How do I get an AI writing assistant to follow a content brief?

Give the assistant a structured brief, not just a topic—and ask it to check its draft against the brief before it answers. Clear instructions, relevant context, and specific output requirements generally make prompts more reliable. (help.openai.com)

A useful prompt template

Write [type of content] for [audience] to help them [goal].

BRIEF
- Topic:
- Key message:
- Audience and what they already know:
- Tone and style:
- Required points:
- Sources or facts to use:
- Things to avoid:
- Length and format:
- Call to action:

RULES
- Treat the brief as the requirements for the draft.
- Do not invent facts, quotes, statistics, or source claims.
- If a requirement is unclear or conflicts with another, ask me before drafting.
- After writing, check the draft against every brief item and revise any misses.

OUTPUT
Return only:
1. The draft
2. A short checklist showing whether each requirement was met

To improve the result

• Make requirements concrete. Instead of “make it engaging,” say “open with a practical problem; use short paragraphs; avoid hype.”
• Separate hard rules from preferences. For example, “must include the three benefits” versus “prefer a conversational tone.”
• Provide examples of the voice or format you want, if you have them.
• Ask for a brief check before drafting if something is ambiguous or contradictory.
• Review the checklist yourself. A model can help catch omissions, but it can still miss a requirement or confidently include an unsupported claim.

You can also paste the brief under a clear heading or between delimiters, so it’s easy to distinguish the instructions from the source material. (help.openai.com)

The most reliable approach is to treat the brief as the prompt itself, then add checks that catch misses. Here's how I'd do it.

1. Paste the whole brief, not a summary. Keyword, audience, angle, structure, tone, and must-cover points should all be in the prompt. One guide argues that a topic alone yields generic writing, while an angle gives the piece a point of view, and a prompt with only the topic will get you a piece that sounds like everyone else's. Some specifics to include: target audience, content goal, search intent, primary keyword, secondary topics, recommended structure, key questions to address, tone, and important information the writer should cover.

2. Spell out constraints, but keep them few. Word limits, required elements, and format rules work well. One source notes that a length limit is the single most consistently useful constraint. Be careful about stacking too many rules: the risk is using too many that pull in different directions. Focus on the few that decide whether the output is usable.

3. Describe the tone concretely. Words like "friendly" are vague. The more you can describe the tone you want rather than just labeling it, the better the calibration. You can also paste examples of tone you want to match, which is often more effective than adjectives.

4. Show a sample of what good looks like. Paste a short excerpt of your house style. A reference passage tends to communicate voice better than instructions do.

5. Tell the AI to check for gaps before drafting. You can ask it to ask up to 5 clarifying questions if anything is missing or ambiguous, and not start the task until you've confirmed. This catches missing brief fields before you get a draft built on guesses.

6. Tell it what to do with unknowns. For factual content, instruct it to use only the source material you provide and to flag anything not covered. One template suggests marking anything the document does not state as "not in the source." This reduces invented claims.

7. Ask for a compliance check, then revise specifically. Have the model check its draft against your brief's list, such as word count, keyword use, required sections, and tone. Then give targeted feedback rather than starting over. Iteration is the normal workflow: review the response, note what worked and what needs to change, and adjust the wording accordingly.

8. Save the stable parts. If your tool supports projects, custom instructions, or saved templates, store your voice guidelines, approved sources, and standard constraints there. Then each brief only needs its specific details. One practitioner recommends this setup so that you're not starting from scratch every time.

A practical template:

You're writing [content type] for [audience]. Goal: [what the reader should do or understand]. Primary keyword: [X]. Angle: [specific point of view]. Structure: [sections/headings]. Must cover: [points]. Must avoid: [claims, jargon, words]. Length: [limit]. Tone: [described, with an example paste]. Use only the source material below; flag anything it doesn't cover. Before writing, ask me any clarifying questions. Afterward, check your draft against this brief and list any misses.

If the output still drifts, the usual cause is a missing piece of context rather than a model limitation. Check which brief element the draft missed, and add that element explicitly to the next prompt.

To get an AI writing assistant to follow a content brief, you must shift your perspective: the brief is a contract, not a suggestion. AI models are literal and perform best when provided with a highly structured, prescriptive blueprint rather than vague requests.

The most effective approach is to move away from "prompting" and toward "engineering" a structured input.

1. Build a Structured "Blueprint"

Instead of writing a loose list of requirements, organize your content brief into distinct, labeled sections. AI models process structured data (like XML tags or clear headers) much more effectively than paragraphs of narrative instructions.

