post about our new service" is a task, not a brief. The other five layers are where the quality comes from.

Why does context change AI writing so much?

Context matters because a language model predicts the most likely next words, and without your specifics the most likely words are the generic ones.

The AI companies say this themselves. Anthropic's guidance suggests treating Claude like a brilliant but new employee who lacks context on your norms and workflows, and testing a prompt by asking whether a colleague with no background could follow it (Anthropic, Be clear, direct, and detailed). OpenAI's advice is to be specific, descriptive and as detailed as possible about the desired context, outcome, length, format and style (OpenAI, Best practices for prompt engineering with the OpenAI API). Both are telling you the same thing. The model cannot read your mind or your client folder.

For agencies, missing context shows up as rework. A senior writer spends twenty minutes turning a bland draft into something on brand, and the time saving disappears. If your team keeps saying the output sounds like everyone else, we cover the causes in Why does your AI output sound generic? and the wider problem of low-effort content in What is AI slop and how do agencies avoid it?

What does a well-briefed AI prompt look like for an agency?

A well-briefed prompt reads like a creative brief: the instruction first, then clearly labelled context, then examples, then the exact output you want back.

Here is the difference in practice. A weak prompt says: "Write three Instagram captions for our café client's new winter menu." A briefed prompt tells the AI who the café serves (office workers in Richmond on a 30-minute lunch break), how the brand talks (dry humour, short lines, no hashtag walls), the goal (more weekday lunch visits), the facts (three new dishes, their prices and opening hours), one caption the client loved last season, and the limits (no "cosy vibes", no claim to the best coffee in Melbourne). Same task. A very different first draft.

Two techniques help. The first is structure. OpenAI recommends putting instructions at the start of the prompt and separating them from the context with clear markers such as ### or quotation marks. Labelled headings like "Audience", "Voice" and "Source material" work just as well, and they make prompts easier for colleagues to reuse. The second is examples. Anthropic explains that a few well-chosen examples improve accuracy and consistency, and that they should be relevant and varied so the model does not copy one too closely (Anthropic, Use examples (multishot prompting)).

One honest caution. Examples are powerful, so choose them carefully. If your only example is a mediocre post, you will get more mediocre posts. And never let AI fill factual gaps. If a statistic or testimonial is not in your source material, tell it to leave a placeholder for a human to complete.

How do you avoid retyping the same context every time?

Save your stable context once, in a project or custom instructions, so each prompt only needs the task and the details that change.

Some context changes with every job, like the source facts and the goal. Some barely changes at all: a client's voice, audience, approved claims and banned phrases. Claude Projects and ChatGPT Projects let you store reference documents and instructions that apply to every chat inside them, and ChatGPT's custom instructions do something similar across your account.

For most agencies, we recommend one short context document per client. Include the audience, voice rules with real examples, banned words, approved claims, compliance notes and two pieces of work the client loved. Keep it to a page or two. Store it in a project for that client and make it the starting point for anyone on the account. When the client's positioning changes, you update one document, not fifty separate prompts.

This is where DIY works well. One motivated person can set up a client project in an afternoon and see better drafts the same week.

Where does DIY context-giving stall for agency teams?

DIY stalls when the context lives in one person's head and the rest of the team keeps prompting the old way.

We see the same pattern across 5 to 50 person agencies. One enthusiast gets great results. Everyone else types a one-line request, gets a generic draft and decides AI does not work for their clients. Output becomes inconsistent across accounts, and senior staff end up rewriting. The fix is a shared habit, not a better tool. Everyone needs to know what context to give, where the shared context lives and when a weak draft should go back to the AI with a better brief rather than to a human for a rewrite.

The rework adds up. Take a hypothetical in-house team and use staff cost of about $70 an hour per person. Five people each spending two extra hours a week fixing under-briefed drafts is 5 × 2 × $70, or $700 a week. That is about $36,400 a year in resource time.

Briefing AI well is a marketing skill before it is a technical one. That is why Daniela Valencia, Reclaim Creative's founder, teaches it first. Dani spent 6+ years in brand and agency marketing before building AI systems, so she knows what a good creative brief looks like and what must stay human. When teams learn the fundamentals on their own processes, results move quickly. A 42 hour a month marketing task at an agency was automated in 20 minutes during a workshop, and a brand strategy sprint went from 6 weeks to 25 minutes. Reclaim Creative helped create AI workflows that solved those problems. For what good team training covers, read What training is recommended for teams adopting AI?, and if you are still deciding which tasks to hand over, see How do you know what to automate in your marketing team, and what not to automate?

What can you do this week to give AI better context?

You can improve your AI writing this week with one client, one repeat task and a one-page context document.

  1. Pick one repeat writing task. Choose something your team produces every week, such as social captions, EDM intros or blog outlines.
  2. Write a one-page context document. Cover audience, voice, goal, channel, limits and two examples of approved work.
  3. Run a side-by-side test. Prompt once with the bare task and once with the document plus the task. Count the edits each draft needs.
  4. Save the better setup as a project. Put the document in a Claude or ChatGPT project so the next person starts from the same context.
  5. Show both drafts at your next team meeting. The difference usually sells the habit better than any policy.

For a wider view of where your agency stands, work through the AI-ready agency checklist or browse our free AI resources. If you want help, see our services and get in touch.

How Reclaim Creative helps

Reclaim Creative's AI Team Workshops teach your team the fundamentals first, including how to brief AI with the right context, then build workflows around your real client work. This hands-on, in-person training runs over 1 to 2 days and starts from $7,000 + GST. If you would like a practical second opinion first, a 60-minute 1:1 AI Advisory session with Daniela is $550 + GST. Every engagement starts with a complimentary Intro Call, so see our services and get in touch.

Frequently asked questions

What context should you give AI before asking it to write?

Give it the audience, brand voice, goal and channel, source facts, an example of good work and the limits to follow. Brief it like a new freelancer.

Does context matter in AI?

Yes. Without context, an AI model fills the gaps with generic, average language, which is why one-line prompts produce bland drafts. In our experience, telling it the audience, voice, goal and source facts changes the result more than switching tools.

What to ask an AI?

Ask AI for a specific piece of work with a clear goal, not a vague topic. A useful request names the audience, format, length and the action you want readers to take, and it includes the facts the AI should work from.

How do I write a prompt?

Start with the instruction, then add labelled context: audience, voice, goal, source material, examples and limits. Finish by stating the exact format you want back, then refine the prompt based on what the first draft gets wrong.

How much context is too much for an AI prompt?

Context becomes a problem when it is irrelevant or contradictory, not simply when it is long. For marketing writing, a one or two page client document is usually enough. Cut anything that does not change how the piece should sound or what it should say.

Should I paste brand guidelines into ChatGPT or Claude?

Yes, but summarise them first. Long brand books bury the rules that matter, so pull out the voice rules, banned words, approved claims and two strong examples. Save that summary in a project so every prompt for that client uses it.

Is it safe to give AI client information?

Check your AI tool's data and privacy settings and your client agreements before pasting in confidential material. Paid business plans generally give more control over how your data is used than free accounts. When in doubt, remove names and sensitive figures from the context.

Do I need to give AI context every time I write?

AI needs context every time, but your team does not need to retype it. Store stable context such as voice, audience and banned phrases in a Claude or ChatGPT project, then add only the task-specific facts to each prompt.

Daniela Valencia, founder of Reclaim Creative
About the author

is the founder of Reclaim Creative, a female-founded AI consultancy built by a marketer with 6+ years of in-house and agency experience. She now helps agencies and founders build AI workflows their teams understand and own. · Instagram