Topic & title generation
Topic directions and title variants generated from the industry question network; topics pass a seven-dimension topic score first.
Platform Agent Pipeline
Topic to publishing, no broken links between
Generate topics and titles from industry intent, write long-form with the project knowledge base, gate with 15-dimension GEO scoring and layered compliance, then publish through automatic or collaborative channels.
What it solves
- AI writing carries two risks at once: bland copy AI ignores, or over-claiming that trips advertising law.
- Quality and compliance should not depend on an editor remembering the rules — they should be machine gates.
- Publishing is only the start; publish without reading back and you have wasted the effort.
How it works
- 15-dimension GEO scoring (deterministic rules; weights configurable; normalized to 100): citations, statistics, FAQ structure, E-E-A-T, entity clarity, structured headings, depth, readability, AI readability, quotes, unique value, originality, duplication check, decision information, promotional tone. "Promotional tone" is the only negative dimension — AI trusts factual statements over self-praise.
- Layered compliance gates: sensitive-word checks (platform + tenant lexicons), industry compliance packs, AI-content labeling, advertising-law red lines.
- Review flow: internal review → client confirmation (optional) → ready to publish, with mobile approval on WeChat mini program.
- Two distinct publishing modes, not conflated: 21 platforms support automatic publishing (official API, including some desktop-local publishing); 10 platforms use collaborative publishing (we generate a publishing guide card, you publish on the platform and paste the link back, and metrics are still read back). Each platform labels its mode under Distribution → Publishing Management — no vague "one-click to every platform" promise.
Capabilities
Six stations on the pipeline
Knowledge-injected writing
Long-form writing grounded in the project knowledge base, with market-pack writing guidance injected when configured.
15-dimension GEO scoring
Deterministic per-dimension scoring with configurable weights; promotional tone is the only negative dimension.
Compliance gates
Sensitive words, industry compliance packs, hard market taboos and AI labeling — layered checks that intercept before publishing.
Review flow & mobile approval
Internal review → optional client confirmation → ready to publish; client confirmation works on the WeChat mini program.
Automatic + collaborative publishing
21 platforms publish automatically (API / desktop-local), 10 publish collaboratively (guide card + receipt backfill); registered links feed citation attribution.
Content quality gate
Publishing is not the finish line — 15 dimensions decide
Every article passes this deterministic score before publishing: 15 dimensions scored individually, summed and capped at 100. Weights are configurable per tenant, and the same article yields the same score on any machine — not a matter of "the AI thought it looked fine".
Citations
15
Statistics
12
FAQ Structure
8
E-E-A-T Signals
16
Entity Clarity
8
Structure (Total-Refine-Total)
8
Content Depth
8
Readability
5
AI Readability (Chunking)
5
Expert Quote
5
Unique Memory Words
5
Originality / Info Gain
5
Duplication Check
5
Decision Readiness
10
Promotional Tone (Negative)
8
Citability
15
Figures are each dimension's maximum (its weight), identical to the admin "GEO scoring rules" page; scores come from rules, with no human grading.