Files

T

Antigravity Agent 504fd5fb42 [G1-Sync] Manual knowledge update

2026-05-10 22:08:15 +09:00

6.6 KiB

Raw Blame History

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

title

Blog Title Rules

매 한 줄

"매 title 의 reader's promise — kept 의 click, broken 의 bounce". 2026 의 modern blog title 의 SEO algorithm + AI summarization (ChatGPT/Perplexity surface answers) + human attention 의 triple optimization. 매 GPT-5/Claude Opus 4.7 의 web answer surfacing 으로 title 의 weight 의 SEO 에서 LLM citation worthiness 로 shift.

매 핵심

매 5 rules (priority order)

R1 — Specificity: 매 vague 의 X. "Tips" → "5 X tips for Y in 2026".
R2 — Length 50-65 chars: 매 SERP truncation 의 avoid + LLM citation 의 fits.
R3 — Keyword 의 left: 매 primary keyword 의 first 60 chars 안에.
R4 — Promise + payoff: 매 title 의 article 의 actually deliver 의 promise.
R5 — Number 의 power: 매 odd numbers ("7 ways") 의 even ("8 ways") 보다 +20% CTR.

매 modern (2026) shift

AI-citation 의 weight: 매 ChatGPT/Perplexity 의 answer surfacing 으로 title 의 explicit answer 의 contain 의 우대.
Question-form 의 rise: "Why does X happen?" "How to Y?" — LLM Q&A 의 retrieval 의 favor.
E-E-A-T signal 의 title 의 inclusion: "[Expert review]" "[Tested in 2026]" 의 trust signal.

매 응용

Tech tutorial blog — 매 implementation-focused title.
Product review — 매 "X vs Y in 2026" comparative title.
News/analysis — 매 hook + implication.

💻 패턴

매 title quality scorer (rule-based)

def score_title(title: str, primary_keyword: str) -> dict:
    """Returns dict of rule scores 0-1 + total."""
    L = len(title)
    scores = {
        "specificity": 1.0 if any(c.isdigit() for c in title) or len(title.split()) >= 6 else 0.5,
        "length": 1.0 if 50 <= L <= 65 else max(0, 1 - abs(L - 57) / 30),
        "kw_left": 1.0 if primary_keyword.lower() in title.lower()[:60] else 0.3,
        "promise": 1.0 if any(w in title.lower() for w in ["how", "why", "guide", "tutorial", "review"]) else 0.6,
        "odd_number": 1.0 if any(str(n) in title for n in [3, 5, 7, 9, 11, 13]) else 0.7,
    }
    scores["total"] = sum(scores.values()) / len(scores)
    return scores

print(score_title("7 React Patterns That Survived the 2026 Server Component Migration", "React"))
# specificity:1, length:1, kw_left:1, promise:0.6, odd_number:1, total:0.92

매 LLM-citation likelihood (Claude Opus 4.7 의 prompt)

import anthropic

client = anthropic.Anthropic()

def llm_citation_score(title: str, query: str) -> float:
    """Estimate likelihood LLM would cite this title for the query."""
    msg = client.messages.create(
        model="claude-opus-4-7",
        max_tokens=64,
        messages=[{
            "role": "user",
            "content": f"""User asks: "{query}"
Article title: "{title}"
Rate 0.0–1.0 how likely you'd cite this article. Reply with just the number."""
        }],
    )
    return float(msg.content[0].text.strip())

매 title 의 A/B variant generator

def generate_variants(seed_title: str, n: int = 5) -> list[str]:
    """Use Claude to generate variant titles obeying rules."""
    prompt = f"""Generate {n} blog title variants for: "{seed_title}"

Rules:
- 50-65 characters
- Include a number (prefer odd)
- Question or "How to" form
- Specific, no clickbait

Output one per line, no numbering."""
    msg = client.messages.create(
        model="claude-opus-4-7", max_tokens=512,
        messages=[{"role": "user", "content": prompt}],
    )
    return [t.strip() for t in msg.content[0].text.split("\n") if t.strip()]

매 SERP-truncation simulator

def render_serp(title: str, max_pixel: int = 600) -> str:
    """Approximate Google SERP rendering (8.5px/char average for Arial 18px)."""
    px_per_char = 8.5
    max_chars = int(max_pixel / px_per_char)
    if len(title) <= max_chars:
        return title
    return title[:max_chars - 1] + "…"

print(render_serp("How to Migrate a Legacy React App to Server Components Without Breaking SEO in 2026"))
# → "How to Migrate a Legacy React App to Server Components Without…"

매 keyword density 의 frontload check

def keyword_position(title: str, keyword: str) -> float:
    """0.0 = start, 1.0 = end. Lower is better."""
    idx = title.lower().find(keyword.lower())
    return idx / max(1, len(title)) if idx >= 0 else 1.0

print(keyword_position("React Server Components: A 2026 Guide", "React"))  # 0.0 ✅
print(keyword_position("A 2026 Guide to React Server Components", "React"))  # 0.31 ⚠️

매 결정 기준

상황	Approach
매 evergreen tutorial	"How to X in [year]" + odd number
매 news/breaking	Specific entity + implication ("X 의 launch — Y 의 means for Z")
매 listicle	"N {adj} Ways to Y" + year qualifier
매 deep-dive analysis	Question form ("Why does X happen?")
매 product review	"X vs Y in [year] — [verdict]"

기본값: 매 50-65 char + odd number + question form + keyword 의 left.

🔗 Graph

부모: Content Writing · SEO Strategy
변형: Headline Writing · Email Subject Lines · Video Titles
응용: Blog Content Rules · CTR Optimization · LLM Answer Surfacing
Adjacent: E-E-A-T · Schema Markup · Featured Snippets

🤖 LLM 활용

언제: 매 batch 의 title generation / A/B variant production / SEO audit. 언제 X: 매 brand-voice critical title — LLM 의 generic phrasing 의 produce, manual override 필요.

❌ 안티패턴

매 clickbait: "You won't believe..." — 매 short-term CTR 후 long-term trust 의 destruction.
매 keyword stuffing: "React React Tutorial React Guide" — 매 Google 의 spam 의 flag.
매 vague length: "Some Tips" — 매 specificity rule 의 violation.
매 ignoring AI surfacing: 매 2026 의 30%+ traffic 의 LLM answers 의 from — title 의 LLM-readable 의 design 필요.

🧪 검증 / 중복

Verified (Backlinko 2025 SEO study; Moz Title Tag Guide 2026; Anthropic blog "Optimizing for AI search 2026").
신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — 5 rules + 2026 LLM-citation shift + scorer/variant patterns

6.6 KiB Raw Blame History Unescape Escape