Content Quality (E-E-A-T 2026)
Content Intelligence Workflow
Before content recommendations, run scripts/analyze-keywords.ts (local-first, no API key). Use it as first-pass evidence for keyword distribution, semantic breadth, local modifier placement, heading coverage, and stuffing risk.
bun run scripts/analyze-keywords.ts <url-or-path> --keyword "<primary keyword>" --synonyms "<syn1,syn2>" --locations "<city1,city2>" --format markdown
It returns density, n-grams, a 0-100 stuffing score, heading coverage, and per-location contextual mentions — purely local HTML parsing.
E-E-A-T Pillars
- Experience: First-hand knowledge signals (case studies, photos, "I tried...")
- Expertise: Author credentials, depth, technical accuracy
- Authoritativeness: Industry recognition, citations, backlinks
- Trustworthiness: Contact info, HTTPS, transparent ownership, fact-checking
Anti-Cannibalization
- One primary keyword per URL
- Different search intents per page (info / navigational / transactional)
- Internal linking respects pillar/cluster topology
Metadata and Heading Rules
- Meta title must be 60 characters or less.
- Meta description must be 150 characters or less.
- Do not include the company or brand name in the meta title unless the client explicitly asks. If needed, present a branded title as an option or exception.
- Meta title and H1 should be semantically similar, but not necessarily identical.
- Meta title and H1 need the primary keyword or a strong variant.
- H2/H3 headings distribute synonyms, long-tail phrases, questions, and sub-intents.
- Avoid repeating the exact keyword across every heading.
Local Semantic Distribution
- The client's target localities (primary
[city]+ neighbouring municipalities/[region]) must appear naturally near service terms, never as a dumped city list. - Prefer sentence-level relevance such as
[service]+[city]+ proof or context. - On a LOCAL page, alternate
[city]with[region]and district/neighbourhood names instead of repeating the same city token.
Keyword Distribution by Zone
Place terms by page zone, not by hitting a density target. Anti-stuffing 2026 is multi-signal, not a fixed percentage (Google's leaked KeywordStuffingScore runs 0-127; risk rises past ~3% density).
| Zone | Primary [service] |
Local modifier [city]/[region] |
Synonyms + entities | Secondary terms |
|---|---|---|---|---|
| Title / H1 | Yes (exact or strong variant) | Local page only | — | — |
| H2 / H3 | Sparingly (1-2) | Local page only, varied | Yes (distribute) | Yes |
| First 100 words / answer capsule | Yes (once) | Local page: once | Yes | — |
| Body | Natural flow | Local page: spread | Yes (bulk of coverage) | Yes |
| Anchors / alt / meta | Variant | Geo-specific on local | Yes | — |
Reference counts for a 1000-1500 word page:
- Primary
[service]: 5-8 occurrences (~1-1.5% — the sweet spot, never above ~3%). - Semantic family (synonyms + named entities): 12-18 occurrences — this carries topical depth, not exact repetition.
- Local modifier
[city]/[region]: 4-6 occurrences on a LOCAL page; 1-2 mentions on a GLOBAL page (zone signal, not stuffing). On the local page, rotate[city]with[region]/district rather than repeating one city.
Keyword Stuffing Detection
Do not use a fixed >3% density threshold as the stuffing rule. Flag keyword stuffing only when multiple signals align:
- Exact keyword repetition
- Repeated n-grams
- Repeated local modifiers
- Low semantic diversity
- Thin content
- Unnatural heading, anchor, or paragraph placement
scripts/analyze-keywords.ts computes these signals into a 0-100 stuffing score.
Copywriting 2026 (citation-eligible writing)
- Answer capsule per H2: open every H2 with a self-contained 40-60 word answer, not only the page's first 100 words. AI Overviews and LLMs extract per-section; each H2 must stand alone as a quotable verbatim answer.
- Hyperlinked statistics: one statistic linked to its primary source every 150-200 words. Naked or undated stats are not citation-eligible.
- Named entities, not pronouns: name key entities explicitly (product, person, place, organization) instead of "it", "they", "this tool". LLMs disambiguate by surface entity mentions, not coreference.
- Keep the anti-AI-slop tone: no "In conclusion...", "It's important to note...", no marketing filler.
AI Content Guidelines
- Disclose AI-assisted content where required
- Human review + first-hand experience injected
- Avoid generic AI-typical structures ("In conclusion...", "It's important to note...")
Entities and Semantics
For entity-based optimization (knowledge graph alignment, sameAs, entity salience, semantic depth), use the seo-entity skill. Anchor each page to a primary entity and its attributes rather than to a keyword string alone.
References
seo-entity— entity SEO, knowledge graph, semantic depthseo-featured-snippets— answer capsule / position 0 formatsskills/seo/06-content-strategy/(eeat-implementation, anti-cannibalization, ai-content-guidelines, keyword-research, keyword-distribution)