SEO Content Brief Generator

What it does

You give it one keyword, your site's topic, the pages you already have and what you can show first-hand. It returns a brief a writer can work from without a call: usually 700-1,000 words, most of them in the outline.

A plain "write me an SEO brief" usually gives you headings and a keyword list. This prompt adds the parts an experienced SEO editor checks first:

• A verdict before the outline. New page, update an existing page, or skip, judged by search intent, so you don't publish two pages that compete for the same search.
• A coverage pass. Before outlining, it works out the questions a practitioner would expect this page to answer (the decisions, the alternatives, the common mistakes, what the "it depends" answers depend on) and gives each one a section.
• Instructions, not topics. Under every heading the writer gets what to explain, the decision or number the reader needs there, the mistake to warn about and the piece of your proof that goes there.
• An angle you can prove. It names the first-hand evidence to collect (measurements, photos, real numbers), which is the part generic pages can't copy. "More comprehensive" is not accepted as an angle.
• A short opening answer that makes sense quoted on its own, in a search snippet, an AI summary or a reader's skim.
• Links that exist. Every page of yours that genuinely fits, placed in the section where it helps, with anchor text, plus the pages that should link back. It never invents a page; a missing one goes on a one-line content-gap note.
• Honest limits: no invented search volumes or statistics, and every assumption listed.

Use it when

• You have a keyword list and want the same brief format for every article before anyone writes.
• You hand articles to freelance writers and want fewer rewrites.
• You are unsure whether a keyword needs a new page or an update to a page you already have.

Skip it when

• You need search volumes, keyword difficulty or live rankings. Use an SEO tool; a chat model cannot verify those numbers.
• The page is a product or category page, where layout, filters and conversion matter more than an outline.
• You want the article itself. This prompt plans it; drafting is a separate step.

The prompt

You are a senior SEO editor. Write a content brief for one keyword. A freelance writer who knows the subject less well than you will work from it without asking you anything, so every part of the outline has to tell them what to write, not only name a topic.

My inputs are at the end. Do not repeat them back. If something is missing, make the simplest reasonable assumption, use it, note it under ASSUMPTIONS and finish the brief. Do not ask me questions first.

Before writing, run two checks and build their results into the brief (do not print them as separate lists):
- Cannibalisation: compare each of my existing pages with this query by search intent (what the searcher wants to do or decide), not by wording.
- Coverage: work out the questions an experienced practitioner would expect a page for this query and this audience to answer: the core steps or criteria, the decisions the reader has to make, the usual alternatives, the common mistakes, and what the "it depends" answers depend on. If I pasted top results, the subtopics most of them share are a baseline this page covers too. Every question goes into a section of the outline, or under LEAVE OUT with the place it belongs instead.

Then write these sections, in this order:

1. VERDICT: "New page", "Update [existing page]" or "Skip", with one sentence why. Choose Update when one of my existing pages already serves the same intent; do not create a second page that competes for the same query. For Update, the brief describes the updated page and says what to keep, rewrite and add. For Skip, say in two lines what to do instead, and stop there.
2. INTENT & FORMAT: what the searcher wants to do or decide, and the format they expect (how-to guide, comparison, list, calculator or tool, definition). If the intent is mixed, say which part this page serves.
3. ANGLE & PROOF: the angle in one sentence: what this page will have that typical pages for this query lack. "More comprehensive" or "more up to date" is not an angle. Then the first-hand proof the writer needs, using only proof I say I have; where proof is missing, write [PROOF NEEDED: what]. If I pasted current top results, base the angle on a gap you can name in them. If I did not, write "Angle not checked against live results" in one line.
4. OPENING ANSWER: 2-3 sentences, about 50 words at most, that answer the query directly and still make sense when quoted on their own. No "In this article".
5. OUTLINE: the core of the brief, so give it at least half of the words. H2 and H3 headings in reading order. For each H2:
   - the heading, a word budget and the reader question it settles;
   - 2-4 bullets telling the writer what the section has to say: the points to make, the decision or number the reader needs (from my material or established practice, never an invented statistic), the mistake to warn about, and the expert terms to explain here rather than just mention;
   - the proof that goes here and the link that belongs here, if any.
   End with a total word range and one sentence on why that length is enough.
6. TITLES & META: 3 titles of at most 60 characters (aim for about 55 and show each count), no clickbait; 1 meta description of at most 155 characters (aim for about 140).
7. LINKS:
   - From this page: every existing page of mine that genuinely helps a reader here, however many that is ("none" is a valid answer), each with anchor text and the H2 it sits in.
   - To this page: which of my existing pages should link here, with anchor text and the spot on that page.
   - Use only pages from my list. If a page I do not have would help, add one line: "Content gap: <page>, <why>".
8. LEAVE OUT: subtopics this page should not cover, one line each, with where they belong instead.
9. ASSUMPTIONS: one short line for each assumption that shaped the brief, including anything you could not verify.

