Prompting for SEO: how to brief an AI writer like an editor

When an article comes back generic, the instinct is to blame the model. But look at what it was given. "Write a 1,500-word blog post about drain maintenance" is the kind of instruction no editor would hand a freelancer, because every editor knows what comes back: something technically on-topic that could have been written for any plumbing company in the country. Freelancers get briefs. AI writers mostly get one-line wishes, and the output shows it.

Anthropic's own documentation tells you to treat Claude as "a brilliant but new employee who lacks context on your norms and workflows." That's the whole framing you need. A new employee with no context doesn't need a cleverer one-line instruction — they need the context. This article covers what to put in the brief; editing the draft so it sounds like you is the companion piece.

Why the brief matters more than the wording

There's a small industry of "magic prompts" that promise better AI articles through secret phrasing. The vendors' actual guidance is plainer. Anthropic's version: the more precisely you explain what you want, the better the result. OpenAI's prompt engineering guide says the same in different words — give the model examples of what good output looks like, mark out the different parts of your input clearly, and be specific about context and format. None of that is wording tricks. It's the information an editor puts in a brief.

Google's position makes the stakes concrete. Since early 2023 its stated policy has been to reward quality content however it's produced — and its spam policies warn that "using generative AI tools or other similar tools to generate many pages without adding value for users" can violate the rules against scaled content abuse. The difference between those two outcomes is decided almost entirely before generation, in what you asked for. A vague prompt produces the average of the internet, and the average of the internet adds no value by definition.

The six parts of an editor's brief

A professional content brief varies by team, but the load-bearing parts are stable. Write these down once per article — fifteen minutes, most of it thinking you'd have to do anyway:

• Audience: who is reading, what they already know, and what they're worried about. "Homeowners who just found water in the basement and are deciding whether to call someone tonight" produces a different article than "people interested in plumbing."
• The job of the article: what the reader should be able to do or decide after reading. This is what search intent means in practice — someone typed a question into Google, and your article either answers the question behind it or it doesn't.
• A working title and the heading outline: write the section headings yourself, in order. The outline is where the article's thinking lives, and it's the single highest-leverage thing to keep in human hands.
• Your facts: the claims, numbers, prices, and examples the article must include, from your own business. The model cannot know that your average callout is $180 or that Tuesday is your busiest day, and everything it substitutes for missing facts is a guess.
• Internal links: which of your existing pages this article should point to, with the anchor text. The model doesn't know your site exists.
• The don'ts: phrases you ban, competitors you won't name, claims you can't legally make, and the tone you're avoiding.

Formatting the brief as a prompt

Both vendors recommend the same mechanical trick: label the parts. Anthropic suggests wrapping sections in tags like <instructions> and <example> so the model can tell your rules apart from your background material; OpenAI recommends the same separation using headings or delimiters. You don't need to be technical about it — a prompt with clearly labeled sections for AUDIENCE, OUTLINE, FACTS TO INCLUDE, LINKS, and RULES beats an eloquent paragraph that mixes them together.

Two additions repay their cost every time. Paste in two or three paragraphs of writing you're proud of and say "match this" — that's the examples technique both vendors push, and it does more for voice than any adjective list. And ask for the outline back before the article, so you can fix the structure while it's still cheap to fix.

The model can only be as specific as the brief. Everything you leave out, it fills in with the average of everyone else's website.

The quality bar is published — use it as your checklist

Google publishes self-assessment questions for what it considers helpful content, and three of them make a brutal, useful review pass for an AI draft: does the content provide "original information, reporting, research, or analysis"? A "substantial, complete, or comprehensive description of the topic"? "Insightful analysis or interesting information that is beyond the obvious"? An unbriefed AI draft typically fails all three, because originality and insight can only come from what you put in the brief.

Two more checks belong in the same pass. Verify every factual claim against a source you trust, because a confident wrong number is the failure mode AI is best at. And read the draft asking where your firsthand experience shows — adding it is usually the difference between a page that ranks and one that doesn't.

What this looks like in practice

The first brief takes half an hour because you're deciding things you've never written down: who your reader is, what you want articles to do, which phrases you hate. The fifth brief takes fifteen minutes, because most of it is reusable — audience, voice samples, and rules carry over, and only the outline, facts, and links are new per article. That reusable core is worth keeping in a document of its own. Teams that skip the brief and generate from one-liners produce content faster, and then discover that volume without value is precisely what fails. The brief is the value. The model is just the typing.