AI Content Writing with Kimi
Write for two consumers at once: Google's ranker and the AI answer engines. This is the system that ranks in search and gets cited by ChatGPT, Perplexity, and Gemini.
1. Research from the live SERP
Pull the live top 10 for your keyword, note People Also Ask and related searches, then parse the top 5 ranking pages. Extract concepts, not word frequencies: entities and subtopics appearing on 3+ of the 5 pages are what Google expects you to cover.
2. Structure for two consumers
Every post is written for Google's ranker and for AI answer engines at once. Title with the primary keyword, a 50-80 word TL;DR that stands alone, 5-7 H2 sections, a conclusion with one specific CTA, and a 5-question FAQ with FAQPage JSON-LD.
3. Write in Answer Capsules
The single most important technique. About 60% of your H2s are phrased as the exact question a real person types, followed immediately by a 30-60 word self-contained direct answer, then 2-4 paragraphs of depth. That capsule is what AI engines lift and cite.
4. Source everything, inline
Every statistic and claim gets an inline contextual link on the relevant phrase. Never a Sources block, never footnotes. Fetch each source and confirm the claim is really there before citing it: unfetched is not a source.
5. Add first-hand experience
AI-only content barely gets cited. Ask specifically: a screenshot, a measured number, a client outcome, a mistake made. Weave it in as first-person narrative. Generic asks get nothing; specific asks get the moat.
6. Interlink from the real sitemap
Fetch the site's sitemap.xml first and build a linking pool. Weave 4-7 internal links with descriptive 2-5 word anchors. Never fake a URL: mark unsure ones as placeholders.
The technique that wins AI citations
AI engines lift a specific sentence and attach the link sitting on the keyword next to it. Win citations by making that sentence trivial to extract and your source impossible to detach from the claim.
The anatomy
H2 = the exact question a real person types
30-60 word capsule: quotable verbatim, self-contained
Depth for humans, capsule for machines
The Rule of One
- One question per H2
- One 2-sentence answer capsule
- One source per claim, inline on the keyword
- One table per post
- One first-hand anecdote per section that needs it
Source quality bar
Primary studies, .gov/.edu, official docs, first-party data. Reject undated listicles and vendor blogs as primary sources. Three confirmed strong sources beat five padded with filler.
Every post, same skeleton
Title
Primary keyword included, SEO title ≤60 characters
TL;DR
50-80 words, fully self-contained summary
Intro
150-200 words, primary keyword in the first 50
H2 sections
5-7 total: ~60% Answer Capsules, ~40% editorial
Conclusion
100-150 words, 2-3 takeaways, one specific CTA
FAQ
5 questions, self-contained answers + FAQPage JSON-LD
The prompts that do the heavy lifting
Copy these straight into Kimi. Replace the bracketed placeholders with your keyword, brief, or draft.
Semantic Extraction (pre-writing)
Target keyword: "[keyword]" 1. Pull the live top 10 results (SERP MCP or web search). Note People Also Ask and related searches. 2. Parse the top 5 organic pages. For each, list the subtopics, entities, and questions answered. 3. Extract CONCEPTS, not word frequencies: entities and subtopics appearing on 3+ of the 5 pages count as expected coverage. If 4 pages discuss "recovery time" in different words, that is one shared subtopic. 4. Deliver: must-cover terms (on 4-5 pages), supporting terms (on 3 pages), and the questions the article must answer.
Why it works: This finds the terms Google expects on the page before you write a word. Skipping it is why most AI content misses rankings.
Answer Capsule Drafter
Write the section for the H2: "[question-form H2]" Rules: - The H2 is the exact question a real person types - Immediately below: a 30-60 word self-contained direct answer. It must make complete sense pulled out of context, because an AI engine will quote it verbatim - Then 2-4 paragraphs of depth: examples, data, nuance - One question per H2. One answer capsule. One source per claim, linked inline on the relevant keyword phrase - 8th-grade reading level: sentences under 20 words, active voice, contractions, no jargon - No em dashes. No "Sources" block. No copied source text Brief and entities to include: [paste]
Why it works: AI engines lift a specific sentence and attach the link sitting next to it. The capsule makes your sentence the one they lift.
Experience Collector (human gate)
Before drafting, interview me section by section for first-hand experience. For each planned H2, ask me specifically: - Do you have a screenshot or a measured number for this? - A client outcome or a before/after? - A mistake you made or something that surprised you? Ask one section at a time. Do not accept generic answers. Then weave my answers into the draft as first-person narrative: "When we implemented this for [client], we saw [specific result]."
Why it works: The phased process is Research, approve, Structure + experience, approve, Draft. The approval stops are where the defensibility enters.
Pre-Delivery Verification Pass
Run this checklist against the draft and fix anything that fails: 1. TL;DR at top, self-contained, 50-80 words 2. Primary keyword in title, first paragraph, and 2-3 H2s 3. ~60% of H2s are answer capsules (question H2 + 30-60 word direct answer) 4. Every statistic and claim has an inline link on the relevant phrase. Flag any claim without one 5. Zero "Sources" blocks, zero footnotes, zero em dashes 6. 4-7 internal links with descriptive 2-5 word anchors. Mark any URL you cannot verify as [INTERNAL LINK: anchor -> topic] 7. 5-question FAQ with 2-4 sentence self-contained answers + FAQPage JSON-LD as a separate block 8. SEO title <=60 chars, meta description <=155 chars with primary keyword + action verb 9. Paragraphs <=4 sentences DRAFT: [paste draft]
Why it works: Verify before claiming done. A report with three real checks beats a draft with twenty assumptions.
The pre-delivery checklist
Draft to a real file, run this pass, then publish through your CMS and verify on the live URL. "I made the change" is not "the change works".
Learn the full content system, live
AI Ranking is a 7,400+ member community teaching AI SEO with live coaching: answer capsules, semantic research, and the workflows from this site, taught end to end.
AI Ranking on Skool: 7,400+ members learning AI search optimization. Free to start.