How to Outline a Blog Post Before You Open the CMS
A website skill to outline a blog post before you open the CMS. Dump the note, name one search job, lock five H2s with proof slots, then run ChatGPT, Claude, or Gemini and do one human pass before the editor can tempt you.
Anyone about to "just start writing" can outline a blog post before you open the CMS in 20 minutes: dump the note, name one search job, lock three to five H2s with proof slots. Run it in ChatGPT, Claude, or Gemini on a plain page so the editor cannot tempt you into a hero image and a half-argument.
The CMS is a factory floor. An outline is the work order. If you open the factory first, you will format a thought you have not decided. That is how 2,000-word posts ship with no claim.
AI Overviews pull H2s as the skeleton of a citation. Claims must match a job. You cannot get that from a blank CMS window.
This loop is a skill, not a ritual outline template. If you still paste "write an outline about X" into a chat and then wander into WordPress, read what a skill prompt actually is and save the card at the end.
Outline a blog post before the CMS in five steps
Dump first. Job second. Headings third. CMS never until the outline can stand alone.
1. Dump raw, off the CMS
Paste the note, the transcript, the links, the half-opinions. Instruct ChatGPT, Claude, or Gemini: "Do not outline yet. Repeat back only what I said, as bullets, in my words. Mark opinions as OPINION. Mark facts as FACT. Mark holes as UNKNOWN. Stop."
A table of contents with no facts is a costume.
Use the same raw-then-shape discipline as turning a messy note into a one-page brief. The brief decides the job. The outline decides the proof.
2. Name one search job
One sentence. Verb plus object. "Teach a stranger to outline a blog post before the CMS." Not "thoughts on content ops."
Send that sentence with the bullet list. Instruct: "If I named more than one job, list them and stop. I will pick one. Do not merge jobs into a definitive guide."
Merging is the silent failure. A merged outline sounds complete and ranks for nothing.
3. Force three to five H2s, each a claim
Now ask for the skeleton. Hard rules:
- H1 restates the job in human words.
- Each H2 is a claim, not a topic word. "Dump raw before you outline" beats "Step one."
- Under each H2, three slots: point, proof, stop rule.
- No H3 yet unless a step list is the point.
Tell the model: "Five H2s max. If a heading cannot be argued with, it is a topic, not a claim. Rewrite it. If proof is missing, write UNKNOWN. Do not invent a case study."
UNKNOWN keeps you out of the CMS. Park the outline in a scrap pile such as the Promptcrates library of reusable prompts, not as a draft URL.
4. Add proof slots and five FAQs you can already answer
Instruct: "For each H2, name the proof I already have: a number, a rule, a quote, a screenshot I can take. Then list five FAQs whose answers are already in the outline. If an FAQ needs new research, drop it."
FAQs are how you win AIO later, not a place to hide a second post. Steal the claim-then-proof pattern from the Promptcrates how-to guides, not a topic list. Cut introduction and conclusion H2s. A sixth H2 "in 2026" is not a claim.
5. One human pass, then and only then open the CMS
Read the H2s out loud as if they will appear in an AI Overview. If you cannot defend a heading, rewrite it yourself.
Check three traps:
- A heading that is a noun cloud.
- A proof slot that says "add later."
- A FAQ you cannot answer without a new interview.
Fix those in your own words. Send the edit back: "Replace only the headings I changed. Do not re-expand."
One pass. Then open the CMS and type from the outline. Do not outline in the editor. Do not add a section because the template has a box for it.
Reusable skill card
Trigger: You are about to open the CMS with notes and a vague title.
Input: The raw dump, one job sentence, a cap of five H2s, plus any facts you can already defend.
Output: A ledger of FACT / OPINION / UNKNOWN, then an outline: H1, three to five claim H2s, point / proof / stop under each, five FAQs already answerable.
Stop: After one read-aloud of the H2s. No CMS until UNKNOWN proof is filled or cut. No merged second job. No intro/conclusion H2s.
Takeaways
- Dump first. Job second. Headings third. CMS last.
- Three to five H2s, each a claim a skeptic can test.
- UNKNOWN proof stays out of the editor. "Add later" is how posts lie.
- Answer engines lift H2s. Save the loop. The draft window is not a thinking tool.
Frequently Asked Questions
Can I outline inside the CMS if I am fast?
No. The CMS invites formatting, images, and a publish button. An outline is a decision document. Decisions die in a draft window.
How many H2s should the outline have?
Three to five. Each H2 is a claim a skeptic can test. A seventh H2 is a second post or a leftover dump.
Should I use ChatGPT, Claude, or Gemini to outline?
Yes, after you name the job. Claude is strict with "do not add sections." ChatGPT is fast at variants. Gemini is fine for a first pass. All three need your human pass.
Where do FAQs go in the outline?
After the claims, as five questions you can already answer. Do not outline questions you still need to research in the CMS.
What if the dump contains three posts?
Split before you outline. One job per outline. Three jobs in one CMS draft is a newsletter, not a guide.