You have a keyword, a Friday deadline and a chatbot that will happily produce 2,000 words in nine seconds. The words will be fine and the page will be forgettable, because the chatbot has only seen what everyone else already published.
The fix is not a cleverer article prompt. It is a better brief. AI prompts for SEO writers earn their keep upstream (briefs, outlines, gap checks) and downstream (refresh audits), while the middle, the part with original information, is yours. This post gives you templates for each stage, a small script that checks a draft against the brief, and the parts of Google's own guidance that apply.
For a full end-to-end workflow from keyword research to publishing, see AI SEO content strategy prompts. This post is narrower: briefs, outlines and refreshes. All prompts here are untested templates; the script output further down is real.
What does Google actually say about AI-assisted content?
Checked on 2026-10-30 against Google Search Central's "Creating helpful, reliable, people-first content" page (it showed "Last updated 2026-10-05 UTC"). The points that matter for a writer:
- Google's ranking systems aim to prioritise helpful, reliable information created to benefit people, "not content that's created to manipulate search engine rankings."
- Its self-assessment questions ask whether content provides original information, reporting, research or analysis, and whether it offers substantial value compared with other pages in the results.
- One of the warning signs is "Are you using extensive automation to produce content on many topics?" Another is "Are you writing to a particular word count because you've heard or read that Google has a preferred word count?" The page's own answer: "No, we don't."
- The "Who, How, and Why" section says that where automation is used to substantially generate content, you should ask whether its use is evident to visitors and whether you explain how and why it was used. It also says using AI-generation for the primary purpose of manipulating rankings violates its spam policies.
None of that says "do not use AI." It says the page has to be useful to a person who arrives directly, and that a pile of automated pages on many topics is the pattern to avoid. Everything below is built so that AI does the structuring and you supply the substance.
How do I write a content brief with AI?
A brief is a decision document. Its job is to make three things explicit before anyone writes: who is reading, what they need to do afterwards, and what this page will contain that the other results don't. Ask the model to interrogate you first, then draft.
I'm a writer preparing a content brief. Do not write the article.
Target keyword: [KEYWORD]
Site and audience: [WHO READS THIS SITE, WHAT THEY ALREADY KNOW]
Business goal for the page: [e.g. reader picks a tool / signs up / understands X]
Step 1. Ask me up to 7 questions you need answered to brief this well
(first-hand material I have, products I've actually used, data I can share,
who the reader is). Wait for my answers.
Step 2. Then produce the brief with these sections:
- Reader and situation: one paragraph on who searches this and why now
- Search intent: informational, comparison, how-to or transactional, and the
one task the reader must be able to do after reading
- Questions the page must answer (6-10, phrased as a reader would ask them)
- Original contribution: what only we can add, taken from MY answers in step 1.
If I gave nothing, say "NONE - do not publish until we have something".
- Claims that need a primary source: list each, with the type of source to check
- What to leave out: topics that would pad rather than help
- How AI was used (one sentence to put on the page if appropriate)
The "NONE" instruction is the point. If you can't name an original contribution, the page will likely be a rewrite of the top results, and that is the pattern Google's guidance warns against ("mainly summarizing what others have to say without adding much value").
Check the draft against the brief with a script
A brief is only useful if you test the draft against it. This tiny checker verifies the mechanical parts: keyword in the title, in the first 100 words, in H2s, description length. I ran it on 2026-10-30 against a deliberately short sample draft I wrote for the purpose.