Use a standardized template with these 8 core fields:

• Intent: What is the primary purpose of this piece? (e.g., educational, lead generation, thought leadership).
• Target Query: The primary keyword or question the content must address.
• Audience: Define the reader’s role, pain points, and current level of knowledge.
• Role/Persona: Define who the AI should act as (e.g., "Act as an expert technical writer with a focus on simplicity").
• Evidence Requirements: Specify what must be cited or included (e.g., "Must include two industry statistics and one customer case study").
• Structural Skeleton: Provide the exact H1, H2, and H3 headings.
• Tone Exemplars: Instead of saying "professional," provide a short snippet of writing that matches your desired brand voice.
• Acceptance Criteria: A "checklist" of non-negotiables (e.g., "No jargon," "Sentences must be under 20 words," "Include a CTA at the end").

2. Use "Content-as-Context"

If the AI is "hallucinating" or ignoring brand guidelines, it often lacks the right context. Do not just tell the AI to "be on-brand."

• Provide Reference Material: Upload your actual style guide, brand voice manual, or high-performing past articles as reference documents.
• Use XML Tags: If you are using advanced models (like Claude or GPT-4), wrap your instructions in tags to help the model distinguish between instructions and input:

  <instructions>
    Write the article based on these guidelines: [Insert Guidelines]
  </instructions>
  <context>
    Refer to these brand voice examples: [Insert Examples]
  </context>
  <outline>
    [Insert H1, H2, H3 structure]
  </outline>

3. The "Upstream" Fix

If the AI is failing to follow the brief, the problem is almost always in the brief, not the AI.

• Don’t iterate on the prompt: Many users waste time tweaking the "system message." If the output is poor, spend that time tightening the brief.
• Acceptance Criteria are key: Include a section that explicitly states what not to do. For example: "Do not use phrases like 'At its core' or 'In the digital age'."
• Human-in-the-loop: Treat the AI output as a "first draft" that requires a human editorial pass for factual accuracy, unique insights, and brand nuances.

4. Optimize the Workflow

• Make it Reusable: Your brief should be a "living asset." Over a few months, your brief template should encode exactly what "good" looks like for your specific brand, including recurring rules and structural preferences.
• Modular Prompting: If the task is complex, break it into smaller prompts. Have the AI generate the outline first, approve it, and then have it write section by section. This prevents the AI from losing track of the instructions halfway through a long document.

Summary: The goal is to provide enough constraints that the AI cannot "guess" or "improvise" beyond the scope of your brand's strategy. By focusing 80% of your effort on the structure of the brief and only 20% on the prompt itself, you will see a drastic improvement in consistency and quality.

Give the assistant the brief as a set of explicit, checkable instructions, not just background information. Clear headings and separators help distinguish the task, context, and requirements; specify the output format and include a good example if you have one. ¹²⁶

A practical workflow

1. Organize the brief into sections: audience, goal, key message, required points, tone, format and length, SEO terms (if relevant), and things to avoid.
2. Rank requirements: mark what’s mandatory versus optional. Be precise about ambiguous terms—for example, define “conversational” with a short example or description.
3. Ask for a plan before the draft. Have the assistant summarize the requirements and propose an outline. Correct anything it missed.
4. Request the draft with clear limits and a defined structure.
5. Check it against the brief. Ask for a requirement-by-requirement audit, then revise any misses. Don’t rely on the assistant’s self-check alone for factual or brand-sensitive claims.

Copy-and-paste prompt

You are a [type of writer]. Write [content type] for [audience] to achieve [goal].

## Brief
- Main message:
- Audience and what they already know:
- Required points:
- Tone and style:
- Length:
- Required structure:
- Keywords or phrases to include:
- Sources or information you may rely on:
- Avoid:
- Optional preferences:

## Rules
- Treat the brief as the source of truth.
- Follow every mandatory requirement. Do not invent facts, quotes, statistics, or sources.
- If a requirement is unclear or the brief conflicts with itself, ask me before drafting.
- Keep the output in the requested format.

First, restate the mandatory requirements and give me a short outline. Wait for approval before drafting.

After approving the outline, prompt:

Draft the content from the approved outline. Then provide a brief checklist showing where each mandatory requirement is met. Flag anything you could not verify or satisfy.

Clear, specific instructions and an explicit format generally give the assistant less room to guess; examples can further clarify the result you want. ²⁶