Finish with at most three lines on anything else the writer needs to start today that the sections above do not cover (for example a term to use consistently, or a reader worry to answer early). Leave it out if there is nothing.

Rules: no search volumes, difficulty scores or traffic estimates. No statistics unless I supplied them. No filler ("in today's world", "ultimate guide"). Length: usually 700-1,000 words, less when the query is simple, never padded. TITLES & META, LEAVE OUT and ASSUMPTIONS together get about 150 words; the outline gets the rest first.

Site topic and audience: [TOPIC AND AUDIENCE]
Target keyword: [KEYWORD]
My existing pages (titles or URLs, or "none"): [PAGE LIST]
What I can show first-hand (tests, data, photos, customer cases, or "nothing yet"): [FIRST-HAND PROOF]
Current top results (titles and H2s of the top 3-5, or "none"): [TOP RESULTS]

How to use

1. Fill the five lines at the end. Be concrete in [FIRST-HAND PROOF]: "we logged 40 shots on two grinders" helps; "we know coffee" doesn't.
2. For [TOP RESULTS], search the keyword in a private browser window and copy the titles and H2 headings of the top 3-5 regular results (skip ads). This input turns the angle from a guess into a real gap and shows what the page has to cover as a baseline. If you leave it as "none", the brief says the angle was not checked.
3. Put your real page titles in [PAGE LIST]. The verdict and every link come from that list, so a missing page means a missed link.
4. Read VERDICT first. If it says Update or Skip, act on that before you commission anything.
5. Fill every [PROOF NEEDED] before writing starts, and give the writer the brief and the proof together. The bullets under each heading are the writer's instructions; hand them over as they are.
6. Check title and meta lengths with a character counter; model counts are estimates.
7. Run one keyword per chat so briefs don't blend. For a shorter brief, ask for it ("same sections, two bullets per H2").

How to run (Oct 2026)

• ChatGPT: paste it as the first message. For a batch, put everything above "Site topic" in a Project's instructions (Project settings > instructions) and send only the five input lines in each chat. Medium on the Thinking slider is enough for one keyword; use High when you paste several competitor outlines. GPT-6 tends to ask questions before starting; the prompt already tells it to assume and continue.
• Claude: paste it into a new chat, or use a Project: the instructions as project instructions and your sitemap or page list as a project knowledge file, so LINKS points to real pages. Claude Opus 5.5 is the default; Claude Sonnet 5.5 is the faster, cheaper choice for a batch.
• Gemini: in the Gemini app, save the instructions part as a skill (type / and its name) and send the inputs each time. In Google AI Studio or the API, Gemini 3.8 Flash at its default medium thinking level fits this job. If you paste long competitor outlines, put them first and end with "Based on the information above, write the brief." Gemini answers briefly by default, so keep the length line in the prompt.
• DeepSeek (API), Qwen (Model Studio or QwenWork), GLM (Z.ai API), Kimi (kimi.com or API): thinking is on by default for DeepSeek-V4.1-Flash and Qwen3.8-Max, and always on for GLM-5.3 and Kimi K3. Use only the final answer, not the reasoning. For a quick batch, lower the effort (reasoning_effort low on DeepSeek, GLM and Kimi; a thinking_budget cap on Qwen); keep the default for an important page.

Example

Input

• [TOPIC AND AUDIENCE]: home espresso for beginners with their first machine
• [KEYWORD]: how to dial in espresso
• [PAGE LIST]: "Best burr grinders under 300 €", "Espresso ratio calculator", "Why is my espresso sour?"
• [FIRST-HAND PROOF]: we logged 40 shots on two grinders while dialing in three coffees
• [TOP RESULTS]: none

Output (excerpt)

Illustrative example written by Meevra to show the target format; your model's wording will differ.

Troubleshooting

• Every keyword gets "New page". Check that [PAGE LIST] holds your real titles, then reply "Compare each existing page by search intent, not by wording, and redo VERDICT."
• The angle is "more comprehensive" or "more up to date". Reply "That is not an angle. Give 3 angles that change the format or the evidence, not the length."
• The outline bullets name topics instead of telling the writer what to say. Reply "Rewrite each H2's bullets as instructions: the point to make, the decision or number the reader needs, the mistake to warn about."
• A link points to a page you don't have. Reply "Use only pages from my list; put anything else on a content-gap line."
• The outline totals more than 1,500 words. Reply "Cut to the shortest article that fully answers the query; drop any H2 a reader in a hurry would skip."
• It quotes search volumes or traffic numbers. Delete them and remind it of the rules line. Get numbers from your SEO tool.
• Title counts look wrong. Count them yourself; models estimate character counts.

© 2026 Meevra · meevra.com · Licensed to the purchaser for personal and internal business use.