import re
KEYWORD = "content brief"
fm, body = re.match(r"---\n(.*?)\n---\n(.*)", draft, re.S).groups()
title = re.search(r'title: "(.*)"', fm).group(1)
desc = re.search(r'description: "(.*)"', fm).group(1)
first100 = " ".join(body.split()[:100]).lower()
h2s = re.findall(r"^## (.*)$", body, re.M)
print("keyword in title:", KEYWORD in title.lower())
print("keyword in first 100 words:", KEYWORD in first100)
print("H2s containing keyword:", sum(KEYWORD in h.lower() for h in h2s), "of", len(h2s))
print("description length:", len(desc), "(target 120-160)")
print("words in body:", len(body.split()))
Output on the sample draft:
keyword in title: True
keyword in first 100 words: True
H2s containing keyword: 1 of 4
description length: 78 (target 120-160)
words in body: 70
Two things to read from that. The description is too short (78 characters), which is a real fix. And "1 of 4" H2s is a prompt to ask whether the other headings are useful, not an order to stuff the keyword in. Use the targets as prompts for judgement; "words in body" is printed for information only, never as a goal.
How do I turn a brief into an outline that isn't a copy of the top results?
Start from the reader's questions, not from the competitors' headings. Then use competitors only to find gaps.
Brief:
<brief>
[PASTE THE BRIEF]
</brief>
Competitor headings I collected (for gap analysis only; do not copy their wording):
<competitors>
[PASTE H2/H3 LISTS FROM 3-5 TOP RESULTS]
</competitors>
Produce an outline:
1. One H2 per reader question from the brief, in the order a reader needs them.
Phrase each H2 as the question or task.
2. Under each H2, list the exact thing the section gives the reader: a template,
a table, a worked example, a checklist, a decision rule. If it only gives
background, mark it CUT.
3. For each section, mark where OUR original material goes: [OURS: ...].
Mark sections with no original material as [GENERIC].
4. Gaps: questions the brief asks that none of the competitors answer well.
5. Overlaps: sections that duplicate each other.
Do not add a conclusion section that restates the post.
Sections tagged [GENERIC] are where readers will drift away and where an AI summary of the topic will compete with you. Either find original material or cut them. Keep the headings sentence-case and answer-first; answer-first rewrite prompts covers how to tighten the opening of each section.
How do I refresh an old post with AI?
Most refresh prompts say "update this article for 2026" and let the model invent whatever sounds current. Do the opposite: audit first, and rewrite only what the audit flags.
Below is a published post. Do not rewrite it yet.
Step 1. List every claim that can go stale: prices, plan limits, version
numbers, dates, named tools, screenshots, statistics, legal or policy statements,
and external links. Quote the sentence each comes from.
Step 2. For each, give the type of primary source I should check (vendor pricing
page, docs, regulator page) - not an answer. You have no browsing in this
conversation, so do not state what the current value is.
Step 3. List sections that are still correct and should be left alone.
Step 4. List what a reader today would likely look for in this topic that the
post does not answer. Mark each as "suggestion, needs my research".
POST:
[PASTE]
Then you do the checking against the actual pages, record the check date, and only afterwards ask for a redraft of the flagged sections, pasting the verified facts as the only allowed source. Two refresh rules worth keeping:
- Leave good sections alone. A rewrite that changes everything also changes what already worked.
- Change the visible date only for substantive edits. Swapping a date to look fresh, with no real change, is the sort of thing that erodes trust with readers; set
updatedwhen facts were actually re-checked or content really changed.
Which parts of SEO writing should I not hand to AI?
- The original contribution. Your tests, screenshots, numbers and opinions can't be generated. If the model writes "in my experience," delete it unless it's true.
- Facts you haven't verified. Prices, limits and policies change; check the primary page and put the date in the post.
- Bylines and credentials. Google's guidance specifically calls fabricated creator profiles deceptive. Use your real name and a real author page.
- Volume for its own sake. Producing many pages on many topics with automation is listed among the warning signs. Fewer, better-evidenced pages are the safer plan.
- Keyword counts and word counts as goals. Use them as a quick check, not a target.
Before publishing, run the page through Google's own questions: would you bookmark it, does it leave someone able to do the task, does it say how it was made if a reader might wonder? For prompts that structure inputs well, giving context is the lesson to revisit, and the prompt library has related templates.
